Top 30+ AI Chip Makers: NVIDIA & Its Competitors

Based on our experience running AIMultiple’s cloud GPU benchmark with 10 different GPU models in 4 different scenarios, these are the top AI hardware companies for data center workloads. Follow the links to see our rationale behind each selection:

30+ AI chip makers by category

*The selected models are based on the latest announcements.

**Broadcom does not sell a branded AI accelerator. Its XPUs are co-designed with, and ship under the name of, the customer that will operate them. Jalapeño was unveiled with OpenAI on June 24, 2026 and is at engineering-sample stage.¹

***The Arm AGI CPU is a data center host CPU rather than an AI accelerator. Arm announced it on March 24, 2026 as its first production silicon. Early systems are available now, with broader availability expected in 2H 2026.²

Sorting is by category. Vendors are ranked by estimated market share within the top 3 categories (i.e., leading producer, public cloud, public AI cloud) because sales numbers or cloud usage can be estimated. Vendors in the last three categories (i.e., AI startup, upcoming producer, other producers) are sorted alphabetically. Vendors in the custom AI ASIC / XPU co-design category are ranked by estimated share of the custom-ASIC design market.

5 mobile AI chip providers

*Most popular & recent chips are selected.

7 edge AI chips

The demand for low-latency processing has driven innovation in edge AI chips. These chips’ processors are designed to perform AI computations locally on devices rather than relying on cloud-based solutions. Vendors quote peak throughput at different numeric precisions, so the table states the precision behind every figure:

*These are the maximum values quoted by each vendor, and they are not directly comparable across rows. A figure quoted at FP4 or INT4 describes a different operation from one quoted at INT8, and rows marked sparse are vendor figures that assume sparsity, so the precision column has to be read with the number. TOPS is tera operations per second, TFLOPS is tera floating-point operations per second, and GOPS is giga operations per second.

Understanding AI chip architectures: GPUs vs ASICs

Not all AI chips are created equal. While the vendors above compete in the same market, they use fundamentally different chip architectures:

• GPUs (Graphics Processing Units) are general-purpose processors that can handle both training and inference across a wide range of AI workloads. NVIDIA and AMD dominate this category.
• ASICs (Application-Specific Integrated Circuits) are custom-designed for specific tasks. Some support both training and inference (Google TPU, AWS Trainium), while others are inference-only (Groq LPU, AWS Inferentia).

Intel has discontinued Gaudi 3 in favor of a GPU-centric roadmap.³ Crescent Island is the replacement entry, an inference-optimized data center GPU that Intel says will sample in the second half of 2026.⁴

Broadcom’s row is a co-designed part rather than a product sold on the merchant market, and it is not yet in production. Both companies state that “Engineering samples of the Jalapeño chip are running ML workloads in the lab at production target frequency and power, including GPT-5.3-Codex-Spark,” and the platform is “designed for initial deployment by the end of 2026”.⁵

Key insight:

Not all ASICs are inference-only. Google TPU, AWS Trainium, Cerebras, and SambaNova support both training and inference, while Groq LPU, AWS Inferentia, and Qualcomm’s announced AI200 and AI250 focus exclusively on inference. The Qualcomm systems cannot be bought yet. Qualcomm states commercial availability in 2026 for the AI200 and 2027 for the AI250.⁶

This distinction matters for buyers: GPUs offer flexibility across different AI workloads, while ASICs deliver better performance per watt but are harder to reprogram as model architectures change.

According to TrendForce⁷ , AI server shipments built on custom ASICs from cloud providers were projected in October 2025 to grow 44.6% in 2026, against 16.1% for GPU-based AI servers. Those figures are AI server shipment growth rates split by accelerator type, not counts of chips shipped. TrendForce has since revised the 2026 mix down, putting ASIC-based systems at about 27% of AI server unit shipments and GPU-based systems at 69.7%, with ASICs reaching roughly 40% by 2030. The ASIC share was revised from 27.8% on April 15, 2026, citing chip validation and tuning delays at Meta and AWS.⁸

Which are the leading AI chip producers?

1. NVIDIA

NVIDIA has been designing graphics processing units (GPUs) for the gaming sector since the 1990s. NVIDIA is a fabless chip manufacturer that outsources most of its chip manufacturing to TSMC. Its main businesses include:

Desktop AI solutions

DGX Spark (formerly Project Digits) is a desktop AI supercomputer for AI engineers and data scientists featuring a Grace Blackwell Superchip with an NVIDIA Blackwell RTX GPU with 6,144 CUDA cores and fifth-generation Tensor Cores with FP4 precision, connected via the NVIDIA NVLink-C2C chip-to-chip interconnect to a high-performance, 20-core NVIDIA Grace CPU, with up to 1 petaflop of AI compute and 128GB of unified memory for on-device agents.^{9 10}

NVIDIA and Microsoft are partnering to deliver a secure Windows platform for on-device agents built on new OS security primitives.¹¹

Datacenter solutions

The company makes AI chips following its Ampere, Hopper, and, most recently, Blackwell architectures. Thanks to the generative AI boom, NVIDIA had excellent results in the past years, reached a trillion in valuation, and solidified its status as the leader of the GPU and AI hardware markets. The following chart shows how NVIDIA’s revenue in this segment has grown over the years and how it has become the company’s primary source of income.

Source: NVIDIA Corporation financial reports.¹²

DGX™ A100 and H100 have been successful flagship AI chips of Nvidia, designed for AI training and inference in data centers. NVIDIA followed up on these with

• H200, B300 and GB300 chips
• HGX servers such as HGX H200 and HGX B300 that combine 8 of these chips
• NVL series and GB200 SuperPod that combine even more chips into large clusters.¹³

Cloud GPUs

Thanks to the strength of its datacenter offering, most cloud players offer NVIDIA hardware as their cloud GPUs. Morgan Stanley analyst Joseph Moore, in a note dated March 2, 2026, put NVIDIA at roughly 85% of AI processor revenue, with custom ASICs above 10% and AMD below 5%.¹⁴ Bloomberg Intelligence, in a January 14, 2026 release, expects NVIDIA to hold 70% to 75% of the AI accelerator market through 2030.¹⁵

NVIDIA also launched its DGX Cloud offering, providing cloud GPU infrastructure directly to enterprises, bypassing cloud providers.

GPUs for graphics

Xbox uses a chipset co-developed by NVIDIA and Microsoft. NVIDIA’s GPUs for retail users include the GeForce series.

Recent developments

DGX Cloud Lepton

Announced on May 19, 2025, at Computex, NVIDIA’s DGX Cloud Lepton is a marketplace that connects AI developers to NVIDIA’s GPU cloud providers, including CoreWeave, Lambda, and Crusoe. It enables flexible access to GPU resources for AI model training and inference, bypassing traditional cloud provider dependencies. This strengthens NVIDIA’s enterprise-focused cloud strategy.¹⁶

NVIDIA Dynamo

NVIDIA Dynamo, announced at GTC 2025, is a new open-source inference framework designed for high-throughput, low-latency deployment of generative AI models in distributed environments, boosting request serving by up to 30x on NVIDIA Blackwell as shown in the figure below. This framework, compatible with popular tools like PyTorch and TensorRT-LLM, utilizes innovations such as disaggregated inference stages and dynamic GPU scheduling to optimize performance and reduce costs. Available on GitHub for developers and included in NVIDIA NIM microservices for enterprise solutions, Dynamo facilitates scalable and cost-effective generative AI serving from single to multi-GPU systems.¹⁷

Figure 1. NVIDIA Dynamo significantly accelerates AI model performance. Specifically, it provides a 30x speedup for the DeepSeek-R1 671B model on the NVIDIA GB200 NVL72 platform. It also more than doubles the performance of the Llama 70B model when using NVIDIA Hopper GPUs.¹⁸

NVIDIA RTX PRO Servers and Enterprise AI Factory

Announced in May 2025 at Computex, NVIDIA introduced RTX PRO Servers powered by RTX PRO 6000 Blackwell Server Edition GPUs, designed for enterprise AI factories. These servers deliver universal acceleration for AI, design, engineering, and business applications, supporting workloads like multimodal AI inference, physical AI, and digital twins on the NVIDIA Omniverse platform.

The NVIDIA Enterprise AI Factory validated design, incorporating RTX PRO Servers, NVIDIA Spectrum-X Ethernet, NVIDIA BlueField DPUs, and NVIDIA AI Enterprise software, enables partners like Cadence, Foxconn, and Lilly to build on-premises AI infrastructure. This initiative accelerates the trillion-dollar IT industry transition to GPU-accelerated AI factories. ¹⁹

NVIDIA Vera Rubin platform

NVIDIA unveiled Vera Rubin, its platform after Blackwell Ultra, at CES 2026 as a six-chip platform.²⁰ At GTC on March 16, 2026 NVIDIA expanded it to seven chips, the Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU, Spectrum-6 Ethernet switch and Groq 3 LPU, and declared them in full production.²¹ Production shipments begin in fall 2026, initially to eight cloud partners, NVIDIA said on May 31, 2026.²² Blackwell and the GB300 NVL72 remained the revenue volume product through Q1 FY2027.

The Vera Rubin NVL72 rack pairs 72 Rubin GPUs with 36 Vera CPUs for 3,600 PFLOPS of NVFP4 inference, 20.7 TB of HBM4 and 260 TB/s of NVLink 6 scale-up bandwidth.²³ The Rubin GPU is reported to be built on TSMC’s N3P with HBM4, a node NVIDIA has not confirmed.

The seventh chip comes from the Groq transaction. Groq announced a non-exclusive licensing agreement with NVIDIA for its inference technology on December 24, 2025, and founder and CEO Jonathan Ross joined NVIDIA with other senior team members.²⁴ CNBC reported the deal at about $20 billion in cash, and at the reported ~$20 billion it would be NVIDIA’s largest transaction to date.²⁵

The Groq 3 LPU is a dedicated inference chip that NVIDIA’s technical blog describes as “the seventh chip of the Vera Rubin Platform”, and Samsung Foundry manufactures it.^{26 27} Groq 3 LPX racks are slated for the second half of 2026, and NVIDIA projects up to 35x higher throughput per megawatt for trillion-parameter models when LPX is paired with Vera Rubin NVL72, a figure NVIDIA footnotes as “Projected performance subject to change”.²⁸

DeepSeek

Release of DeepSeek’s R1 showed that state-of-the-art models could be trained with a relatively small number of GPUs. This led to a reduction in NVIDIA’s stock price. Though this is not investment advice, this can be positive for NVIDIA since the more utility computing power provides, the more widely it should be used (i.e., Jevons paradox²⁹ ).

However, given that GPU systems’ performance improves multiple times annually due to advances in chip design and interconnect, buyers would be wise not to buy beyond their annual needs, as this can lead to owning outdated systems.

Tariffs & export restrictions

Four separate measures govern advanced AI chip sales to China, and they are frequently conflated.

Licensing changed first. On January 13, 2026 the US Bureau of Industry and Security announced a final rule moving license applications for the NVIDIA H200, AMD MI325X and similar chips from a presumption of denial to case-by-case review, subject to conditions covering exporter certifications, know-your-customer checks and independent pre-shipment testing.³⁰ A volume cap runs alongside it: aggregate shipments of a given chip to China and Macau may not exceed 50% of the quantity of that same product the exporter has shipped to US customers for US end-use.³¹

Two further measures are both set at 25% and are routinely treated as one. President Trump announced a 25% US government revenue share on H200 sales to China on December 8, 2025 and restated it on January 14, 2026, while the White House announced a distinct 25% Section 232 tariff on advanced semiconductors at the same performance thresholds on January 14, 2026, with carve-outs including US data centres and US R&D.³²

China has moved in the other direction, though less formally. Reuters reported on November 5, 2025, citing unnamed sources, that authorities had barred foreign AI chips from state-funded data centres, and no official Chinese regulation to that effect has been published.

Permission has not translated into sales. On July 14, 2026, Under Secretary of Commerce Jeffrey Kessler told a congressional hearing that “very few shipments against licenses for H200s and equivalents have taken place. It’s a very small quantity of chips.” NVIDIA had booked zero China data center compute revenue as of both its February 2026 and its May 2026 results.

