Nvidia Competitors in 2026: Who Is Chasing the AI Chip Lead

Last Updated: November 6, 2026

Summary: Nvidia's main competitors in 2026 are AMD with its Instinct GPUs, Broadcom and Marvell with custom AI chips built for cloud giants, the in-house chips of Google, Amazon, Microsoft and Meta, Intel, inference specialists such as Cerebras, and Huawei in China. None is close to catching Nvidia yet. Nvidia reported $89.0 billion in data center revenue for the quarter ended July 26, 2026, and Morgan Stanley estimated in March 2026 that Nvidia held roughly 85% of AI processor revenue.

The competition is real, though, and it is growing faster than at any point in the AI boom. AMD's data center sales nearly doubled in its latest quarter, Broadcom's AI chip revenue tripled, and every major cloud company now designs its own silicon.

This guide covers who sells what, the latest numbers for each rival, and why Nvidia's lead stays wide. Figures come from company filings, official announcements or named analysts, checked in November 2026. Nothing here is investment advice.

Nvidia's biggest competitors fall into five groups: rival GPU makers, custom chip designers, cloud companies building their own chips, inference startups, and Chinese chipmakers. Nvidia names most of them itself. Its fiscal 2026 annual report lists AMD, Huawei and Intel as GPU and accelerator competitors, and Alibaba, Alphabet, Amazon, Baidu, Huawei and Microsoft as cloud companies with their own in-house chips.

The contest is unusual because several rivals are also Nvidia's biggest customers. Microsoft, Amazon, Google and Meta buy huge numbers of Nvidia GPUs while building chips meant to replace some of them.

Nvidia's lead is still very large. In its second quarter of fiscal 2027, which ended July 26, 2026, Nvidia reported revenue of $96.2 billion, up 106% from a year earlier, according to its earnings release. Data center revenue was $89.0 billion, up 117% year over year, and the company guided to $108.0 billion in total revenue for the following quarter.

Market share estimates vary by method:

To put the revenue gap in perspective, Nvidia's data center revenue for one quarter was about 10 times AMD's entire data center segment in AMD's latest quarter, and about 5 times Broadcom's quarterly AI chip revenue. The quarters end on different dates, but the gap is clear. For more on Nvidia's numbers, see our Nvidia statistics page.

AMD is the closest direct rival in GPUs and is growing quickly, but it remains far smaller. AMD reported third quarter 2026 revenue of $13.4 billion on November 3, 2026, with a record $8.5 billion from its data center segment, up 97% year over year, according to the company's results release. That segment includes EPYC server CPUs as well as Instinct GPUs.

AMD credited part of the growth to the launch of its Instinct MI450 Series GPUs. Its Helios rack-scale systems pair MI450 GPUs with 6th Gen EPYC CPUs, the same rack-level approach Nvidia sells with its NVL72 systems. AMD says Anthropic, Meta, Microsoft, OpenAI and Oracle are running workloads on the first Helios systems.

AMD's biggest single commitment comes from OpenAI. In October 2025, the two companies announced a deal for OpenAI to deploy 6 gigawatts of AMD GPUs over multiple generations, starting with 1 gigawatt of MI450 GPUs in the second half of 2026. AMD gave OpenAI a warrant for up to 160 million AMD shares that vests as deployments scale.

AMD is also buying into specialized inference. In August 2026, it agreed to acquire Taalas, a Toronto startup that builds inference silicon, for undisclosed terms.

Intel is a competitor on paper but a minor one in AI accelerators today. Intel still sells its Gaudi 3 accelerator and supports Gaudi 2, according to an October 2026 note from Jon Peddie Research.

Intel's next attempt is Crescent Island, an inference-only data center GPU based on its Xe3P architecture with 160GB of LPDDR5X memory. At Hot Chips 2026, Intel said customer sampling would begin that quarter, with an official launch slated for 2027, TechSpot reported.

Intel's relationship with Nvidia is now partly cooperative. In September 2025, Nvidia agreed to invest $5 billion in Intel common stock at $23.28 per share, and the companies said they would jointly develop custom data center and PC chips, ABC7 News reported.

Broadcom and Marvell compete with Nvidia indirectly by designing custom AI chips for cloud companies that would otherwise buy more Nvidia GPUs. These application-specific chips (ASICs) can be cheaper and more power-efficient at very large scale.

Broadcom is the larger of the two by far. For its fiscal third quarter ended August 2, 2026, Broadcom reported AI semiconductor revenue of $16.7 billion, up 221% year over year, in its earnings release. Broadcom expects $21.7 billion of AI revenue in the fourth quarter. Broadcom is also the supply partner behind Google's TPUs: Anthropic's April 2026 agreement for about 3.5 gigawatts of next-generation TPU capacity starting in 2027 runs through Broadcom, The Next Web reported.

Marvell is smaller but growing. Marvell reported record revenue of $2.739 billion for its fiscal second quarter ended August 1, 2026, up 37%, with data center growth accelerating to 46%, according to its results release. CEO Matt Murphy said he expects "a significant acceleration in our Custom business" beginning in the second half of fiscal 2027.

Yes. All four hyperscalers run their own AI chips alongside Nvidia GPUs, aiming for lower costs and less dependence on one supplier rather than a clean break. The scale of their spending, covered in our AI spending statistics, is why even partial substitution matters.

Google's Tensor Processing Units are the most established alternative. At Cloud Next in April 2026, Google announced its eighth generation as two chips: TPU 8t for training and TPU 8i for inference, IT Pro reported. Google claims TPU 8t is about 2.8 times more powerful than the previous Ironwood generation and offers double the performance per watt.

