What are the top AI chip startups by revenue today?

The ranking of top startups in the AI chip market by revenue

Below is a table ranking all the companies in this market by their current revenue scale. You can find our methodology at the end of this page.

If you want a deeper understanding of the market and its current dynamics, get our report on AI chip funding, valuations and who's ahead.

Which AI chip startups make the most revenue today?

Cerebras currently has the strongest full-year revenue figure among the independent AI-chip challengers we can measure cleanly, but several Chinese GPU companies are now much closer than most Western rankings suggest.

Cerebras reported $510 million of revenue for 2025. Behind it, Moore Threads generated RMB1.74 billion, roughly $240 million, in only the first half of 2026. MetaX reported RMB1.64 billion for 2025, while Biren, Enflame and Iluvatar CoreX have each moved well past RMB900 million in half-year revenue.

That creates an awkward definition problem. Moore Threads, MetaX and several other companies started as venture-backed AI-chip startups but have since listed on public markets. Excluding them would produce a cleaner "private startup" list, but it would also hide where the biggest challenger businesses have actually emerged. We therefore include recent startup graduates in the main commercial comparison and make the distinction clear when it affects the conclusion. Startups graduating to bigger businesses is a recent shift, and we look at how the market has moved over the past year in our AI chips report.

We also keep revenue separate from bookings, contracts and deployment value. That is why Etched's more than $1 billion of customer contracts does not put it above Cerebras yet.

Company	Strongest current commercial figure	Period	What it actually is
Cerebras	$510M	2025	Revenue
Moore Threads	RMB1.74B	H1 2026	Revenue
MetaX	RMB1.64B	2025	Revenue
Biren Technology	RMB1.24B	H1 2026	Revenue
Enflame	RMB1.12B	H1 2026	Revenue
Iluvatar CoreX	~RMB946M	H1 2026	Revenue
SambaNova	Around $100M	2025 estimate	Estimated revenue
Zhonghao Xinying	RMB598M	2024	Revenue
Rebellions	~KRW35B	2025	Revenue
FuriosaAI	KRW5.7B	2025	Revenue
Etched	>$1B	Current contracts	Signed customer contracts
Tenstorrent	~$150M	Disclosed in 2024	Deals closed

Is Cerebras still the biggest AI chip startup by revenue?

Yes. Cerebras currently has the best combination of scale, clean accounting evidence and a directly comparable revenue figure.

Its IPO filings show revenue rising from $78.7 million in 2023 to $290.3 million in 2024 and $510 million in 2025. That is 76% year-on-year growth in the latest full year.

The mix is useful too. Cerebras generated about $358 million from hardware and another $152 million from cloud and other services. We are looking at a real operating business rather than a large backlog being presented as if it had already become sales.

Cerebras has since moved beyond the classic private-startup stage, but leaving it out would make the AI-chip revenue picture less useful. Among companies built specifically to challenge incumbent AI-compute architectures, it remains the cleanest benchmark for current commercial scale.

Is Moore Threads catching Cerebras?

Moore Threads is the company putting the most immediate pressure on Cerebras in the revenue ranking.

Moore Threads generated RMB1.74 billion in the first six months of 2026, up 147% year over year. Its 2025 revenue had already reached about RMB1.51 billion, after growing more than 240%.

The pace inside 2026 is even more interesting. First-quarter revenue was RMB738 million, which means the second quarter alone was close to RMB1 billion. We do not annualize that figure and pretend it is a full-year disclosure, but Moore Threads is clearly operating at a much higher revenue level now than its 2025 accounts suggested.

The company is also much closer to breakeven. Its latest half-year loss narrowed dramatically while sales continued to climb. Moore Threads looks less like an experimental domestic GPU supplier these days and more like a scaled semiconductor business that is still growing unusually fast.

How much revenue does MetaX make?

MetaX generated RMB1.64 billion of revenue in 2025, putting it firmly among the largest Nvidia challengers created during China's recent GPU startup wave.

The growth curve is striking. MetaX went from roughly RMB53 million in 2023 to RMB743 million in 2024 and then more than doubled again in 2025. Its first-quarter 2026 revenue reached RMB562 million, up about 75% year over year.

The 2025 full-year number is already becoming slightly stale for understanding MetaX's current size. Still, it remains the cleanest annual figure available, so we keep it rather than inventing a 2026 run-rate from one quarter.

MetaX has reached a scale where the main questions are margins, software maturity, production capacity and whether growth can hold up as local competition gets much more crowded.

How big is Biren Technology now?

Biren has already become a nine-figure-dollar AI-chip company on a half-year basis.

The company reported roughly RMB1.24 billion of revenue in the first half of 2026. Its core intelligent-computing business accounted for almost all of that amount, which makes the figure unusually relevant for an AI-chip ranking.