Inference Market Competition

While NVIDIA dominates the AI “training” market, competition is heating up in “inference,” the deployment of AI models for real-world tasks. Companies like AMD and Qualcomm, alongside inference specialists including Cerebras, SambaNova, d-Matrix and Positron, are developing chips that aim to provide more cost-effective inference solutions, with a particular focus on lower power consumption. The field has thinned as well as grown: Untether AI shut down in June 2025 and filed for bankruptcy that October, and Groq’s inference technology was licensed to NVIDIA in December 2025.

New “reasoning” AI techniques demand more computing power. NVIDIA believes that reasoning will favor its architecture in the long run and expects the inference market to eventually dwarf the training market in size, even if its market share is smaller. ³³

2. AMD

AMD is a fabless chip manufacturer with CPU, GPU, and AI accelerator products.

AMD launched MI300 for AI training workloads in June 2023 and is competing with NVIDIA for market share. Startups, enterprises and tech giants adopted AMD hardware in 2023, when NVIDIA hardware was difficult to procure amid the generative AI boom triggered by the launch of ChatGPT.^{34 35 36}

In 2025, AMD acquired a team of AI hardware and software engineers from Untether AI, a developer of energy-efficient AI inference chips, along with compiler startup Brium, strengthening its AI compiler, kernel development and chip design work. In June 2026 AMD announced the acquisition of MEXT, whose predictive-memory technology is designed to make flash behave more like DRAM and cut data center memory cost; AMD disclosed no price.^{37 38}

AMD launched the Instinct MI350X and MI355X on June 12, 2025 at Advancing AI 2025.³⁹ The MI355X is built on CDNA4 at TSMC 3nm with 288 GB of HBM3E and 10.1 PFLOPS of MXFP4, and AMD positions it against NVIDIA’s Blackwell B200 rather than the H200. AMD followed on May 7, 2026 with the Instinct MI350P, a passive PCIe 5.0 card carrying 144 GB of HBM3E and a 600 W maximum TBP.⁴⁰

AMD has signed four large-scale Instinct agreements since October 2025. It announced a definitive multi-generation agreement with OpenAI on October 6, 2025 covering 6 GW, with the first 1 GW of MI450 Series capacity from the second half of 2026.⁴¹ It named Oracle Cloud Infrastructure launch partner on October 14, 2025 for an initial 50,000 MI450 Series GPUs from calendar Q3 2026.⁴² It agreed with Meta on February 24, 2026 to deploy up to 6 GW from the second half of 2026, using a custom Instinct GPU based on the MI450 architecture rather than a stock MI450.^{43 44} And it announced a partnership with Anthropic on July 22, 2026 to deploy up to 2 GW of MI450 Series GPUs in Helios racks from the first half of 2027, with a stated commitment to invest up to $5 billion in Anthropic in the future.⁴⁵

AMD is also working with Hugging Face to enable data scientists to use its hardware more efficiently.⁴⁶

The software ecosystem is critical as hardware performance relies heavily on software optimization. After a public AMD and NVIDIA disagreement over benchmarking H100 and MI300, the latest benchmarks put MI300 better or on par with H100 for inferencing on a 70B LLM.⁴⁷

MI400 series

AMD launched the Instinct MI400 Series at Advancing AI 2026 on July 22-23, 2026 in San Francisco.⁴⁸ The MI455X is the AI-focused member of the family, with a product page launch date of July 23, 2026. AMD lists it as CDNA5 on “TSMC 2nm | 3nm FinFET” with 320 billion transistors, 432 GB of HBM4, 23.3 TB/s of peak memory bandwidth and 40.3 PFLOPs of OCP MXFP4.⁴⁹ Lisa Su said Helios, the rackscale system built around the MI400 Series, is in full production with first shipments starting at the end of Q3 2026, so nothing had shipped as of the event.⁵⁰

A third family member, the MI440X, was introduced at CES 2026 on January 5, 2026 for on-premises enterprise AI in a compact, eight-GPU form factor, and AMD has published no specifications for it since.⁵¹ Further out, AMD’s roadmap places the Instinct MI500 Series in 2027 and the MI600 Series in 2028, with the CDNA 6 architecture, 2nm process technology and HBM4E memory detail for MI500 given at CES 2026.⁵²

The MI430X is the HPC and sovereign-AI member of the MI400 Series. AMD announced it on November 19, 2025 during SC25 week rather than at Advancing AI 2026, giving 432 GB of HBM4 and 19.6 TB/s at the time, and updated the numbers on July 23, 2026. AMD’s product page now lists up to 288 TFLOPs of hardware-based peak theoretical FP64, up to 9.2 PFLOPs of peak FP4 and MXFP4, 432 GB of integrated HBM4 and up to 23.3 TB/s of peak theoretical memory bandwidth.⁵³ The distinguishing feature against the MI455X is full-rate FP64 for scientific computing rather than low-precision AI throughput.

AMD states that the MI430X is “expected to be available in 2027”, so it is an announced part rather than a shipping one, and it has published no transistor count, board power or lithography for it. Two figures in AMD’s own materials disagree: the MI400-series landing-page FAQ still states up to 19.6 TB/s where the MI430X specification page gives 23.3 TB/s, and the FAQ on the MI430X page contains a typo reading 2.3 TB/s.⁵⁴

AMD names three target systems for the MI430X: Discovery at Oak Ridge National Laboratory, planned for 2028; Alice Recoque, France’s first exascale system; and Herder at HLRS Stuttgart, announced with HPE on December 2, 2025. Lux, the Oak Ridge AI factory cluster announced on October 27, 2025, is built on MI355X and is not an MI430X system.

Software

AMD’s AI software runs on the ROCm stack. SemiAnalysis, in a December 22, 2024 benchmark of MI300X against H100 and H200, gave AMD a 0% chance of breaking NVIDIA’s CUDA moat, finding that CUDA worked out of the box for most tasks while AMD software required significant configuration. ⁵⁵

The same analysts have since revised that call, first to a meaningful chance in April 2025 and then, on July 25, 2026, to “a great chance of success as long as AMD solves the two major risks we outline below”.⁵⁶ The risks it names are the Helios rack production ramp, where weak SerDes require up to 85% of the backplane to be retimed with over 550 Broadcom ethernet retimers per rack, and a persistent lack of stable internal GPU clusters for software development and automated-testing CI.

At Advancing AI 2026 AMD committed to a fixed six-week feature-release cadence in place of its previous roughly quarterly cycle,⁵⁷ and introduced ROCm.ai, an AI-driven development platform intended to let coding agents such as Claude, Codex and Cursor understand AMD platforms and ROCm natively.

Ecosystem

Like NVIDIA, AMD is selectively investing in users of its solutions to drive adoption of its hardware. ⁵⁸

Since late 2025 AMD has also tied equity to its largest customer commitments, issuing OpenAI and Meta each a performance-based warrant for up to 160 million AMD common shares at $0.01 per share, vesting in tranches against Instinct shipment milestones.^{59 60}

3. Intel

Intel is the most significant player in the CPU market and has a long history of semiconductor development. Unlike NVIDIA and AMD, Intel uses its own foundry to build its chips.

Gaudi 3 is Intel’s most recent AI accelerator and has no announced successor. ⁶¹ Intel’s sales guidance for Gaudi3 was ~$500M for 2024, a target Intel publicly abandoned in November 2024.⁶² Anil Nanduri, Intel’s vice president of product management and go-to-market for data center AI accelerators, said at CES 2026 that Intel “did not meet the needs of the frontier AI training market, and we didn’t meet the market needs.”⁶³

Under new CEO Lip-Bu Tan (appointed March 2025), Intel’s AI strategy has clarified around rack-scale solutions.⁶⁴ Intel cancelled its Falcon Shores GPU to pivot to Jaguar Shores, a rack-scale AI accelerator whose confirmed details are limited to the Gaudi branding and SK hynix HBM4 memory, with no process node stated.⁶⁵ At CES 2026 Intel’s roadmap referred to “the Shores product line” without the Jaguar name, and an Intel spokesperson repeated that nomenclature to CRN, which CRN reports as an observation because Intel has not announced a rename. Intel made fresh AI hardware announcements at Computex 2026 including its Xeon 6+ processor on the 18A node and the Crescent Island data center GPU.^{66 67}

Crescent Island is an Xe3P-based, inference-optimized data center GPU, first announced on October 14, 2025 at the OCP Global Summit with 160GB of LPDDR5X and customer sampling expected in the second half of 2026.⁶⁸ Intel detailed it at Computex 2026 as a 350W air-cooled PCIe card supporting data types from FP4 through FP64, with up to 480GB as the maximum partners may build to rather than the card’s own capacity.⁶⁹

4. Qualcomm

Qualcomm is a fabless designer best known for mobile SoCs, and it has announced a data center product line built around inference. Every product in that line carries a future availability date, and only the older Cloud AI 100 family is available to buy today.

AI200, AI250 and the data center roadmap

On October 27, 2025, Qualcomm announced the AI200 and AI250, two rack-scale AI inference solutions now marketed under the Dragonfly brand.⁷⁰ Qualcomm’s release states that the AI200 “supports 768 GB of LPDDR per card” and will be “commercially available in 2026”, while the AI250 debuts a memory architecture “based on near-memory computing” delivering “greater than 10x higher effective memory bandwidth” and will be “commercially available in 2027”. Both racks use direct liquid cooling and 160 kW of rack-level power. The 10x figure is a vendor claim about an unreleased product.

At its Investor Day on June 24, 2026, Qualcomm unveiled three further data center products.⁷¹ The Dragonfly C1000 is a data center CPU expected to reach commercial availability in 2028, and Qualcomm named Meta as its customer under a multi-generation collaboration covering data center CPUs rather than AI accelerators. Qualcomm High Bandwidth Compute (HBC) is a near-memory architecture that “bonds compute with highly-accelerated memory bandwidth in a 3D-stacked silicon solution”, making it a compute architecture rather than a memory part, and HBC Gen 1, paired with the AI250, is expected to reach commercial sampling in mid-2027. The Dragonfly AI300 is an inference accelerator with commercial sampling expected in 2028. At the same event Qualcomm set a data center revenue target of “more than $15 billion by fiscal 2029”, a company target rather than a reported result.⁷²

Qualcomm expects initial shipments to a hyperscaler custom silicon customer, which it has not named, later in calendar 2026. CEO Cristiano Amon repeated that expectation with the Q2 FY2026 results on April 29, 2026, which is a forward-looking statement rather than confirmation that shipments have taken place.

Acquisitions and the Cloud AI 100 family

Qualcomm completed its acquisition of Alphawave on December 18, 2025. On June 24, 2026 it announced that it “has reached an agreement to acquire Modular Inc”, a transaction expected to close in the second half of 2026 subject to regulatory approvals, so that acquisition has not completed.⁷³

Before the Dragonfly line, Qualcomm’s data center inference accelerators were the Cloud AI 100 cards. In Qualcomm’s own comparison table, the Cloud AI 100 Pro PCIe card draws 75W and is rated at up to 400 TOPS INT8, a data center part rather than an edge chip.

Which public cloud providers produce AI chips?

5. AWS

AWS produces Trainium chips for model training and Inferentia chips for inference. Although AWS is the market leader in public cloud, it began developing its own chips after Google.

Hundreds of thousands of Trainium2 chips are used to form the Project Rainier cluster, which powers LLM developer Anthropic’s models. The count at the cluster’s October 2025 launch was nearly half a million chips, and Anthropic stated on April 20, 2026 that it uses over one million Trainium2 chips.⁷⁴

Trainium3 and Trn3 UltraServers reached general availability on December 2, 2025.⁷⁵ Trainium3 is AWS’s fourth-generation AI chip and its first built on a 3nm process, rated at 2.52 PFLOPS of FP8 compute, 144 GB of HBM3e and 4.9 TB/s of memory bandwidth per chip. A Trn3 UltraServer scales to 144 chips for 362 FP8 PFLOPS, 20.7 TB of HBM3e and 706 TB/s of memory bandwidth, up to 4.4x the performance of a Trn2 UltraServer.

