Which AI chip startup is growing the fastest?
In our AI chip market deck, you will find everything you need to understand the market
SUMMARY
Etched is the fastest-growing private AI chip startup overall today, because no rival combines its scale of signed commercial demand, production progress, customer deployment and organizational expansion quite as strongly.
The race is accelerating because inference spending has become a live infrastructure market. AI infrastructure budgets are rising so quickly that challengers can grow very fast without taking much share from Nvidia.
Etched’s strongest number is not revenue but more than $1 billion of signed customer contracts. That is serious commercial demand, although it should not be confused with revenue already recognized.
Jane Street gives Etched a much more important proof point than another funding round: production hardware is being deployed inside a real customer data center after testing. The catch is that Jane Street is also an investor, so the next unrelated production customers will matter more.
Etched’s valuation has outrun the evidence we can verify. Going from a roughly $34 million seed valuation to $21 billion is a more than 600-fold increase, while public revenue disclosure remains thin.
Rebellions has the cleaner record on realized sales. Its revenue rose from roughly KRW 2.7 billion in 2023 to about KRW 32 billion last year, which makes it the strongest answer if the question is narrowed to proven revenue growth.
FuriosaAI could produce the most spectacular annual revenue multiple. Its KRW 80–100 billion target would imply roughly fourteenfold to seventeen-and-a-half-fold growth from last year, but that remains a target rather than a finished result.
DeepX is a smaller business, yet its order curve is unusually sharp: more than 60% of its first year of commercial purchase orders arrived in the latest four months. That looks like genuine acceleration, not just a slow accumulation of early customers.
Tenstorrent, d-Matrix and Positron are harder to rank because they disclose less financial data, but all three have crossed the line from prototypes into production systems. That makes them more serious than their quieter media profiles suggest.
The next phase of this race will be decided less by benchmark claims and more by manufacturing, repeat orders and customer concentration. Etched is ahead today, but its lead will look much more durable once several unrelated customers deploy, reorder and scale the hardware.
This market map, featured in our AI chip market deck, highlights top companies and startups in the AI chip market
Which AI chip startup is growing the fastest?
Why is the AI chip startup race suddenly moving so fast?
AI chip startups are growing unusually fast right now because inference has turned into a huge live infrastructure market rather than a future opportunity companies are waiting for.
Gartner recently estimated that worldwide spending on AI-optimized infrastructure-as-a-service will reach about $42.3 billion this year, almost double the previous year's $21.5 billion. The more revealing detail is the split: inference is expected to account for $23.3 billion, ahead of the $19 billion spent on training.
The hardware numbers are moving in the same direction. TrendForce recently raised its forecast for global AI-server shipment growth from more than 28% to nearly 31%. It also estimates that the nine largest cloud companies will spend more than $886 billion on capital expenditure this year, roughly 90% more than a year earlier.
Custom AI chips are already absorbing a meaningful part of that spending. Broadcom's latest quarterly AI semiconductor revenue reached $16.7 billion, up 221% year over year, with custom accelerators accounting for most of that business. Broadcom expects another jump to $21.7 billion in the following quarter.
That creates an unusually forgiving environment for challengers. An AI chip startup does not need to take meaningful share from Nvidia to double or triple. The underlying compute market is growing so quickly that several architectures can expand at once.
What should count as the fastest-growing AI chip startup?
For this comparison, the fastest-growing AI chip startup is the private company expanding its real commercial footprint fastest across revenue, customer orders, deployments and production.
Revenue deserves the most weight when we have it. A dollar of recognized revenue tells us that a customer received something and the seller earned money from it.
Private semiconductor companies rarely disclose enough revenue for a perfect comparison, so we also have to look at signed orders, purchase orders, chips already shipped and systems running at customers. Those measures become much weaker when they are still pilots, reservations or non-binding partnerships.
Percentage growth also needs context. Going from $2 million to $20 million is tenfold growth, but a company adding $500 million of business from a much larger base can be expanding much faster in economic terms.
We are also keeping the field genuinely private. Cerebras would otherwise deserve attention after generating roughly $510 million of revenue last year, but it is now publicly traded. Blaize is public too. Groq remains private, although its recent shift toward running an inference cloud makes it less comparable with a pure chip vendor.
That leaves companies such as Etched, FuriosaAI, Rebellions, DeepX, SambaNova, Tenstorrent, d-Matrix and Positron at the center of the comparison.
| Measure | What we learn from it | How much weight we give it |
|---|---|---|
| Recognized revenue | How much business has already been delivered | Very high |
| Commercial purchase orders | Whether testing is becoming real buying | High |
| Signed future contracts | How much demand has been committed | High, with caution |
| Production deployments | Whether customers are actually running the hardware | High |
| Units shipped or production ramp | Whether the company can move beyond prototypes | High |
| Funding and valuation | How much growth investors expect | Supporting evidence only |
As this chart shows, and as featured in our AI chip market deck, search interest in AI chips has grown significantly
Is Etched actually the fastest-growing AI chip startup today?