Amazon Web Services builds Trainium chips and Inferentia chips, now in their second generation with Inferentia2 for inference. AWS made Trainium3 UltraServers generally available after unveiling them at re:Invent 2025, according to Data Center Dynamics. Trainium3 is AWS's first 3nm AI chip, with up to 144 chips per UltraServer and up to 4.4 times the compute of the prior generation, per AWS. AWS CEO Matt Garman called Trainium "a multi-billion-dollar business," and AWS says Trainium4 will support Nvidia's NVLink Fusion interconnect.

Microsoft unveiled Maia 200, its second-generation AI accelerator, in January 2026. Built on TSMC's N3 process with 216GB of HBM3e memory, Maia 200 delivers 10.1 petaOPS at FP4 and offers 30% better performance per dollar than the latest hardware in Microsoft's fleet, according to Microsoft's Azure engineering blog.

Meta introduced four chips in its MTIA family in March 2026: MTIA 300, 400, 450 and 500, MarketScreener reported. MTIA 300 already trains ranking and recommendation models; the later chips target generative AI inference and should be in service by 2027. Meta still buys Nvidia and AMD hardware in parallel.

Inference specialists compete on speed and cost for running models, but the two best-known names took opposite paths: Cerebras went public, while much of Groq's technology went to Nvidia.

Cerebras priced its IPO at $185 per share in May 2026, raising at least $5.5 billion under the ticker CBRS, SiliconANGLE reported. Its WSE-3 is a wafer-sized chip with 44GB of on-chip SRAM. Cerebras has a $20 billion agreement to provide OpenAI with 750 megawatts of compute over three years. Customer concentration is a known risk: its prospectus showed the Mohamed bin Zayed University of Artificial Intelligence accounted for 62% of 2025 revenue.

Groq signed a non-exclusive licensing deal with Nvidia in December 2025, and Nvidia hired founder Jonathan Ross and president Sunny Madra, SiliconANGLE reported. A Groq investor told CNBC the transaction was worth $20 billion. Groq stays independent and runs its GroqCloud service.

The takeaway: Nvidia has shown it will license or absorb promising inference technology rather than let it become a threat.

Huawei is Nvidia's strongest competitor in China, helped by US export controls on Nvidia chips. In the first half of 2026, Nvidia's share of China's data center AI accelerator market fell to 34% while Huawei reached 28%, according to an IDC report cited by Reuters and TechNode.

Nvidia's CEO gave a starker view. Jensen Huang said in a May 2026 podcast interview, "In China, we have now dropped to zero," Tom's Hardware reported. The same report said Huawei projects about $12 billion in AI processor revenue for 2026, driven by its Ascend 950PR inference chip, with a training-focused 950DT scheduled for the fourth quarter.

Nvidia's filings confirm the squeeze. Its outlook for the quarter after July 2026 assumed no data center compute revenue from China.

Each main competitor, its flagship chip and its latest sourced data point:

Nvidia's lead rests on two things rivals struggle to copy quickly: its CUDA software ecosystem and the scarce manufacturing capacity needed to ship AI chips in volume.

Nvidia says more than 7.5 million developers worldwide use CUDA and its other software tools, and its computing platform supports 6,000 applications, according to its annual report. Years of AI code is written for CUDA, so switching chips often means rewriting software.

Rivals are attacking that gap. AMD's ROCm software stack plays a similar role for AMD GPUs, and with ROCm 7 in September 2025, AMD claimed roughly 3.5 times better inference performance on MI300X compared with ROCm 6, The Register reported. Hyperscalers sidestep CUDA by tuning their own software for TPUs, Trainium or Maia.

Advanced AI chips depend on scarce manufacturing and memory capacity. Rivals still need factory slots, high-bandwidth memory and packaging to ship in volume.

The answer depends on which part of the market you watch. Here is how the main challengers line up by threat type:

For a broader look at who leads across AI, see our AI market share 2026 breakdown and our list of the biggest AI companies.

Who is Nvidia's biggest competitor?
AMD is Nvidia's biggest direct competitor in AI GPUs, with record data center revenue of $8.5 billion in Q3 2026. Broadcom is the biggest competitor by AI chip revenue, reporting $16.7 billion in AI semiconductor sales for its quarter ended August 2, 2026, from custom AI accelerators and networking.

What is Nvidia's market share in AI chips?
Morgan Stanley estimated in March 2026 that Nvidia held roughly 85% of AI processor revenue, with custom ASICs just above 10% and AMD below 5%. Bloomberg Intelligence projects Nvidia will hold 70% to 75% of the AI accelerator market through 2030.

Is Google's TPU a threat to Nvidia?
Google's TPU is one of the most credible alternatives because Google uses it internally and rents it to outside customers such as Anthropic. Google announced TPU 8t and TPU 8i in April 2026, and Anthropic agreed to about 3.5 gigawatts of TPU capacity starting in 2027.

Why is Nvidia hard to compete with?
Nvidia's CUDA software is used by more than 7.5 million developers, so switching chips often means reworking code.

Does Nvidia have competition in China?
Yes. Huawei held 28% of China's data center AI accelerator market in the first half of 2026 versus 34% for Nvidia, according to IDC. US export controls and Chinese restrictions have sharply limited Nvidia's sales there.

Nvidia still dominates AI chips by a wide margin, with $89.0 billion in quarterly data center revenue and analyst share estimates between 70% and 85%. No single competitor is close to matching its scale or software ecosystem.

The pressure is coming from several directions at once: AMD's GPU deals, custom chips from Broadcom and the hyperscalers, inference specialists, and Huawei in China. Watch the share estimates rather than the headlines. Morgan Stanley's forecast of a one to two point decline in 2026 is the kind of gradual erosion to track.

Written by the AI Business Weekly Editorial Team. Reviewed, edited and published by Sameer Khan. This article was AI-assisted, then reviewed by Sameer Khan before publishing.