AWS announced Trainium4 as a roadmap item at re:Invent 2025, with support for NVIDIA NVLink Fusion. On the Q1 2026 earnings call, Amazon CEO Andy Jassy said Trainium4 is about 18 months from broad availability and is largely reserved already.⁷⁶

6. Google Cloud Platform

Google Cloud TPU is the purpose-built machine learning accelerator chip that powers Google products like Translate, Photos, Search, Assistant, and Gmail. Google announced TPUs in 2016.⁷⁷ Trillium TPU is the 6th generation.⁷⁸

Google has introduced Ironwood. This generation is specifically designed for complex “thinking models” like LLMs and MoEs, offering massive parallel processing (4,614 TFLOPS of FP8 compute per chip) and scaling up to 42.5 Exaflops in 9,216-chip pods.⁷⁹ Ironwood, also called TPU v7 or TPU7x, entered preview on November 24, 2025 and reached general availability on March 31, 2026.⁸⁰

Ironwood delivers significant advancements over Trillium, including 2x better power efficiency, 6x the High Bandwidth Memory capacity (192 GB/chip), 4.5x the HBM bandwidth (7.37 TB/s per chip), and 1.5x the Inter-Chip Interconnect speed (1.2 TB/s bidirectional, 9.6 Tb/s).⁸¹ It also features an enhanced SparseCore for large embeddings. Google also produces the much smaller Edge TPU for different needs, designed for deployment on edge devices like smartphones and IoT hardware.

At Cloud Next on April 22, 2026, Google announced its eighth TPU generation as two purpose-built chips, TPU 8t for training and TPU 8i for inference.⁸² A TPU 8t superpod scales to 9,600 chips, two petabytes of shared HBM and 121 ExaFLOPS at FP4. Both chips are announced rather than shipping, and Google says they will be generally available later in 2026.

7. Alibaba

Alibaba designs AI accelerators through its chip unit T-Head (PingTouGe), on a self-developed parallel processing unit (PPU) architecture with a custom ICN interconnect.⁸³

T-Head publicly detailed the Zhenwu 810E on January 29, 2026, alongside the launch of its official website. The chip carries 96 GB of HBM2e and 700 GB/s of inter-chip bandwidth across 7 independent ICN links, performance the company describes as comparable to NVIDIA’s H20, and Alibaba Cloud deploys it in 10,000-card clusters. This was a specification disclosure rather than a product launch. The 810E is not new 2026 silicon, and the 10,000-card deployments predate the announcement.

At the Alibaba Cloud Summit in Hangzhou on May 19-20, 2026, Alibaba announced the Zhenwu M890, with 144 GB of memory, 800 GB/s of inter-chip interconnect bandwidth, native support from FP32 down to FP4, and three times the performance of the 810E. It ships inside the Panjiu AL128 supernode, which holds 128 accelerators per rack and uses the ICN Switch 1.0 at 25.6 Tbps. Alibaba states that cumulative deliveries of the Zhenwu series exceed 560,000 units, adopted by more than 400 external customers across 20 industries. That figure covers the whole series, not M890 units.

Alibaba’s earlier Hanguang 800 inference chip dates from 2019. However, some North American, European, and Australian organizations (e.g., those in the defense industry) may not prefer to use Alibaba Cloud for geopolitical reasons.

8. IBM

IBM announced its deep learning chip, the artificial intelligence unit (AIU), in 2022.⁸⁴ . IBM is considering using these chips to power its Watsonx generative AI platform.⁸⁵

The IBM AIU builds on the IBM Telum Processor, which powers the AI processing capabilities of IBM Z mainframe servers. At launch, Telum processors’ highlighted use cases included fraud detection.⁸⁶

IBM also demonstrated that merging compute and memory can lead to efficiencies. These were demonstrated in the North Pole processor prototype.⁸⁷

The IBM AI accelerator that ships commercially is the Spyre Accelerator, announced on October 7, 2025 and generally available from October 28, 2025 on z17 and LinuxONE 5, and from early December 2025 on Power11.⁸⁸ Spyre carries 32 accelerator cores and 25.6 billion transistors on a 5nm process, packaged as a 75W PCIe card with 128 GB of LPDDR5. Up to 48 cards cluster in a Z or LinuxONE system, against 16 in Power.

9. Huawei

Huawei’s HiSilicon Ascend 910C is part of the Ascend 910 family of chips introduced in 2019.

Due to sanctions, AI labs in China can not buy the newest, highest-performance chips from US firms like NVIDIA or AMD. Therefore, they are experimenting with Ascend 910C. Huawei’s cloud is hosting DeepSeek models, and a researcher at DeepSeek claims that it can reach 60% of NVIDIA H100 inference performance. ⁸⁹

Huawei debuted the Ascend 950PR chip alongside the Atlas 350 accelerator card at the Huawei China Partner Conference 2026 in Shenzhen on March 20, 2026.⁹⁰ The Atlas 350 card is specified at 1.56 PFLOPS of FP4 compute, 112 GB of Huawei’s proprietary HiBL 1.0 HBM at 1.4 TB/s of memory bandwidth, 600 W of power (about 1.5 times the H20) and 2 TB/s of interconnect. Those memory figures belong to the card. The Ascend 950PR silicon itself is specified at 128 GB and 1.6 TB/s. Zhang Dixuan, head of Huawei’s Ascend computing business, claims the card delivers 2.8 times the performance of NVIDIA’s H20 (2.87 times in some reports). That is a vendor claim, it has not been independently verified, and it is not a like-for-like comparison, because the Hopper-class H20 has no native FP4 support.

The in-house memory is Huawei’s own statement. At Huawei Connect 2025 in Shanghai on September 18, 2025, Deputy Chairman Eric Xu said the 950 series would use “two proprietary HBMs… HiBL 1.0 and HiZQ 2.0”, describing HiBL 1.0 as “our proprietary, low-cost HBM that is more cost-effective than HBM3E and HBM4E”.⁹¹

Huawei has published one Ascend roadmap, in that same keynote. It lists Ascend 950PR in Q1 2026, Ascend 950DT in Q4 2026, Ascend 960 in Q4 2027 and Ascend 970 in Q4 2028. The Ascend 950DT is specified for 144 GB of HiZQ 2.0 memory, 4 TB/s of memory bandwidth, 2 TB/s of interconnect and decode plus training workloads, but as of July 2026 it is a roadmap commitment rather than a launched product. Huawei showed the Atlas 950 SuperPoD at WAIC 2026 on July 18, 2026, with commercial rollout still set for Q4 2026. Reports of an “Ascend 920” trace back to April 2025 supply-chain coverage (DigiTimes Asia, April 17, 2025; Commercial Times, April 21, 2025) describing a rumored SMIC N+3 / 6nm replacement for the H20. Huawei has never announced, specified or shipped such a part, and it appears nowhere on the published roadmap.

Which cloud AI providers produce their own chips?

These providers do not have public clouds with comprehensive capabilities like the hyperscalers. They provide limited cloud services, typically focused on AI inference. We were able to sign up for these services without talking to sales teams:

10. Groq

Groq was founded by former Google employees and built its business around the LPU (Language Processing Unit), a chip architecture aimed at low-latency inference rather than training.

On December 24, 2025, Groq announced that it had entered a non-exclusive licensing agreement with NVIDIA covering Groq’s inference technology, and that founder and CEO Jonathan Ross, president Sunny Madra and other senior team members had joined NVIDIA.⁹² Groq stated that it continues as an independent company. CNBC put the transaction at a reported ~$20 billion in cash, a figure attributed to Alex Davis, CEO of Groq investor Disruptive, who also said NVIDIA acquires all of Groq’s assets except the GroqCloud business.⁹³ NVIDIA has published no press release on the deal, NVIDIA CFO Colette Kress declined to comment on the price, and Reuters noted that neither company confirmed it. Jensen Huang told employees, “While we are adding talented employees to our ranks and licensing Groq’s IP, we are not acquiring Groq as a company.” NVIDIA unveiled the resulting product, the NVIDIA Groq 3 LPU, at GTC on March 16, 2026 as the seventh chip of the Vera Rubin platform.⁹⁴

Groq now operates as an inference cloud provider, chaired by Alex Davis of Disruptive and led by CEO Adam Winter. In June 2026 the company said it runs 13 data centers across North America, Europe, the Middle East and APAC, and is scaling its inference cloud toward 200 MW of capacity by 2027.⁹⁵

Groq has raised roughly $2.4 billion in disclosed funding and shipped products including the GroqChip™ Processor and the GroqCard™ Accelerator. Its $640 million Series D, led by BlackRock Private Equity Partners and announced on August 5, 2024, valued the company at $2.8 billion.⁹⁶ It announced a further $750 million at a $6.9 billion post-money valuation on September 17, 2025, led by Disruptive.⁹⁷ On June 22, 2026 it announced $650 million in growth capital led by Disruptive and Infinitum, with reinvesting existing investors, without disclosing a valuation, which leaves $6.9 billion as its last confirmed valuation.

The company is focused on LLM inference and released benchmarks for Llama-2 70B.⁹⁸

Recently, Groq secured a significant $1.5 billion investment commitment from Saudi Arabia to expand the delivery of its advanced AI chips to the country. This investment will be used to expand Groq’s existing data center in Dammam, Saudi Arabia, built in partnership with Aramco Digital.⁹⁹

In its first quarter of 2024, Groq reported more than 70,000 new developers on its cloud platform and more than 19,000 new applications built on it.¹⁰⁰ By June 2026 the company said it serves more than five million developers.¹⁰¹

On March 1, 2022, Groq acquired Maxeler, which has high-performance computing (HPC) solutions for financial services.¹⁰²

11. SambaNova Systems

SambaNova Systems was founded in 2017 to develop high-performance, high-precision hardware-software systems for high-volume generative AI workloads. It raised a 350 million dollar Series E round in February 2026.¹⁰³ On July 8, 2026 it completed the first close of a $1B Series F at an $11 billion post-money valuation, led by General Atlantic.¹⁰⁴

In February 2026, SambaNova unveiled the SN50 chip, its latest Reconfigurable Data Unit (RDU), claiming a max speed 5x faster than competitive chips and 3x lower total cost of ownership compared to GPUs for agentic AI workloads. The SN50 delivers 5x more compute per accelerator and 4x more network bandwidth than the previous generation SN40L, and supports a three-tier memory architecture for models up to 10 trillion parameters and context up to 10 million tokens.¹⁰⁵ The SN50 was announced on February 24, 2026, with shipments to customers due in the second half of 2026. As of July 2026 no SN50 has shipped.

SoftBank Corp. will be the first customer to deploy SN50 within its next-generation AI data centers in Japan.

SambaNova also announced a planned multi-year strategic collaboration with Intel to deliver AI inference solutions, combining SambaNova’s systems with Intel Xeon processors, Intel GPUs, and Intel networking to power scalable inference infrastructure as an alternative to GPU-centric solutions.

SambaNova Systems leases its platform to businesses through SambaCloud. This AI platform-as-a-service approach makes their systems easier to adopt and encourages hardware reuse for the circular economy.¹⁰⁶

Which are the leading AI chip startups?

We would also like to introduce some startups in the AI chip industry whose names we may hear more often in the near future.

12. Cerebras

Cerebras was founded in 2015 and is the only major chip maker focusing on wafer-scale chips. ¹⁰⁷ Wafer-scale chips have advantages in parallelism compared to GPUs, thanks to their higher memory bandwidth. However, designing and manufacturing such chips is an emerging technology.

Cerebras chips include:

• WSE-1 with 1.2 trillion transistors and 400k processing cores.
• WSE-2, with 2.6 trillion transistors and 850k cores, was announced in April 2021. It leveraged TSMC’s 7nm process
• WSE-3, featuring 4 trillion transistors and 900k AI cores, was announced in March 2024. It leverages TSMC’s 5nm process¹⁰⁸

WSE-3 is the current shipping generation, and Cerebras has not announced a successor.

Cerebras’ system works with pharmaceutical companies such as AstraZeneca and GlaxoSmithKline and research labs that rely on it for simulations. It also targets LLM makers since its chips can lower inference costs for frontier models.

Cerebras also offers its chips on its cloud to enterprises.

Cerebras is a public company. It priced its IPO at $185.00 per share on May 13, 2026 and began trading on Nasdaq under the ticker CBRS on May 14, 2026, selling 34.5 million Class A shares for approximately $6.38 billion in gross proceeds, which the company describes as the largest semiconductor IPO of all time.¹⁰⁹ The stock closed its first trading day at $311.07, up 68%.