Etched is currently the strongest overall answer because its customer demand, production activity and company size are all expanding at the same time and at an unusually large scale.
When Etched raised its Series A in 2024, the company had only 35 employees and customers had reserved hardware worth tens of millions of dollars. Etched now has more than 400 employees and says it has secured more than $1 billion of signed customer contracts.
The company has also crossed the line that kills many semiconductor stories: somebody outside the company is using production hardware. Jane Street received Etched's first production rack and is deploying the system inside its own data center after testing the chip.
So there is more here than investor enthusiasm. In roughly two years, Etched has gone from a 35-person company working toward production silicon to a 400-plus-person manufacturer with a live customer installation and a contract book measured in billions rather than tens of millions.
DeepX can beat that growth on some percentage measures. FuriosaAI could post a much larger revenue multiple this year. Rebellions has already accumulated a stronger history of recognized sales. None of those companies, however, currently combines Etched's absolute commercial demand with such a rapid transition into production.
That is why Etched ranks first today.
If you want more recent data on this point, please see our latest AI chip market report.
Did Etched’s business really grow as fast as its $21 billion valuation?
Etched's business has grown exceptionally fast, but the company's $21 billion valuation has moved even faster than the commercial results we can actually verify.
The valuation curve is extreme. Etched's seed valuation was reportedly around $34 million. The company was valued at about $5 billion after a financing in late 2025, then at $10.3 billion in a Series C this summer. Only weeks later, another $700 million round led by Jane Street put the valuation at $21 billion.
From roughly $34 million to $21 billion is an increase of more than 600 times.
The operating progress underneath that valuation is substantial. Etched achieved working first-pass silicon in less than three years from its seed round, moved systems into customer testing and then into its first production deployment. The company is already developing three hardware generations in parallel rather than waiting for the first product cycle to finish.
Etched has also raised about $1.9 billion in total, giving it the balance sheet to hire, secure manufacturing and build full rack-scale systems rather than selling a chip on its own.
Still, we have no disclosed revenue figure remotely comparable with the valuation. Investors are pricing Etched as though today's early production ramp will become a very large semiconductor business quickly.
That may eventually be right. For now, the operating growth supports a big valuation increase, while the final leap to $21 billion still relies heavily on what customers are expected to buy next.
This chart, featured in our AI chip market deck, shows annual VC investment in AI chip startups
Are Etched’s billion-dollar contracts already revenue?
Etched's billion-dollar contract book is commercially important, but those contracts should not be treated as revenue the company has already earned.
Etched describes the figure as signed customer contracts. That wording is stronger than expressions of interest, reservations or vague partnership announcements. Customers have made formal commitments to buy Etched systems.
Revenue comes later as those obligations are fulfilled under the contracts. Etched has only recently entered customer production, so a meaningful part of the contracted value almost certainly sits ahead of the company rather than inside historical financial results.
We also do not know the average contract duration, delivery schedule, cancellation provisions or customer concentration. A contract spread over three years has very different implications from the same dollar amount delivered over twelve months.
There is still plenty to like here. Semiconductor startups often spend years talking about evaluations without showing that customers have moved into actual purchasing. Etched has clearly gone beyond that stage.
We simply should not turn a large backlog into an invented revenue number. Until Etched reports meaningful recognized sales, Rebellions and some Korean rivals remain easier to judge on financial execution.
If you want more recent data on this point, please see our latest AI chip market report.
Does Jane Street prove Etched’s AI chip works in production?
Jane Street gives Etched real production validation today, although one sophisticated customer cannot tell us yet how easily Etched can repeat that deployment across the rest of the market.
Jane Street received Etched's first production rack after testing the chip and said it was pleased with the early results. The system is now being deployed against Jane Street's own workloads inside its data center.
For a young semiconductor company, that is a meaningful jump from engineering samples. Rack-scale hardware has to survive much more than a controlled benchmark: networking, power, cooling, software integration, model compatibility and operational reliability all start to matter.
Jane Street is also a demanding customer. High-frequency trading firms spend heavily on compute and tend to care obsessively about latency and performance.
There is one complication. Jane Street is also an investor in Etched and led its latest financing round. We should therefore treat the customer win and investment as related pieces of evidence rather than two independent endorsements.
The next few customers will tell us far more. If unrelated AI labs and cloud providers put Etched hardware into production and reorder it, the company will have moved from promising first deployment to repeatable product-market fit.