The change from a year earlier was enormous. Biren's revenue increased by roughly twenty times as larger GPU deliveries and computing-cluster deployments began showing up in the accounts. Large GPU and cluster deployments lifted Biren twenty-fold, and we follow which technologies are gaining traction in our AI chips market report.

For years, Biren was discussed mainly through funding rounds, sanctions, benchmark claims and product launches. Its recent results make that framing outdated. There is now enough recognized revenue to judge the commercial execution directly.

Is Enflame much bigger now than its 2025 revenue suggests?

Yes. Enflame's current business is already substantially larger than its 2025 full-year figure makes it look.

Enflame produced RMB1.12 billion of revenue in the first half of 2026 after generating roughly RMB990 million during all of 2025. First-half sales increased about 279% from a year earlier.

The company's own outlook pointed to RMB2.3 billion to RMB3 billion of revenue over the first nine months of 2026. We do not use that forecast in the ranking because it had not yet become recognized revenue, but it shows how fast management expects the business to move.

Enflame's biggest weakness is concentration. Tencent-related business represented more than four-fifths of 2025 revenue. The sales are real, but losing or delaying one huge buyer would have an unusually large effect on the company.

How much revenue does Iluvatar CoreX make?

Iluvatar CoreX generated about RMB946 million of revenue in the first half of 2026, making it another Chinese AI-chip company that has quietly crossed into serious commercial scale.

Revenue increased about 192% from the same period a year earlier. More importantly for this ranking, almost all of the company's sales came from GPGPU products rather than unrelated semiconductor activities.

Inference products were the larger part of the mix, with training products making up the rest. Iluvatar also reported serving hundreds of customers.

That gives Iluvatar one of the cleaner revenue profiles in the Chinese challenger group. Its international profile is smaller than Cerebras, Tenstorrent or SambaNova, but its disclosed sales are already larger than those of many better-known private competitors.

Are Chinese AI chip startups now making more money than US challengers?

Outside Cerebras, the biggest documented challenger revenue figures currently come overwhelmingly from China.

Moore Threads, MetaX, Biren, Enflame and Iluvatar CoreX all have recent revenue disclosures that translate into roughly nine figures in US dollars. SambaNova is the only large private US peer we found around the same broad level, and its revenue figure is less transparent because it relies on outside estimates rather than filed financial statements.

The pattern is too broad to dismiss as one exceptional company. Several Chinese GPU businesses accelerated at the same time as domestic cloud operators, telecom companies and AI-compute projects increased purchases of local hardware. Domestic buyers turning to local hardware opened up room, and we map where the biggest opportunities are in our report covering AI chips.

That does not tell us that these chips are technically better than US alternatives or Nvidia. Revenue answers a narrower question: who is actually getting substantial quantities of accelerator hardware into paying customers today?

Company	Recent revenue growth evidence	Latest revenue used
Biren	Roughly 20x YoY in H1	RMB1.24B H1 2026
Enflame	+279% YoY in H1	RMB1.12B H1 2026
Iluvatar CoreX	+192% YoY in H1	~RMB946M H1 2026
Moore Threads	+147% YoY in H1	RMB1.74B H1 2026
MetaX	+121% in 2025	RMB1.64B FY2025
Cerebras	+76% in 2025	$510M FY2025
Rebellions	Roughly 3.4x in 2025	~KRW35B FY2025

Is SambaNova really a $100 million AI chip company?

SambaNova appears to be around the $100 million revenue level, but we have less confidence in that figure than in the public-company results above.

Outside estimates place SambaNova's 2025 revenue at roughly $100 million. SambaNova itself has said it finished 2025 with record revenue and bookings, but the company has not published a precise annual revenue figure in the same way Cerebras or listed Chinese companies have.

There is plenty of evidence that SambaNova remains commercially active. It has raised more capital, announced the SN50 accelerator, signed a multi-year relationship with Intel and added major customers for on-premise AI infrastructure.

Those developments strengthen the case that SambaNova has a substantial business. They still do not turn an outside estimate into an audited number, so we rank the company more cautiously.

How much revenue does Zhonghao Xinying make?

Zhonghao Xinying generated RMB598 million of revenue in 2024, one of the stronger historical revenue figures among smaller Chinese AI-accelerator companies.

The business had grown from about RMB82 million in 2022 to RMB485 million in 2023, and it became profitable during that expansion.

The problem is freshness. Its first-half 2025 results were much weaker, with revenue around RMB102 million and a substantial loss. Management pointed to seasonality, but the older RMB598 million number cannot simply be treated as the company's current annual scale.

We keep Zhonghao Xinying in the ranking because an audited historical revenue figure is still more useful than a speculative estimate. Its exact current position is less certain than the companies publishing 2026 half-year numbers.

Is Etched already one of the biggest AI chip startups?

Etched has become one of the biggest AI-chip startups by contracted business, although recognized revenue still has to catch up.