Q1 2026 revenue was $193.4 million, up 94% year over year, and Cerebras guides FY2026 core revenue to $855-865 million.¹¹⁰ Cerebras signed a 750 MW inference agreement with OpenAI on January 14, 2026, which its own Q1 release values at more than $20 billion.¹¹¹

13. d-Matrix

d-Matrix follows a novel approach, ditching the traditional von Neumann architecture in favor of in-memory computing. While this approach has the potential to resolve the bottleneck between memory and compute, it is new and unproven. In November 2025, d-Matrix raised $275M in a Series C co-led by Bullhound Capital, Triatomic Capital, and Temasek, with Microsoft’s M12 participating as a follow-on investor, valuing the company at $2 billion.^{112 113}

Corsair, the company’s AI inference platform, entered full production on June 9, 2026, built on an SRAM-based in-memory compute chiplet architecture.¹¹⁴

The 10x speedup attached to Corsair comes from partner-run benchmarking rather than independent testing. The tester, Gimlet Labs, is a commercial partner of d-Matrix, and the benchmark post carries d-Matrix’s co-founder and CTO on the byline. The measured result is a 2-10x end-to-end request-latency improvement on gpt-oss-120b for one speculative-decoding configuration, not a blanket 10x over GPUs.

14. Rebellions

A Korea-based startup raised $124M in 2024 and is focused on LLM inference.¹¹⁵

Rebellions merged with another Korean semiconductor design firm, SAPEON, and reached a unicorn valuation in 2024.¹¹⁶

In July 2025, Rebellions secured investment from tech giant Samsung as part of a funding round targeting up to $200 million, ahead of a planned initial public offering (IPO). ¹¹⁷

Rebellions closed a $250 million Series C in September 2025 and a $400 million pre-IPO round on March 30, 2026 led by Mirae Asset Financial Group and the Korea National Growth Fund, taking total funding to about $850 million at a valuation of roughly $2.34 billion.¹¹⁸

The IPO has not happened. CEO Sunghyun Park said on July 8, 2026 that Rebellions targets the first or second quarter of 2027, leaning toward a KOSPI main-board listing.¹¹⁹

REBEL-Quad is the second-generation accelerator, four chiplets built on Samsung’s 4-nanometer process, and Rebel100 is its productized commercial name.

15. Tenstorrent

Tenstorrent’s latest Blackhole Tensix Processor delivers 664 TFLOPS (BLOCKFP8) of performance, paired with 32GB of GDDR6 memory and 512 GB/s memory bandwidth.

The P150a card is priced at $1,399 and features four QSFP-DD 800G ports for multi-card scaling. The entry-level P100a model starts at $999.¹²⁰

From January 2026, all Blackhole p150 cards ship with 120 Tensix cores instead of 140, and firmware release v19.5.0 applied the same reduction retroactively to cards already sold. Tenstorrent puts the cost at 1-2% on typical workloads.¹²¹

Tenstorrent offers a fully open-source software stack. The company raised $700M at a valuation of more than $2.6 billion from investors, including Jeff Bezos, in December 2024. ¹²²

16. Positron

Positron was founded in 2023 and focuses exclusively on transformer model inference. The company takes an ASIC approach, building purpose-built hardware optimized specifically for transformer architectures rather than general-purpose GPU computing.

Products:

• Atlas (shipping now): A transformer inference server featuring 8x Positron Archer Transformer Accelerators with 256 GB total HBM. The company claims >4x performance per watt and >3x performance per dollar compared to NVIDIA Hopper systems, benchmarked on Llama 3.1 8B with BF16 compute.¹²³
• Titan (coming 2027): A next-generation system with 8+ TB memory powered by 4x Asimov custom chips, designed to support up to 16 trillion parameter models and 10 million+ token context windows in an air-cooled 4U form factor.¹²⁴
• Asimov (coming 2027): Custom inference accelerator silicon with 2+ TB memory per chip.

Positron raised a 230M+ Series B round in early 2026 with investors including QIA, Arm Holdings, Arena, and Jump Trading¹²⁵

Atlas is currently used by networking, gaming, content moderation, CDN, and Token-as-a-Service companies. Positron claims its Atlas system demonstrated 3x lower end-to-end latency for trading inference workloads versus comparable H100 systems while consuming one-third of the power.

Positron’s chips are designed, fabricated, and assembled in the United States.

17. _etched

Their approach sacrifices flexibility for efficiency by burning model architectures into their chips.

Etched emerged from stealth on June 30, 2026 with first-pass A0 silicon back from TSMC’s N4P process, more than $1 billion in signed customer contracts and $800 million raised.¹²⁶ On July 23, 2026 it announced a $300 million Series C at a $10.3 billion valuation, led by Sequoia.¹²⁷

The company no longer uses the Sohu product name. It now describes itself as building frontier inference clusters that run DeepSeek, Qwen, Mamba and Llama, so the design is no longer transformer-only.¹²⁸

Two questions stay open. Etched has published no third-party benchmark and names no customer behind the contracts it reports, so its performance figures remain internal measurements. The obsolescence question the original Sohu design raised now applies to the wider model set, since a chip built around a fixed group of architectures has to be re-targeted once those architectures are superseded.

18. Taalas

Taalas was founded in early 2023 and takes the most extreme approach to AI chip specialization: hard-wiring individual models directly into custom silicon, producing what the company calls “Hardcore Models.”¹²⁹ The company claims it can transform any previously unseen AI model into custom silicon within two months.

Taalas’ architecture unifies storage and compute on a single chip at DRAM-level density, eliminating the need for HBM, advanced packaging, 3D stacking, liquid cooling, or high-speed I/O. The company describes this as a radical simplification of the hardware stack.

Products:

• HC1 (available now): A technology demonstrator hard-wired with Llama 3.1 8B, built on TSMC 6nm with 53 billion transistors. Taalas claims 17,000 tokens per second per user, which it says is nearly 10x faster than the current state of the art, while costing 20x less to build and consuming 10x less power in a 2.5 kW air-cooled server. However, the model uses aggressive custom 3-bit and 6-bit quantization, which introduces quality degradations compared to GPU baselines.¹³⁰
• HC2 (roadmap target for winter 2026, not yet a product): A second-generation platform with higher density, faster execution, and standard 4-bit floating-point formats to address the quantization limitations of HC1.

Taalas has raised $219 million in total, including a $169 million round announced on February 19, 2026, and reports spending 30 million dollars to bring its first product to market with a team of 24 people.

19. Extropic

Extropic raised a $14M round in late 2023 to use thermodynamics for computing. Its hardware page lists three parts. X0 is a silicon prototype from Q1 2025, and XTR-0 is a development platform shipped in limited numbers to selected researchers in Q3 2025. Both were unveiled on October 29, 2025. Z1, the first production-scale chip, is listed as early access in 2026.¹³¹

20. Vaire

Vaire is a UK-based startup pioneering reversible computing, an innovative approach that aims to create near-zero energy chips. Unlike traditional computing, where energy is lost as heat, reversible computing recycles a significant portion of energy for subsequent computations.

Vaire has demonstrated a test chip that can recover 50% of its energy, showing the technology’s potential to reduce the energy consumption of AI workloads and circumvent the physical limitations, or thermal wall, that are challenging modern semiconductor manufacturing. ¹³²

21. Fractile

Fractile is a UK-based AI inference chip startup that emerged from stealth in July 2024 with $15 million in funding to challenge NVIDIA on frontier model inference.¹³³

The company is building processors that physically interleave memory and compute on the same die, which it claims solves the simultaneous low-latency and high-throughput requirement that GPUs cannot meet for frontier model inference. Fractile claims its design can run frontier models up to 25x faster and at 1/10th the cost of existing solutions, with the goal of serving thousands of tokens per second to thousands of concurrent users.

Fractile is headquartered in London with hardware engineering in Bristol, and has been profiled by the Financial Times in March 2025 as part of a wave of inference-focused startups challenging NVIDIA’s dominance.¹³⁴

Fractile announced a $220 million Series B on May 13, 2026, co-led by Accel, Factorial Funds and Founders Fund, and separately committed £100 million to UK operations in February 2026.¹³⁵ Reporting on the round put the valuation at about $1 billion, a figure the company has not confirmed.

What are the upcoming AI hardware producers?

Though these are compelling AI hardware solutions, there are currently limited benchmarks on their effectiveness since they are newcomers to the market.

22. Apple

Apple’s Project ACDC is reported to be focused on building chips for AI inference in data centers.¹³⁶ The server chip is reportedly codenamed Baltra and developed with Broadcom, and Apple has never officially acknowledged it. Reuters reported on July 15, 2026, citing The Information, that the project has been pushed back and that Apple’s in-house AI servers still run on M2 Ultra chips.^{137 138} Apple is already a major chip designer with its internally designed semiconductors used in iPhones, iPads, and MacBooks. M5 launched on October 15, 2025, and M5 Pro and M5 Max were announced on March 3, 2026 with a Neural Accelerator in each GPU core. Apple is strengthening its on-device AI strategy with the Core AI framework, which runs models entirely on Apple silicon with no server dependencies, supported by an open-source Core AI models repository on GitHub.^{139 140}

23. Meta

Meta Training and Inference Accelerator (MTIA) is a family of processors for AI workloads such as training Meta’s LLaMa models.

Meta renamed the MTIA family on March 11, 2026. The part the company previously called Next Gen MTIA is now MTIA 200. It is based on TSMC 5nm technology, is claimed to offer 3x the performance of MTIA v1, and is hosted in racks containing up to 72 accelerators.¹⁴¹ MTIA 300 is in production for ranking and recommendation training, and MTIA 400, 450 and 500 follow across 2026 and 2027.¹⁴²

MTIA is currently for Meta’s internal usage. However, in the future, if Meta launched an LLaMa-based enterprise generative AI offering, these chips could power such an offering.

On April 14, 2026 Meta and Broadcom announced a multi-year, multi-generation strategic partnership extending through 2029 in which Broadcom delivers technology supporting Meta’s MTIA chips. “The initial commitment exceeds 1GW and is the first phase of a sustained, multi-gigawatt rollout,” and Broadcom’s release states that the rollout includes “the industry’s first 2nm AI compute accelerator” alongside Broadcom’s Ethernet scale-up, scale-out and scale-across networking.¹⁴³ On its Q2 FY2026 earnings call Broadcom described the partnership as targeting 3 GW through the end of 2028, with the initial 1 GW order delivering from the second half of 2027.¹⁴⁴

24. Microsoft Azure

At Hot Chips 2024, Microsoft unveiled Maia 100, their first custom AI accelerator designed to optimize large-scale AI workloads in Azure through hardware and software co-optimization. Built on TSMC’s N5 process with advanced memory and interconnect technology, Maia 100 targets high throughput and diverse data formats, offering developers flexibility via its SDK for quick deployment of PyTorch and Triton models. Microsoft launched Maia 200 (codenamed Braga) on January 26, 2026, as an inference-focused AI accelerator for Azure, designed to reduce AI token costs and deliver 30% better performance per dollar over existing systems.¹⁴⁵

Maia 200 is built on TSMC 3nm with more than 140 billion transistors, 216 GB of HBM3e at 7 TB/s, 272 MB of on-chip SRAM and more than 10 petaFLOPS of FP4 compute within a 750 W SoC TDP. It is deployed in Azure’s US Central region near Des Moines, Iowa, with US West 3 near Phoenix next.¹⁴⁶

25. OpenAI

OpenAI designed its first AI chip with Broadcom, and the leadership of its chip team has experience designing TPUs at Google.¹⁴⁷ OpenAI has confirmed TSMC as the foundry, but neither OpenAI nor Broadcom has confirmed a process node.¹⁴⁸ Korean supply-chain reporting has used the codename “Titan” for the chip and placed it on TSMC’s N3 process, though neither company uses that name.¹⁴⁹

Samsung has secured an agreement to supply HBM4 memory for the OpenAI chip, reportedly allocating over 50% of its Pyeongtaek foundry capacity to HBM4 base dies for this purpose.¹⁵⁰ OpenAI and Broadcom announced a collaboration for 10 gigawatts of custom AI accelerators in October 2025, with OpenAI designing the accelerators and systems and Broadcom developing and deploying them. Racks are “targeted to start in the second half of 2026, to complete by end of 2029.” The announcement states that the parties “have signed a term sheet” for the racks, so it covers a term sheet alongside pre-existing co-development and supply agreements rather than a disclosed definitive purchase contract, and neither company has published a financial value for the collaboration.¹⁵¹ On June 24, 2026 the two unveiled Jalapeño, described as “OpenAI’s first Intelligence Processor,” co-developed “from initial design to manufacturing tape-out in just nine months.” Both companies state that “Engineering samples of the Jalapeño chip are running ML workloads in the lab at production target frequency and power, including GPT-5.3-Codex-Spark,” and that the platform is “designed for initial deployment by the end of 2026,” which places the chip at post-tape-out engineering-sample stage rather than in mass production. Celestica is named for board, rack and system expertise, while Broadcom supplies the silicon implementation and Tomahawk networking.¹⁵² On its Q2 FY2026 earnings call Broadcom said silicon has been delivered, production is expected in late 2026, and OpenAI is contracted for 1.3 GW in 2027 within the 10 GW-by-2029 arrangement.¹⁵³ Broadcom’s side of the arrangement, including its other custom-accelerator customers, is covered in the Broadcom section below.