Right now, Jane Street proves that Etched has built usable production hardware. Broader commercial proof is still coming.
This chart, featured in our AI chip market deck, shows how Nvidia is leading in AI chips
Is DeepX growing faster than Etched in edge AI chips?
DeepX is growing extraordinarily fast in edge AI right now, and its recent order velocity is stronger than the headline revenue forecasts suggest.
DeepX began mass-producing its DX-M1 on-device AI processor about a year ago. By the end of July, the Korean startup had secured 77 commercial purchase orders worth more than $13 million across more than ten countries and regions.
The pace has accelerated sharply. Forty-eight of those 77 orders arrived during the most recent four-month period. In other words, more than 60% of all the orders accumulated during the first year of mass production came in roughly the final third of that period.
DeepX had spent two to three years letting more than 400 companies evaluate samples before mass production started. We can now see some of that evaluation pipeline converting into buying. The company has also signed 27 semiconductor distribution partners and works with more than 50 hardware and module companies.
The international mix is encouraging as well. DeepX says overseas business now represents more than half of cumulative revenue.
Its previous full-year revenue was only about KRW 3.3 billion. Converting $13 million of announced purchase orders into won gives a value more than five times that historical annual revenue base, although purchase orders and recognized revenue obviously remain different accounting measures.
DeepX therefore has a credible claim to being one of the fastest-growing AI chip startups by percentage. Its commercial base is still much smaller than the leading data-center challengers, so we would not put it first overall.
Could FuriosaAI beat every AI chip startup on revenue growth?
FuriosaAI could deliver one of the biggest revenue jumps in the entire AI chip sector this year if its current production plan turns into sales.
The Korean startup generated about KRW 5.7 billion of revenue last year. FuriosaAI now says it is targeting between KRW 80 billion and KRW 100 billion.
The lower end would represent roughly fourteenfold growth. Hitting KRW 100 billion would push the increase to about seventeen and a half times in one year.
There is a physical production plan behind the target. FuriosaAI expects to manufacture around 20,000 units of its second-generation RNGD accelerator, with about 7,000 intended for customer sales. The rest will largely support ecosystem development.
Commercial use has already started. Samsung SDS offers RNGD through its NPU-as-a-service product on Samsung Cloud Platform. FuriosaAI hardware is also supporting AI-generated overview features for Daum users, giving the company exposure to an application serving millions of people rather than only laboratory tests.
Outside Korea, RNGD is available through infrastructure at Equinix in Lisbon, and FuriosaAI is involved in a planned 15-megawatt AI data-center project in Sweden.
We would still call the KRW 80–100 billion figure a target rather than an accomplished result. But unlike a revenue goal attached to a chip still sitting in development, FuriosaAI is currently mass-producing the product and already has live workloads.
If FuriosaAI reaches the top of its plan, it will have a better case for fastest percentage revenue growth than Etched.
If you want more recent data on this point, please see our latest AI chip market report.
This chart, featured in our AI chip market deck, shows annual funding in AI chip startups
Is Rebellions already more commercially proven than Etched?
Rebellions is currently more proven on realized AI chip sales than Etched, with several years of revenue growth already behind it.
Rebellions generated roughly KRW 2.7 billion of revenue in 2023, around KRW 10.3 billion in 2024 and approximately KRW 32 billion last year.
That means annual revenue expanded almost twelvefold in two years.
The growth came with physical chip sales. Rebellions sold thousands of ATOM and ATOM-Max accelerators before moving toward its second-generation Rebel100 platform.
Industry estimates now put this year's potential revenue above KRW 80 billion as Rebel100 deliveries begin. We treat that figure more cautiously because it is an outside forecast rather than reported revenue.
A recent international project makes the expansion more interesting. Rebellions signed a partnership with British startup Callosum to pursue the deployment of more than 100 NPU racks in a UK government-backed AI research cluster. The companies say later phases could involve additional racks.
That would be a much larger deployment format than selling individual accelerator cards and could help Rebellions move beyond its Korean customer base.
There is one reason to be careful with the historical comparison. Rebellions merged with SK Telecom-backed Sapeon Korea and later acquired AI software company SqueezeBits, so the company we see today is broader than the original standalone Rebellions business.
Even with that caveat, Rebellions has the clearest multi-year sales record among the private startups in this race.
Is SambaNova really growing again?
SambaNova is clearly growing again, although the lack of current revenue data makes it impossible to rank the company above Etched, FuriosaAI or Rebellions with much confidence.
The comeback has been dramatic on the financing side. SambaNova raised more than $350 million earlier this year around the launch of its SN50 platform. Five months later, it completed the first $1 billion close of another funding round at an $11 billion post-money valuation.