The company disclosed more than $1 billion of signed customer contracts when it emerged from stealth. Since then, Etched has moved another step forward: its first rack has been delivered to Jane Street, and the company says it is scaling production for the rest of its backlog.

That makes the Etched story stronger than it was only a few months earlier. The company now has working silicon, real rack-scale systems, large customer commitments and its first customer delivery.

We still would not write "$1 billion revenue" next to Etched. Contracts can be recognized over time, deliveries can span several periods and contract value does not necessarily equal revenue already booked in the accounts. Etched could move very high in this ranking once financial disclosures catch up with its order book.

Does Tenstorrent really make $150 million in revenue?

No. Tenstorrent's widely repeated $150 million figure refers to deals closed, not annual revenue.

Tenstorrent used the number itself when announcing its Series D financing, describing roughly $150 million of closed deals as evidence of commercial traction.

That is still meaningful. Tenstorrent sells AI systems and also licenses AI and RISC-V intellectual property, so commercial contracts can come from several different parts of the business. Selling systems and licensing IP are two ways to earn from the same chips, so we compare which business models are winning in our AI chips report.

The distinction is simple but important: a signed deal can produce revenue over multiple years or depend on milestones. Until Tenstorrent publishes a proper annual revenue figure, $150 million should stay labeled as deals rather than sales.

How much real business does Positron have now?

Positron already has meaningful commercial deployments, even though the company has not published a clean annual revenue number.

CEO Mitesh Agrawal told EE Times that Positron was deploying multiple tens of millions of dollars' worth of systems and racks into Oracle's cloud infrastructure. Positron now says more than 50 racks have been deployed at Oracle.

That puts the company well beyond the evaluation-board stage. Oracle is running actual inference infrastructure built around Positron hardware, and the deployment has continued to expand.

We would still avoid converting deployment value directly into recognized revenue. But among younger US accelerator startups, Positron has some of the strongest evidence of real hyperscaler adoption currently available.

How much revenue does Rebellions make?

Rebellions generated about KRW35 billion of revenue in 2025, roughly 3.4 times the previous year's level.

The Korean NPU company made around KRW10.3 billion in 2024 and only KRW2.7 billion in 2023. Revenue increased more than tenfold in two years.

Thousands of ATOM and ATOM-Max chips were sold during the latest expansion, while Rebellions continued moving newer products toward larger AI-infrastructure deployments.

Rebellions remains much smaller than the leading Chinese GPU companies, but it is already one of the more substantial private or recently private AI-chip businesses outside China and the US.

Is FuriosaAI finally generating meaningful revenue?

FuriosaAI is now selling its chips commercially, although its 2025 revenue was still small at KRW5.7 billion.

What makes FuriosaAI interesting today is the size of the planned jump. The company has targeted KRW80 billion to KRW100 billion of revenue as production of its Renegade accelerator ramps.

Reaching the top of that range would represent roughly seventeen times last year's sales. FuriosaAI also expects production volume to rise sharply as the accelerator moves into customer deployments.

For now, we keep KRW5.7 billion in the ranking because that is actual reported revenue. The larger figure remains a target until it shows up in financial results.

Which AI chip startups have actually crossed $100 million in revenue?

Cerebras, Moore Threads, MetaX, Biren, Enflame and Iluvatar CoreX all have strong evidence of at least roughly $100 million of recognized annual or half-year revenue.

SambaNova probably belongs around the threshold too, although its figure is externally estimated rather than directly filed.

That shortlist says a lot about how the market has changed. A few years ago, people often judged AI-chip startups through venture funding, benchmark performance or the size of their engineering teams. We can now compare a meaningful group using real sales. Real sales let us judge companies on more than funding and benchmarks, and we look at which ones lead today in our report on AI chips.

Etched could join the very top group quickly if its billion-dollar contract book turns into recognized revenue at anything close to its disclosed scale. Positron and Tenstorrent also have substantial commercial evidence, but the public metrics are still different enough that forcing them into the same revenue table would create false precision.

What happened to Groq in the AI chip startup ranking?

Groq no longer fits neatly into a ranking of independent AI-chip hardware startups.

Groq signed a non-exclusive technology licensing agreement with Nvidia, while founder Jonathan Ross and other members of the Groq team moved to Nvidia. Groq remained an independent company, but its business increasingly shifted toward operating an AI inference cloud.

That change has become even clearer lately. Groq now describes itself as an inference neocloud, operates data centers across several regions and has joined Nvidia's cloud partner program. Its current infrastructure plans also include Nvidia hardware.

Groq therefore still belongs in the broader AI-inference market, but comparing today's Groq directly with Cerebras, Moore Threads or Etched as if they were still pursuing the same business model would be misleading.

Are the biggest AI chip startups profitable yet?