What are other AI chip producers?

26. Graphcore

Graphcore is a British company founded in 2016. The company announced its flagship AI chip as IPU-POD256. Graphcore has already been funded with around $700 million.

The company has strategic partnerships with data storage corporations like DDN, Pure Storage, and Vast Data. Graphcore’s AI chips serve research institutes like Oxford-Man Institute of Quantitative Finance, University of Bristol and Berkeley University of California.

The company’s long-term viability was at risk as it was losing ~$200M per year.¹⁵⁴ SoftBank announced its acquisition of Graphcore on July 11, 2024, and neither party officially disclosed the price, reported at around $600 million.¹⁵⁵ SoftBank injected a further $457 million into Graphcore on April 10, 2026.¹⁵⁶

27. Mythic

Mythic was founded in 2012 and is focused on edge AI. Mythic follows an unconventional path, an analog compute architecture, that aims to deliver power-efficient edge AI computing.

It has developed products such as M1076 AMP and MM1076 key card, and has already raised about $165 million in funding.¹⁵⁷

Mythic laid off most of its staff and restructured its business with its funding round in March 2023.¹⁵⁸

Mythic raised an oversubscribed $125 million round led by DCVC on December 17, 2025, with Honda Motor and Lockheed Martin among the strategic investors, and acquired Videantis in May 2026.¹⁵⁹

28. Speedata

Founded in 2019 in Tel Aviv, Speedata develops an Analytics Processing Unit (APU) designed to accelerate big data analytics and AI workloads. It’s an APU that targets Apache Spark workloads, with plans to support other major data analytics platforms.

Speedata raised $44M in a Series B round in June 2025, led by Walden Catalyst Ventures, 83North, and others, bringing its total funding to $114M. The company claims its APU outperforms general-purpose processors and GPUs by replacing racks of servers with a single chip, offering superior performance and energy efficiency for data processing.¹⁶⁰

29. Axelera AI

Founded in July 2021 in Eindhoven, Netherlands, Axelera AI specializes in AI hardware acceleration technology for computer vision and generative AI. The company is developing Titania, an AI inference chiplet based on its Digital In-Memory Computing (D-IMC) architecture, designed to accelerate AI workloads from edge to cloud.

Axelera AI raised more than $250 million on February 24, 2026 in a round led by Innovation Industries, with BlackRock and SiteGround Capital joining and Bitfury, CDP Venture Capital, the European Innovation Council Fund, SFPIM and Invest-NL among existing investors. The company puts its total at more than $450 million in equity, grants and venture debt since incorporating in July 2021, and describes the round as the largest investment ever in an EU AI semiconductor company.^{161 162} That earlier total included up to €61.6 million from the EuroHPC Joint Undertaking and member states under the DARE Project in March 2025, and a $68 million Series B before it.

Its Europa AIPU, announced on October 21, 2025, is rated at up to 629 TOPS INT8 across 8 second-generation AI cores and 16 RISC-V processors at around 45 W. The launch release said shipments would begin in the first half of 2026; an Andes Technology release in June 2026 described Europa as sampling to lead customers. Titania, the D-IMC inference chiplet, remains targeted for 2028.

See more of our benchmarks and data-driven insights in Google Search.

Add as preferred source

Which vendors co-design custom AI ASICs?

Several accelerators listed above are not designed alone by the company whose name is on them. Broadcom co-designs custom accelerators, which it calls XPUs, with the customer that will operate them, and supplies the Ethernet switching and fabric silicon that connects them. It is a supplier to several vendors covered earlier in this article rather than a competitor to them, which is why it sits in a category of its own.

30. Broadcom

Broadcom does not sell a branded AI accelerator to the general market. Its AI business is custom silicon co-designed with a small number of large buyers, plus merchant networking silicon sold to the same buyers.

For Q2 FY2026 (quarter ended May 3, 2026, reported June 3, 2026), Broadcom reported record total revenue of $22,187 million, up 48% year over year. President and CEO Hock Tan stated: “Q2 semiconductor revenue from AI of $10.8 billion grew 143% year-over-year, above our forecast, driven by increasing demand for custom AI accelerators and AI networking.”¹⁶³

On the earnings call the same day, Tan gave figures that do not appear in the press release: “bookings for AI semiconductors were over $30 billion against the $10.8 billion we shipped”; full-year FY2026 AI semiconductor revenue “of $56 billion, up approximately 180% from fiscal 2025”; and, for FY2027, “we reiterate our AI semiconductor revenue guidance to be in excess of $100 billion”, which is a reiteration rather than a raise. Q3 FY2026 guidance is total revenue of about $29.4 billion, up 84%, and AI semiconductor revenue of $16.0 billion, up more than 200%.¹⁶⁴

On that call Broadcom walked through six core AI customers but named four of them, Google, Anthropic, OpenAI and Meta. The other two remain undisclosed and were described only as two customers with $6 billion in purchase orders received to date, with shipments beginning in late 2026 and accelerating into 2027. Published lists that name all six are analyst inference rather than a Broadcom disclosure.¹⁶⁵

Broadcom’s disclosure varies by customer:

• Google has a multi-generation TPU and AI-networking agreement, described on the call only as a “very, very substantial” dollar commitment, with no gigawatt figure given. Broadcom’s Form 8-K with an event date of April 6, 2026 discloses a Long Term Agreement under which Broadcom develops and supplies custom TPUs for Google’s future generations, plus a Supply Assurance Agreement covering networking components “through up to 2031”.¹⁶⁶
• Anthropic takes over 1 GW of TPU-based compute in 2026, plus an April agreement for a further 5 GW of next-generation compute beginning in 2027, according to the earnings call. The same Form 8-K states that “Broadcom, Google and Anthropic PBC have expanded their current strategic collaboration under which Anthropic, beginning in 2027, will access through Broadcom approximately 3.5 gigawatts as part of the multiple gigawatts of next generation TPU-based AI compute capacity committed by Anthropic.” The filing adds: “The consumption of such expanded AI compute capacity by Anthropic is dependent on Anthropic’s continued commercial success.” The 3.5 GW figure comes from Broadcom’s filing, while Anthropic’s own post quantifies the commitment as “multiple gigawatts”. The filing also frames the arrangement as an expansion of an existing collaboration rather than a new three-way contract.¹⁶⁷
• For OpenAI, silicon has been delivered, production is expected in late 2026, and OpenAI is contracted for 1.3 GW in 2027 within the larger 10 GW-by-2029 arrangement, according to the earnings call. The chip itself is covered in the OpenAI section above.¹⁶⁸
• Meta has a multi-generation MTIA XPU partnership targeting 3 GW through the end of 2028, with an initial 1 GW order delivering from the second half of 2027, according to the earnings call. The April 14, 2026 announcement describes a multi-year, multi-generation strategic partnership extending through 2029 in which Broadcom delivers technology supporting Meta’s MTIA chips: “The initial commitment exceeds 1GW and is the first phase of a sustained, multi-gigawatt rollout.” Broadcom’s release states that the rollout includes “the industry’s first 2nm AI compute accelerator” and is built on Broadcom’s XPU platform together with its Ethernet scale-up, scale-out and scale-across networking. Mark Zuckerberg is quoted in the release: “As we roll out more than 1GW of our custom silicon to start and then multiple gigawatts over time…” Hock Tan is transitioning off Meta’s board to an advisor role.¹⁶⁹

On February 26, 2026 Broadcom “announced it has begun shipping the industry’s first 2nm custom compute SoC built on its 3.5D eXtreme Dimension System in Package (XDSiP) platform”. That first SoC is for Fujitsu, supporting the FUJITSU-MONAKA processor program, rather than a hyperscaler XPU, and Broadcom says XPUs for its broader customer base ship from the second half of 2026.¹⁷⁰ In networking, Broadcom described Tomahawk 6 as “the first and only shipping 102.4T Tomahawk 6 now shipping in production volume” in its March 12, 2026 OFC release, and Jericho 4, a 51.2 Tbps fabric router launched in August 2025, is shipping and was cited on the Q2 FY2026 call as enabling the largest deployments across multiple hyperscalers.

On June 9, 2026 Broadcom announced the AI XPV Platform with Apollo (NYSE: APO) and Blackstone’s (NYSE: BX) Credit & Insurance Business as initial anchor investors. Broadcom’s release states that the Platform “is designed to enable more than 20 gigawatts in compute capacity using Broadcom’s XPUs and networking solutions customized for leading frontier AI labs, including Anthropic and OpenAI, through 2028”, and that it “launches today with an initial tranche of $35 billion led by Apollo, in partnership with Blackstone, to facilitate Anthropic’s previously-announced capacity expansion of more than 1 gigawatt of compute infrastructure expected to deploy in Fluidstack-based sites starting in mid-2026”. The $35 billion is the launch tranche and is tied to more than 1 GW for Anthropic, while the more than 20 gigawatts is the Platform’s design target through 2028 and is not funded by that tranche.¹⁷¹

Third-party estimates of the custom-ASIC share of the AI processor market depend on what is being counted. Bloomberg Intelligence, in a January 14, 2026 note, splits the custom-ASIC market between Broadcom at 60% to 80% and Marvell at 20% to 25%. Morgan Stanley analyst Joseph Moore, in a March 2, 2026 note, puts custom ASICs at above 10% of AI processor revenue, with NVIDIA at around 85% and AMD below 5%. TrendForce puts ASIC-based AI servers at about 27% of 2026 AI server unit shipments, revised down from 27.8% on April 15, 2026, citing chip validation and tuning delays at Meta and AWS.

31. Marvell

Marvell designs custom AI silicon, which it calls XPUs, for hyperscalers, alongside the interconnect, optics and networking silicon that connects those chips. Bloomberg Intelligence estimated in January 2026 that Broadcom holds 60-80% of the custom AI ASIC market and Marvell 20-25%.

Marvell reported record net revenue of $8.195 billion for fiscal 2026, which ended January 31, 2026, up 42% year over year, with GAAP EPS of $3.07 and non-GAAP EPS of $2.84. Fourth-quarter revenue was $2.219 billion.¹⁷² For the following quarter, reported on May 27, 2026, Marvell posted record revenue of $2.418 billion, up 28% year over year and 9% sequentially, record data center revenue of $1.83 billion, up 27% year over year, and record operating cash flow of $639 million.

On that call, CEO Matt Murphy raised the company’s outlook. These are management guidance figures rather than reported results. Marvell guided to approximately $11.5 billion for fiscal 2027, around 40% growth and more than $500 million above the prior view, and approximately $16.5 billion for fiscal 2028, about $1.5 billion above the prior view of roughly $15 billion. Marvell also reaffirmed a target of more than $10 billion in custom business revenue in fiscal 2029.

On March 31, 2026, NVIDIA and Marvell announced a strategic partnership that connects Marvell to the NVIDIA AI factory and AI-RAN ecosystem through NVLink Fusion, together with collaboration on silicon photonics. The announcement states that “NVIDIA has invested $2 billion in Marvell.” Under the partnership, Marvell provides custom XPUs and NVLink Fusion-compatible scale-up networking, while NVIDIA provides the Vera CPU, ConnectX NICs, BlueField DPUs, NVLink interconnect, Spectrum-X switches and rack-scale compute.¹⁷³

Marvell closed three acquisitions between February and April 2026, all on the optical and interconnect layer:

• Celestial AI, announced December 2, 2025 at $3.25 billion payable at closing, $1.0 billion in cash plus approximately 27.2 million Marvell shares. The acquisition completed on February 2, 2026 and reduced Marvell’s cash by $1 billion. Marvell expects initial Celestial AI revenue in the second half of fiscal 2028, a $500 million annualized run rate in the fourth quarter of fiscal 2028 and $1 billion annualized by the fourth quarter of fiscal 2029.¹⁷⁴
• XConn Technologies, announced January 6, 2026 at approximately $540 million, roughly 60% cash and 40% stock. The acquisition completed on February 10, 2026 and reduced Marvell’s cash by $325 million.
• Polariton Technologies, a Swiss ETH Zurich spinoff working on plasmonics-based silicon photonics, acquired on April 22, 2026 on undisclosed terms.

Marvell’s position at Amazon is disputed. Analysts and supply-chain reporting say Marvell designed Trainium2 but lost the Trainium3 design to Alchip. SemiAnalysis wrote in December 2025 that “Marvell ends up being the big loser from this. While they designed Trainium2, they lost the design bakeoff with Alchip for this generation,” attributing the loss to Trainium2 execution problems including timeline slippage and RDL interposer issues that Alchip had to fix, and describing Trainium3 as an Annapurna front-end design with Alchip handling back-end physical and package design.¹⁷⁵ On December 8, 2025, Benchmark downgraded Marvell to Hold, with analyst commentary that Marvell had lost both Trainium 3 and Trainium 4 to Alchip, and the stock fell about 7%. Two limits apply to that reporting. The sourced technical account covers Trainium3 only, and the Trainium 4 loss is the sell-side analyst’s assertion. Marvell, Amazon’s Annapurna Labs and Alchip have confirmed none of it, and other analysts including JPMorgan pushed back.

The size of the account under discussion explains the market reaction. AWS Trainium3 reached general availability on December 2, 2025 at 2.52 PFLOPS FP8, 144 GB of HBM3e and 4.9 TB/s per chip, and Amazon disclosed on its first-quarter 2026 earnings call on April 29, 2026 that it holds over $225 billion in revenue commitments for Trainium and that its chips business is at an annual revenue run rate above $20 billion.

Which CPUs run alongside AI accelerators?

Accelerators do not run on their own. Every AI rack pairs them with host CPUs, and during 2026 that pairing became a designed choice rather than a default one. NVIDIA’s Vera CPU carries 88 custom Olympus cores, which are Arm compatible, with 176 threads via Spatial Multithreading and 1.5 TB of LPDDR5X. Google’s Axion is the host CPU for both eighth-generation TPUs, the first time Google has replaced x86 in that role, and its N4A VMs became generally available in late January 2026. AMD’s Helios racks pair Instinct GPUs with 6th Gen EPYC “Venice” x86 CPUs. Qualcomm announced the Dragonfly C1000 data center CPU on June 24, 2026 with Meta as a customer, with commercial availability expected in 2028.

Arm

Arm announced on March 24, 2026 that it is “extending into production silicon products for the first time in the company’s history,” starting with the Arm AGI CPU, a data center CPU for agentic AI infrastructure. Meta “serves as the lead partner and co-developer,” and the two companies say they are committed to collaborating across multiple generations of the AGI CPU roadmap.^{176 177}

Arm states the CPU carries up to 136 Arm Neoverse V3 cores, 6 GB/s of memory bandwidth per core at sub-100ns latency and a 300 W TDP, scaling to 8,160 cores per air-cooled rack and more than 45,000 cores per liquid-cooled rack. Arm claims more than 2x performance per rack versus x86 and up to $10 billion in capex savings per gigawatt. These are Arm’s own figures and have not been independently measured.¹⁷⁸

ASRock Rack, Lenovo, Quanta and Supermicro are the announced OEM/ODM partners, with early systems available now and broader availability expected in the second half of 2026. Alongside Meta, Arm names Cerebras, Cloudflare, F5, OpenAI, Positron, Rebellions, SAP and SK Telecom as committed customers, four of which appear as vendors elsewhere in this article.¹⁷⁹

Arm continues to license IP and Compute Subsystems (CSS). The AGI CPU adds a silicon product line rather than replacing the licensing business, which means several companies covered here are now both Arm licensees and Arm customers. Meta is the clearest case of the same blurring on the buyer side. It designs its own MTIA silicon, is Arm’s AGI CPU lead partner and co-developer, is an announced Qualcomm data center CPU customer, and buys from AMD and Broadcom.

Foundry partners and TSMC’s role

As the world’s leading pure-play foundry, TSMC manufactures semiconductors based on customer designs rather than creating its own chips, distinguishing it from companies like NVIDIA and AMD. While Samsung Foundry and Intel Foundry Services compete in this space, TSMC maintains a technological edge.

N2, TSMC’s 2nm node, entered volume production in the fourth quarter of 2025 and accounted for 3% of wafer revenue in the second quarter of 2026, against 33% for 5nm, 30% for 3nm and 11% for 7nm.¹⁸⁰ TSMC’s roadmap places A16 in 2027, A14 in the second half of 2028, and A13 and A12 in 2029, as presented at its North America Technology Symposium on April 22, 2026.¹⁸¹ TSMC fabricates AI accelerators for the following designers:

Alibaba’s Hanguang 800 inference chip, produced at TSMC in 2019, is not included. US export controls now bar TSMC from producing advanced AI chips for Chinese designers, and foundries must review any shipment at 7nm and below to Chinese designers.¹⁸² Alibaba’s 2026 AI silicon is fabricated domestically, with capacity at SMIC the stated constraint.¹⁸³

SemiAnalysis reported on March 12, 2026 that front-end N3 wafer capacity has displaced CoWoS packaging as the dominant bottleneck in AI chip supply, with CoWoS tight but easing. TrendForce expects the CoWoS supply-demand gap to narrow from about 20% to about 10% by the end of 2026.¹⁸⁴ Micron reached high-volume production of HBM4 36GB 12H for NVIDIA’s Vera Rubin platform on March 16, 2026, at pin speeds above 11 Gb/s.¹⁸⁵

Expansion plans

TSMC’s announced US investment totals $265 billion across 12 leading-edge fabrication and packaging facilities, after an incremental $100 billion Arizona commitment announced on July 16, 2026, up from the $165 billion announced in March 2025.¹⁸⁶

The March 2025 reports of a TSMC-led joint venture to run Intel’s foundry division did not produce a deal. TSMC denied active discussions on April 17, 2025, and on September 26, 2025 stated that it “has never entered into talks with any company on establishing a joint venture or engaging in the licensing or transfer of technology.”¹⁸⁷ Intel was recapitalized instead. SoftBank agreed to invest $2 billion in Intel in August 2025, the US government took a 9.9% stake for $8.9 billion later that month, and NVIDIA announced a $5 billion investment in September 2025.¹⁸⁸

What are the AI chip makers in China?

Due to US sanctions preventing many Chinese companies from acquiring the most advanced AI chips from AMD and NVIDIA, Chinese buyers have increased their purchases from local producers.

Other than Huawei and Alibaba covered above, these are the leading AI chip producers in China:

• Cambricon (SHA: 688256) reported FY2025 revenue of RMB 6.4972bn, up 453.21% year over year, and net profit attributable to shareholders of RMB 2.0592bn, its first full-year profit, against a RMB 452m loss in 2024. Non-GAAP net profit was RMB 1.770bn, and the annual report was filed on March 12, 2026. Q1 2026 revenue, reported April 29, 2026, reached RMB 2.885bn, up 159.56% year over year and 52.65% quarter over quarter, with net profit attributable of RMB 1.013bn (up 185.04%) and non-GAAP net profit of RMB 934m (up 238.56%). On June 30, 2026 the shares closed 7.66% higher at CNY 1,595.55, taking market capitalization past RMB 1 trillion (about RMB 1.013trn, or US$139-147bn), the first STAR Market company to reach that level.
• Baidu Kunlunxin unveiled two chips at Baidu World in Beijing on November 13, 2025. The Kunlun M100 is optimized for large-scale inference and was slated for launch in early 2026, and the Kunlun M300, built for training and inference of ultra-large multimodal models, is targeted for 2027. Baidu also presented two supernode systems, Tianchi 256 (available in H1 2026) and Tianchi 512 (available in H2 2026), both built from Baidu’s existing Kunlun P800 chips rather than from the M100 or M300. Baidu flagged thousand-card-scale supernodes for 2028.
• Biren, founded by NVIDIA alumni, listed on the Hong Kong Stock Exchange main board on January 2, 2026, following a December 22, 2025 prospectus. It sold 284.8 million H shares at HK$19.60, the top of the marketed range, raising HK$5.58bn (about US$717m). The stock opened at HK$35.70 and closed at HK$34.46, up about 76%. It is the first mainland-Chinese GPU company to list in Hong Kong.
• Moore Threads (SHA: 688795) listed on the STAR Market on December 5, 2025 at an IPO price of CNY 114.28, raising about RMB 8bn gross and RMB 7.58bn net, roughly US$1.1-1.13bn. The stock opened up about 469% at CNY 650 and closed its first day up about 425% at CNY 600.5. FY2025 revenue was RMB 1.505bn, up 243.37% year over year, with gross profit of RMB 987m, per the annual report filed April 27, 2026. The company remains unprofitable and has been on the US Entity List since 2023.
• MetaX listed on the STAR Market on December 17, 2025 at CNY 104.66 a share, raising about RMB 4.2bn (US$586-596m) and rising roughly 693-755% on debut. It has since confidentially filed for a secondary Hong Kong listing.
• Iluvatar CoreX (HKEX: 9903) listed on the Hong Kong main board on January 8, 2026 at HK$144.60 per H share, raising about HK$3.68bn through 25.43 million H shares.
• Enflame has not listed. Its STAR Market application was approved by the listing review committee on June 15, 2026 and its IPO registration was approved by the CSRC on July 9-10, 2026, clearing it to raise RMB 6bn (about US$883m), but the shares have not begun trading as of July 2026.¹⁸⁹ Disclosed financials show 2025 revenue of RMB 990m against a net loss of RMB 1.2bn, with accumulated losses of RMB 4.4bn. Tencent holds 20.3% of the company and accounted for 83.8% of its 2025 revenue.
• SMIC, the foundry most of these designers depend on, had N+3 (5nm-class) in volume production as of June 2026, 7nm capacity of roughly 45,000 to 60,000 wafers per month, and $9.3bn of 2025 revenue.

Domestic share of China’s AI accelerator market

An IDC report seen by Reuters puts total 2025 shipments by NVIDIA, AMD and Chinese vendors at approximately 4 million AI accelerator cards in China.¹⁹⁰ The metric is unit shipments of accelerator cards, not servers and not revenue.

Within the Chinese 41%, Huawei led with about 812,000 chips, roughly half of all domestically branded shipments. Alibaba’s T-Head was second with about 265,000, and Baidu Kunlunxin and Cambricon tied third at about 116,000 each. Hygon, MetaX and Iluvatar CoreX accounted for 5%, 4% and 3% of Chinese-vendor shipments respectively.¹⁹¹

The 41% is a share of unit shipments, not of revenue.

FAQs

Further reading

For hands-on performance comparisons of the chips covered in this article, see our benchmarks:

• Multi-GPU benchmark: How NVIDIA’s B200, H200, H100, and AMD’s MI300X scale across 1, 2, 4, and 8-GPU configurations for LLM inference, with throughput, latency, and cost-per-token analysis.
• GPU concurrency benchmark: How NVIDIA’s B200, H200, H100, and AMD’s MI300X handle 1 to 512 concurrent requests, including system throughput, per-query speed, end-to-end latency, and tokens-per-dollar at each concurrency level.
• CUDA vs ROCm: How AMD’s ROCm compares with NVIDIA’s CUDA on framework support and portability, the software question behind AMD’s position in this article.
• Edge AI chip makers: The vendors in the edge table above covered one by one, with the use cases each part targets.
• Cloud GPU providers: 60+ providers that rent the accelerators covered here by the hour, for teams that do not buy hardware.

Cite this research

Pick the format that matches where you're publishing. Pasting the link version into your CMS preserves the backlink.

@misc{dilmegani2026,
  author = {Dilmegani, Cem and Sarı, Ekrem},
  title  = {{Top 30+ AI Chip Makers: NVIDIA & Its Competitors}},
  year   = {2026},
  month  = jul,
  howpublished    = {\url{https://aimultiple.com/ai-chip-makers}},
  note   = {AIMultiple. Retrieved July 27, 2026}
}

Reference Links

1.

https://investors.broadcom.com/news-releases/news-release-details/openai-and-broadcom-unveil-llm-optimized-intelligence-processor

2.

https://newsroom.arm.com/news/arm-agi-cpu-launch

3.

https://www.crn.com/news/components-peripherals/2026/how-intel-s-ai-chip-strategy-is-coming-into-shape-after-years-of-struggles

4.

https://newsroom.intel.com/artificial-intelligence/intel-to-expand-ai-accelerator-portfolio-with-new-gpu

5.

https://investors.broadcom.com/news-releases/news-release-details/openai-and-broadcom-unveil-llm-optimized-intelligence-processor

6.

https://www.qualcomm.com/news/releases/2025/10/qualcomm-unveils-ai200-and-ai250-redefining-rack-scale-data-cent

7.