Several AI-chip challengers now have serious revenue, but profitability remains inconsistent.

Cerebras produced $510 million of revenue in 2025 and reported positive GAAP net income, although its filings also show a non-GAAP loss once unusual valuation effects are removed. Moore Threads has pushed close to breakeven as its revenue expanded. Iluvatar CoreX reported a profit in its latest half-year.

Other companies remain deep in investment mode. MetaX was still losing money despite RMB1.64 billion of annual revenue, while Enflame continued spending heavily as sales climbed.

That is normal for companies competing in accelerators. New architectures require repeated chip generations, software investment, expensive packaging and large manufacturing commitments. Revenue tells us who has found customers; it still does not tell us who has built a durable economic model.

Does Enflame's dependence on Tencent make its revenue less impressive?

Enflame's revenue is real, but its unusually high dependence on Tencent makes that revenue less diversified than the headline number suggests.

Tencent-related business represented more than 80% of Enflame's 2025 sales. For a young semiconductor company, one customer can easily absorb a huge production run, so concentration is not automatically a sign of weak demand.

It does create risk. A purchasing delay, product transition or strategy change at one buyer can move Enflame's annual revenue far more than it would at a supplier with dozens of similarly sized customers. A single large buyer is a risk for any young chip company, and it belongs with the biggest unsolved problems we cover in our report covering AI chips.

We keep the full revenue number while treating customer concentration as one reason to be careful about extrapolating today's growth indefinitely.

Are most new AI chip startups still pre-revenue?

Yes. A small group now has hundreds of millions of dollars in sales, but much of the AI-chip startup market is still somewhere between first silicon and early customer deployments.

d-Matrix has already entered production and is shipping its Corsair inference platform, but it has not published a useful revenue number. MatX, Fractile, Neurophos, Lumai and several photonic-computing startups remain earlier in the commercial cycle.

Etched shows how quickly that can change. It moved from stealth and first silicon to more than $1 billion of contracts and a customer rack delivery in a short period. Positron has also progressed from a young accelerator company into a deployment of more than 50 racks at Oracle.

So "pre-revenue" covers companies at very different stages. Some still need to prove their chips work in customer environments; others are already shipping hardware and simply have not disclosed enough financial information to rank them properly.

Company	Current commercial stage	Revenue figure we can use?
Etched	First customer rack delivered; >$1B contracts	Not yet
Positron	50+ Oracle racks deployed	Not cleanly
Tenstorrent	Products and licensing deals	No current annual figure
d-Matrix	Volume production and shipments	No
MatX	Pre-scale commercialization	No
Fractile	Moving toward first customer systems	No
Neurophos	Commercialization development	No
Lumai	Customer evaluation stage	No

Is AI chip startup revenue really accelerating this fast?

Yes. The recent growth is broad enough that we are seeing a real commercialization wave rather than one lucky company.

Cerebras grew revenue 76% in 2025. Moore Threads increased first-half 2026 revenue 147%. Iluvatar CoreX grew almost 192%. Enflame grew 279%. Biren's half-year increase was dramatically larger still. MetaX more than doubled annual revenue in 2025.

Those companies use different architectures and sell into different customers. We are seeing wafer-scale systems, general-purpose GPUs, GPGPUs and dedicated AI accelerators all convert into larger commercial businesses.

China currently accounts for much of the visible acceleration because its listed companies disclose detailed financials and domestic demand for local AI hardware is booming. The next question is whether younger US companies such as Etched and Positron can produce a similar jump once current contracts and deployments begin appearing as reported revenue.

So which AI chip startups are making the most money today?

Cerebras currently has the strongest proven revenue position, while Moore Threads is the clearest fast-moving challenger and China's broader GPU group now dominates the next tier of disclosed sales.

Cerebras remains our number one because $510 million of full-year revenue is both large and cleanly documented. Moore Threads follows with RMB1.74 billion in only six months and could materially change the ranking once a complete 2026 year is available. MetaX, Biren, Enflame and Iluvatar CoreX already sit at a level that many better-known Western startups have never publicly demonstrated.

Among still-private companies, the picture is less tidy. SambaNova appears to be around $100 million of revenue but lacks the same disclosure quality. Rebellions has a smaller but fast-growing revenue base. Positron has large Oracle deployments without a comparable annual revenue number. Tenstorrent has meaningful deal value rather than disclosed sales.

Etched is the company most likely to disrupt the ranking next. More than $1 billion of customer contracts, working hardware and a first customer delivery give it a much stronger commercial position than a normal pre-revenue chip startup. We still need to see how quickly those contracts turn into reported revenue.

The broader conclusion has changed significantly. AI-chip startups are no longer mainly companies raising huge rounds while waiting for customers. A real revenue hierarchy has formed, and several challengers are already selling accelerators at a scale measured in hundreds of millions of dollars.