ASIC Set to Outpace GPU? NVIDIA’s Scale-Up and Beyond | TrendForce

TrendForce

8.

https://www.trendforce.com/presscenter/news/20260120-12887.html

9.

Personal AI Supercomputer Powered by Blackwell | NVIDIA DGX Spark

10.

NVIDIA and Microsoft Reinvent Windows PCs for the Age of Personal AI | NVIDIA Newsroom

11.

Introducing a powerful new chapter for Windows PCs, accelerated by NVIDIA RTX Spark | Windows Experience Blog

12.

NVIDIA Corporation - Financial Reports

13.

DGX GB200: AI Infrastructure for State-of-the-Art AI Models | NVIDIA

14.

https://uk.investing.com/news/stock-market-news/nvidia-reclaims-the-top-chip-stock-spot-at-morgan-stanley-4533620

15.

https://www.bloomberg.com/company/press/ai-accelerator-market-looks-set-to-exceed-600-billion-by-2033-driven-by-hyperscale-spending-and-asic-adoption-according-to-bloomberg-intelligence/

16.

Nvidia Pushes Further Into Cloud With GPU Marketplace - WSJ

The Wall Street Journal

17.

NVIDIA Dynamo, A Low-Latency Distributed Inference Framework for Scaling Reasoning AI Models | NVIDIA Technical Blog

NVIDIA Developer

18.

NVIDIA Dynamo, A Low-Latency Distributed Inference Framework for Scaling Reasoning AI Models | NVIDIA Technical Blog

NVIDIA Developer

19.

NVIDIA RTX PRO Servers Speed Trillion-Dollar Enterprise IT Industry Transition to AI Factories | NVIDIA Newsroom

20.

NVIDIA Vera Rubin Opens Agentic AI Frontier | NVIDIA Newsroom

21.

https://nvidianews.nvidia.com/news/vera-rubin-full-production-agentic-ai-factory

22.

NVIDIA Computex 2026 News Bytes: Vera Rubin Now In Production, DGX Station Gets Windows - ServeTheHome

ServeTheHome

23.

https://www.nvidia.com/en-us/data-center/vera-rubin-nvl72/

24.

https://groq.com/newsroom/groq-and-nvidia-enter-non-exclusive-inference-technology-licensing-agreement-to-accelerate-ai-inference-at-global-scale

25.

Nvidia buying AI chip startup Groq for about $20 billion, biggest deal

CNBC

26.

https://www.nvidia.com/en-us/data-center/lpx/

27.

https://semiconductor.samsung.com/news-events/tech-blog/architecting-the-ai-era-samsung-electronics-and-nvidia-define-the-future-at-gtc-2026/

28.

https://www.nvidia.com/en-us/data-center/lpx/

29.

Jevons paradox - Wikipedia

Contributors to Wikimedia projects

30.

https://www.bis.gov/press-release/department-commerce-revises-license-review-policy-semiconductors-exported-china

31.

https://www.bis.gov/press-release/department-commerce-revises-license-review-policy-semiconductors-exported-china

32.

Trump imposes 25% tariff on Nvidia AI chips and others, citing national security | Nvidia | The Guardian

The Guardian

33.

Nvidia CEO to defend AI dominance as competition intensifies | Reuters

Reuters

34.

Announcing AI2 OLMo, an Open Language Model Made by Scientists, for Scientists | by Ai2 | Ai2 Blog | Medium

Ai2 Blog

35.

Training LLMs at Scale with AMD MI250 GPUs | Databricks Blog

36.

Training 221B Parameter Korean LLM on 1,200 AMD MI250 GPU Cluster – Moreh

37.

https://www.amd.com/en/blogs/2026/amd-acquires-mext-for-memory-optimization.html

38.

Exclusive: AMD Acquires Team Behind AI Chip Startup Untether AI

39.

https://ir.amd.com/news-events/press-releases/detail/1255/amd-unveils-vision-for-an-open-ai-ecosystem-detailing-new-silicon-software-and-systems-at-advancing-ai-2025

40.

https://www.hpcwire.com/off-the-wire/amd-launches-instinct-mi350p-pcie-cards-for-enterprise-ai-inference/

41.

https://ir.amd.com/news-events/press-releases/detail/1260/amd-and-openai-announce-strategic-partnership-to-deploy-6-gigawatts-of-amd-gpus

42.

https://www.oracle.com/news/announcement/ai-world-oracle-and-amd-expand-partnership-to-help-customers-achieve-next-generation-ai-scale-2025-10-14/

43.

Meta and AMD Partner for Longterm AI Infrastructure Agreement

Meta

44.

AMD, Meta Announce Expanded Strategic Partnership to Deploy 6 Gigawatts of AMD GPUs - AMD Newsroom

Advanced Micro Devices, Inc.

45.

https://ir.amd.com/news-events/press-releases/detail/1292/amd-and-anthropic-announce-strategic-partnership-to-deploy-up-to-2-gigawatts-of-amd-instinct-mi450-series-gpus

46.

AMD + : Large Language Models Out-of-the-Box Acceleration with AMD GPU

Hugging Face

47.

Achieving Top Inference Performance with the NVIDIA H100 Tensor Core GPU and NVIDIA TensorRT-LLM | NVIDIA Technical Blog

NVIDIA Developer

48.

https://ir.amd.com/news-events/press-releases/detail/1294/aai-2026-amd-delivers-full-stack-compute-for-the-agentic-ai-era

49.

https://www.amd.com/en/products/accelerators/instinct/mi400/mi455x.html

50.

https://ir.amd.com/news-events/press-releases/detail/1294/aai-2026-amd-delivers-full-stack-compute-for-the-agentic-ai-era

51.

AMD and its Partners Share Their Vision for 'AI Everywhere, for Everyone' - AMD Newsroom

Advanced Micro Devices, Inc.

52.

https://ir.amd.com/news-events/press-releases/detail/1294/aai-2026-amd-delivers-full-stack-compute-for-the-agentic-ai-era

53.

https://www.amd.com/en/products/accelerators/instinct/mi400/mi430x.html

54.

https://www.amd.com/en/products/accelerators/instinct/mi400/mi430x.html

55.

MI300X vs H100 vs H200 Benchmark Part 1: Training - CUDA Moat Still Alive

SemiAnalysis

56.

https://newsletter.semianalysis.com/p/can-amd-break-the-cuda-moat-amd-advancing

57.

https://www.phoronix.com/news/AMD-ROCm-Six-Weeks

58.

Exclusive | AMD Invests in Drug-Discovery Company Absci in Push to Sell AI Chips - WSJ

The Wall Street Journal

59.

https://ir.amd.com/news-events/press-releases/detail/1260/amd-and-openai-announce-strategic-partnership-to-deploy-6-gigawatts-of-amd-gpus

60.

https://ir.amd.com/news-events/press-releases/detail/1279/amd-and-meta-announce-expanded-strategic-partnership-to-deploy-6-gigawatts-of-amd-gpus

61.

Intel Breaks Down Proprietary Walls to Bring Choice to Enterprise GenAI Market - Intel Newsroom

Intel Corporation

62.

https://www.crn.com/news/components-peripherals/2026/how-intel-s-ai-chip-strategy-is-coming-into-shape-after-years-of-struggles

63.

https://www.crn.com/news/components-peripherals/2026/how-intel-s-ai-chip-strategy-is-coming-into-shape-after-years-of-struggles

64.

Intel Appoints Lip-Bu Tan as Chief Executive Officer - Intel Newsroom

Intel Corporation

65.

https://www.crn.com/news/components-peripherals/2026/how-intel-s-ai-chip-strategy-is-coming-into-shape-after-years-of-struggles

66.

Intel’s ‘Make-or-Break’ Jaguar Shores Rack-Scale AI Lineup Expected to Be Finalized in H1 2026, With Talks Underway With Alchip

67.

Intel Computex 2026 Keynote Live Coverage - ServeTheHome

ServeTheHome

68.

https://newsroom.intel.com/artificial-intelligence/intel-to-expand-ai-accelerator-portfolio-with-new-gpu

69.

https://www.crn.com/news/components-peripherals/2026/how-intel-s-ai-chip-strategy-is-coming-into-shape-after-years-of-struggles

70.

https://www.qualcomm.com/news/releases/2025/10/qualcomm-unveils-ai200-and-ai250-redefining-rack-scale-data-cent

71.

https://www.qualcomm.com/news/releases/2026/06/qualcomm-unveils-comprehensive-data-center-roadmap-for-the-agent

72.

https://investor.qualcomm.com/news-events/press-releases/news-details/2026/Qualcomm-Accelerates-Diversification-with-Comprehensive-Strategy-for-Data-Center-and-Sees-Multiple-Inflection-Points-Over-the-Next-3-to-5-Years/default.aspx

73.

https://investor.qualcomm.com/news-events/press-releases/news-details/2026/Qualcomm-to-Acquire-Modular/default.aspx

74.

https://www.anthropic.com/news/anthropic-amazon-compute

75.

https://aws.amazon.com/about-aws/whats-new/2025/12/amazon-ec2-trn3-ultraservers/

76.

https://www.aboutamazon.com/news/company-news/amazon-ceo-andy-jassy-amazon-chips-business-q1-2026-earnings

77.

Google supercharges machine learning tasks with TPU custom chip | Google Cloud Blog

Google Cloud

78.

Introducing Trillium, sixth-generation TPUs | Google Cloud Blog

Google Cloud

79.

Ironwood: The first Google TPU for the age of inference

Google

80.

https://docs.cloud.google.com/tpu/docs/release-notes

81.

https://docs.cloud.google.com/tpu/docs/tpu7x

82.

https://blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/eighth-generation-tpu-agentic-era/

83.

https://www.alibabagroup.com/document-1994119844504535040

84.

IBM’s new AIU artificial intelligence chip - IBM Research

IBM

85.

Can an in-house AI chip save IBM's cost of operating WatsonX?

TechHQ

86.

Telum Processor: IBM’s newest chip - IBM Research

IBM

87.

‘Mind-blowing’ IBM chip speeds up AI

Nature Publishing Group UK

88.

https://newsroom.ibm.com/2025-10-07-ibm-introduces-the-spyre-accelerator-for-commercial-availability

89.

Tech war: China’s chip firms embrace DeepSeek in AI self-sufficiency drive | South China Morning Post

South China Morning Post

90.

https://www.tomshardware.com/pc-components/gpus/huawei-unveils-new-atlas-350-ai-accelerator-with-1-56-pflops-of-fp4-compute-and-up-to-112gb-of-hbm-claims-2-8x-more-performance-than-nvidias-h20

91.

https://www.huawei.com/en/news/2025/9/hc-xu-keynote-speech

92.

https://groq.com/newsroom/groq-and-nvidia-enter-non-exclusive-inference-technology-licensing-agreement-to-accelerate-ai-inference-at-global-scale

93.

Nvidia buying AI chip startup Groq for about $20 billion, biggest deal

CNBC

94.

https://www.nvidia.com/en-us/data-center/lpx/

95.

https://groq.com/newsroom/groq-raises-usd650m-to-scale-its-ai-inference-cloud-business

96.

https://groq.com/newsroom/groq-raises-640m-to-meet-soaring-demand-for-fast-ai-inference

97.

https://groq.com/newsroom/groq-raises-750-million-as-inference-demand-surges

98.

Groq Sets New Large Language Model Performance Record of 300 Tokens per Second per User on Meta AI Foundational LLM, Llama-2 70B

Cision PR Newswire

99.

AI chip startup Groq secures $1.5 billion commitment from Saudi Arabia | Reuters

Reuters

100.

Real-time AI Inference Demand Accelerates on GroqCloud | Groq is fast, low cost inference.

101.

https://groq.com/newsroom/groq-raises-usd650m-to-scale-its-ai-inference-cloud-business

102.

Groq Acquires Dataflow Systems Pioneer Maxeler Technologies

Cision PR Newswire

103.

SambaNova Unveils Fastest Chip for Agentic AI, Collaborates with Intel, and Raises $350M+

SambaNova

104.

https://www.generalatlantic.com/media-article/sambanova-completes-first-close-of-1-billion-financing-at-11-billion-valuation/

105.

Introducing the SN50 RDU: Purpose-Built for Agentic Inference

SambaNova

106.

SambaCloud | Full-Stack AI Platform for Large Open-Source Models

107.

Cerebras Systems - Wikipedia

Contributors to Wikimedia projects

108.

Cerebras Systems Unveils World’s Fastest AI Chip with Whopping 4 Trillion Transistors - Cerebras

109.

https://www.cerebras.ai/press-release/cerebras-systems-announces-closing-of-initial-public-offering

110.

https://investors.cerebras.ai/news-releases/news-release-details/cerebras-systems-announces-strong-first-quarter-2026-results

111.

https://investors.cerebras.ai/news-releases/news-release-details/cerebras-systems-announces-strong-first-quarter-2026-results

112.

d-Matrix Raises $275 Million to Power the Age of AI Inference - d-Matrix

d-Matrix

113.

Bullhound Capital leads $275m investment into AI inference leader d-Matrix - Bullhound Capital

Bullhound Capital

114.

d-Matrix Corsair AI Inference Platform Enters Full Production to Meet Customer Demand - d-Matrix

d-Matrix

115.

Korean AI chipmaker Rebellions Closes $124M Series B Fundraise - Rebellions

Rebellions

116.

Rebellions and SAPEON Korea Sign Definitive Merger Agreement - Rebellions

Rebellions

117.

Samsung backs AI chip startup Rebellions ahead of IPO

CNBC

118.

https://rebellions.ai/newsroom/rebellions-closes-400-million-pre-ipo-and-launches-rebelrack-and-rebelpod-to-accelerate-global-expansion/

119.

https://www.cnbc.com/2026/07/08/rebellions-ipo-south-korea-ai-chips.html

120.

https://tenstorrent.com/hardware/cards

121.

https://github.com/tenstorrent/tt-zephyr-platforms/blob/main/doc/release/release-notes-19.5.md#p150-tensix-core-count

122.

Jeff Bezos Is Betting on AI Chip Startup Tenstorrent to Take on Nvidia (NVDA) - Bloomberg

Bloomberg

123.

https://www.positron.ai/atlas

124.

https://www.positron.ai/titan

125.

Positron | About

126.

https://www.globenewswire.com/news-release/2026/06/30/3319922/0/en/etched-emerges-from-stealth-with-working-chip-800m-raised-and-over-1b-in-customer-contracts.html

127.

https://www.reuters.com/technology/ai-chip-startup-etched-raises-300-million-103-billion-valuation-2026-07-23/

128.

https://www.globenewswire.com/news-release/2026/06/30/3319922/0/en/etched-emerges-from-stealth-with-working-chip-800m-raised-and-over-1b-in-customer-contracts.html

129.

The path to ubiquitous AI | Taalas

130.

Products | Taalas

131.

https://extropic.ai/hardware

132.

A startup working on 'reversible computing' chip for AI says initial tests show a 50% energy savings | Fortune

Fortune

133.

https://www.fractile.ai/news/startup-with-radical-concept-for-ai-chips-emerges-from-stealth

134.

https://www.fractile.ai/news/how-inference-is-driving-competition-to-nvidias-ai-chip-dominance

135.

https://www.fractile.ai/news/fractile-raises-220m-to-build-the-next-generation-of-inference-hardware

136.

Secret Apple 'Project ACDC' to pioneer AI chips for data centers

AppleInsider

137.

https://www.reuters.com/business/apple-chasing-ai-chip-company-deals-information-reports-2026-07-15/

138.

https://www.techzine.eu/news/infrastructure/142924/apple-considers-acquiring-server-chip-companies/

139.

Core AI - Apple Developer

140.

GitHub - apple/coreai-models: Model export recipes, Python primitives, and Swift runtime utilities for on-device AI · GitHub

141.

Our next generation Meta Training and Inference Accelerator

142.

https://ai.meta.com/blog/meta-mtia-scale-ai-chips-for-billions/

143.

https://investors.broadcom.com/news-releases/news-release-details/broadcom-announces-extended-partnership-meta-deploy-technology

144.

https://www.fool.com/earnings/call-transcripts/2026/06/03/broadcom-avgo-q2-2026-earnings-transcript/

145.

Maia 200: The AI accelerator built for inference - The Official Microsoft Blog

146.

Maia 200: The AI accelerator built for inference - The Official Microsoft Blog

147.

OpenAI and Broadcom announce strategic collaboration to deploy 10 gigawatts of OpenAI-designed AI accelerators | OpenAI

148.

OpenAI and Broadcom unveil LLM-optimized inference chip | OpenAI

149.

[News] OpenAI Reportedly to Deploy Custom AI Chip on TSMC N3 by End-2026, Second-Gen Planned for A16

TrendForce

150.

[News] Samsung Reportedly Allocates 50%+ of Pyeongtaek Foundry Capacity to HBM4 Base Die; Said to Win OpenAI Deal

TrendForce

151.

OpenAI and Broadcom announce strategic collaboration to deploy 10 gigawatts of OpenAI-designed AI accelerators | OpenAI

152.

https://investors.broadcom.com/news-releases/news-release-details/openai-and-broadcom-unveil-llm-optimized-intelligence-processor

153.

https://www.fool.com/earnings/call-transcripts/2026/06/03/broadcom-avgo-q2-2026-earnings-transcript/

154.

GRAPHCORE LIMITED filing history - Find and update company information - GOV.UK

155.

Graphcore joins SoftBank Group to build next generation of AI compute

Graphcore

156.

https://www.cnbc.com/2026/05/12/softbank-graphcore-ai-chip-investment.html

157.

Mythic company information, funding & investors | Dealroom.co

158.

AI chip startup Mythic rises from the ashes with $13M, new CEO | TechCrunch

TechCrunch

159.

https://mythic.ai/whats-new/mythic-to-challenge-ais-gpu-pantheon-with-100x-energy-advantage-and-oversubscribed-125m-raise/

160.

Speedata, a chip startup competing with Nvidia, raises a $44M Series B | TechCrunch

TechCrunch

161.

https://axelera.ai/news/axelera-ai-secures-more-than-250-million-funding-on-global-commercial-growth

162.

https://www.reuters.com/business/european-ai-chip-startup-axelera-raises-additional-250-million-2026-02-24/

163.

https://investors.broadcom.com/news-releases/news-release-details/broadcom-inc-announces-second-quarter-fiscal-year-2026-financial

164.

https://www.fool.com/earnings/call-transcripts/2026/06/03/broadcom-avgo-q2-2026-earnings-transcript/

165.

https://www.fool.com/earnings/call-transcripts/2026/06/03/broadcom-avgo-q2-2026-earnings-transcript/

166.

https://www.sec.gov/Archives/edgar/data/1730168/000119312526144028/d87999d8k.htm

167.

https://www.sec.gov/Archives/edgar/data/1730168/000119312526144028/d87999d8k.htm

168.

https://www.fool.com/earnings/call-transcripts/2026/06/03/broadcom-avgo-q2-2026-earnings-transcript/

169.

https://investors.broadcom.com/news-releases/news-release-details/broadcom-announces-extended-partnership-meta-deploy-technology

170.

https://www.globenewswire.com/news-release/2026/02/26/3245680/19933/en/broadcom-ships-3-5d-face-to-face-compute-soc-powering-ai-revolution.html

171.

https://investors.broadcom.com/news-releases/news-release-details/broadcom-apollo-and-blackstone-establish-landmark-strategic

172.

https://investor.marvell.com/news-events/press-releases/detail/1011/marvell-technology-inc-reports-fourth-quarter-and-fiscal-year-2026-financial-results

173.

https://investor.marvell.com/news-events/press-releases/detail/1019/nvidia-ai-ecosystem-expands-as-marvell-joins-forces-through-nvlink-fusion

174.

https://investor.marvell.com/news-events/press-releases/detail/1005/marvell-completes-acquisition-of-celestial-ai

175.

https://newsletter.semianalysis.com/p/aws-trainium3-deep-dive-a-potential

176.

https://newsroom.arm.com/news/arm-agi-cpu-launch

177.

https://www.cnbc.com/2026/03/24/arm-launches-its-own-cpu-with-meta-as-first-customer.html

178.

https://newsroom.arm.com/news/arm-agi-cpu-launch

179.

https://newsroom.arm.com/news/arm-agi-cpu-launch

180.

https://pr.tsmc.com/english/news/3326

181.

https://pr.tsmc.com/english/news/3326

182.

https://www.alibabagroup.com/document-1994119844504535040

183.

https://www.alibabagroup.com/document-1994119844504535040

184.

https://newsletter.semianalysis.com/p/the-great-ai-silicon-shortage

185.

https://investors.micron.com/news-releases/news-release-details/micron-high-volume-production-hbm4-designed-nvidia-vera-rubin

186.

https://www.commerce.gov/news/press-releases/2026/07/trump-administration-secures-additional-100-billion-us-semiconductor

187.

https://www.taipeitimes.com/News/biz/archives/2025/09/27/2003844488

188.

https://www.intc.com/news-events/press-releases/detail/1748/intel-and-trump-administration-reach-historic-agreement-to

189.

https://www.caixinglobal.com/2026-07-10/ai-chipmaker-enflame-wins-approval-for-883-million-star-market-ipo-102462536.html

190.

https://telecom.economictimes.indiatimes.com/news/devices/chinese-chipmakers-surge-to-41-of-ai-accelerator-market-as-nvidias-dominance-declines/129966271

191.

https://telecom.economictimes.indiatimes.com/news/devices/chinese-chipmakers-surge-to-41-of-ai-accelerator-market-as-nvidias-dominance-declines/129966271

Cem Dilmegani

Principal Analyst

Follow On

Cem has been the principal analyst at AIMultiple since 2017. AIMultiple informs hundreds of thousands of businesses (as per similarWeb) including 55% of Fortune 500 every month.

Cem's work has been cited by leading global publications including Business Insider, Forbes, Washington Post, global firms like Deloitte, HPE and NGOs like World Economic Forum and supranational organizations like European Commission. You can see more reputable companies and resources that referenced AIMultiple.

Throughout his career, Cem served as a tech consultant, tech buyer and tech entrepreneur. He advised enterprises on their technology decisions at McKinsey & Company and Altman Solon for more than a decade. He also published a McKinsey report on digitalization.

He led technology strategy and procurement of a telco while reporting to the CEO. He has also led commercial growth of deep tech company Hypatos that reached a 7 digit annual recurring revenue and a 9 digit valuation from 0 within 2 years. Cem's work in Hypatos was covered by leading technology publications like TechCrunch and Business Insider.

Cem regularly speaks at international technology conferences. He graduated from Bogazici University as a computer engineer and holds an MBA from Columbia Business School.

View Full Profile

Technically reviewed by

Ekrem Sarı

AI Researcher

Follow On

Ekrem is an AI Researcher and Data Analyst at AIMultiple. He designs and runs hands-on benchmarks for AI and LLM systems.

View Full Profile

Comments 2

Share Your Thoughts

Your email address will not be published. All fields are required. Comments are left in their original language.

Name

Email Address

Comment

0/450

Post Comment

Dave

Aug 29, 2022 at 05:49

You forgot to include Tesla with their DOJO supercomputer. From the ground-up, the supercomputer was specifically designed for machine learning and image recognition - which means that every component was designed for it including, but not limited to, PCI board design, CPU, RAM, cooling, power, scalable hardware design and software. If I'm not mistaken, the AI is also the second most widely tested and used in the "wild", just below that of Google due to Google using it in their Search.

Reply

Cem Dilmegani

Sep 06, 2022 at 13:52

Thank you for your feedback, Dave! Here we are only covering companies that sell the chips that they produce. Therefore, companies like Tesla that build supercomputers for their own use or companies that embed chips in their products are out of our scope.

Reply

thayyil

Mar 19, 2022 at 11:48

surprised that brainchip (akida) missing in this report. any reasons?

Reply

Cem Dilmegani

Nov 18, 2022 at 07:36

All included companies here raised $100+M. Last time we collected the data, that wasn't the case for akida. Why don't you reach out to us at info@aimultiple.com and let's discuss why it should be included. Thank you!

Reply