Customer names have improved at the same time.
JPMorganChase selected SambaNova as an inference-infrastructure partner for secure AI workloads running inside the bank. SoftBank is the first deployment partner for the next-generation SN50, which SambaNova plans to ship during the second half of the year.
Those are meaningful enterprise customers for a company that had spent several years looking much less threatening than its original hype suggested.
What we still cannot see is how many systems SambaNova is selling, what the current order book looks like or how quickly revenue is moving. A company can raise $1 billion because investors expect a comeback long before the comeback appears in financial statements.
SambaNova has earned its place back among the serious AI chip challengers. Calling it the fastest grower today would require numbers the company has not made public.
This chart, featured in our AI chip market deck, compares the main business model options for AI accelerator chip companies
Is Groq still an AI chip startup, or is it really an AI cloud company now?
Groq is currently one of the fastest-growing AI inference platforms, but the company has moved far enough toward cloud infrastructure that we would no longer use it as the cleanest answer to an AI chip startup question.
The recent user growth is substantial. Groq said it was serving more than five million developers in June. By August, that figure had passed six million.
That is at least 20% additional developer adoption in less than two months.
Groq also operates 13 data centers across North America, Europe, the Middle East and Asia-Pacific and says those systems process trillions of AI tokens every week.
Capital has followed the expansion. Groq raised $650 million in June and another $350 million in August, giving the company $1 billion of new funding in less than two months.
The business has changed, though. After Groq entered a major licensing agreement with Nvidia and several top technical leaders moved to Nvidia, the remaining company began describing itself much more explicitly as an inference neocloud. Its infrastructure business can increasingly be understood as selling fast inference capacity rather than simply selling its own LPU chips.
The distinction sounds technical, but it changes our ranking. If we were asking which private AI inference platform is adding developers fastest, Groq would be near the top. For a comparison centered on startups commercializing new AI processors, Etched, Rebellions and FuriosaAI are cleaner candidates.
If you want more recent data on this point, please see our latest AI chip market report.
Are Tenstorrent, d-Matrix and Positron quietly catching up?
Tenstorrent and d-Matrix are currently much further into commercial shipping than their lower media profiles suggest, while Positron reached a production customer unusually quickly for such a young chip startup.
Tenstorrent's Blackhole-based Galaxy systems are now in production and shipping in volume. The company says customers are already running clusters with 36 Galaxy systems connected as one computer. Cirrascale has also added Galaxy Blackhole hardware to its commercial AI cloud, giving outside customers bare-metal access.
d-Matrix has moved its Corsair inference accelerator into full production too. Parasail is deploying Corsair alongside Nvidia Hopper and Blackwell GPUs, while Gimlet has also announced production-oriented use of the hardware. The company has been expanding beyond silicon lately, acquiring GigaIO's data-center business and then Wallaroo.ai within four months to add networking and inference-deployment software.
Positron has a shorter history. According to the company's development timeline, it deployed its first full-scale production rack to a major cloud provider only 22 months after the company was started. Atlas systems have since been used for networking, gaming, content moderation, content delivery and token-as-a-service workloads. Positron raised more than $230 million earlier this year at a valuation above $1 billion.
We cannot rank these companies properly on revenue because none publishes enough financial detail. Their production progress still makes them much more serious challengers than startups whose main achievements remain benchmark claims and fundraising announcements.
| Startup | Where it is commercially today | Recent development worth watching |
|---|---|---|
| Tenstorrent | Blackhole Galaxy in production and volume shipping | Commercial cloud availability and 36-system clusters |
| d-Matrix | Corsair in full production | Parasail deployment plus two acquisitions in four months |
| Positron | Production Atlas racks deployed | Reached first major cloud production rack in month 22 |
This chart, featured in our AI chip market deck, shows how revenue is split across customer segments in the AI chip market
Can Etched manufacture fast enough to keep growing this quickly?
Etched's biggest test now is manufacturing: the company has created demand faster than it has had time to prove it can deliver hardware repeatedly and at large scale.
Etched's processor is manufactured on TSMC's N4P process, and the company sells complete inference clusters rather than bare chips. Scaling therefore involves far more than receiving good wafers. Etched has to assemble chips, boards, networking, software, power and cooling into reliable production racks.
The industry around it is also getting busier. TrendForce recently increased its AI-server growth forecast to nearly 31%, while hyperscalers are raising infrastructure spending at an extraordinary rate. More AI servers mean more competition throughout semiconductor and system supply chains.
Some rivals deliberately designed around those bottlenecks. d-Matrix, for example, manufactures Corsair on the more mature TSMC N6 process and uses an architecture built around SRAM, organic substrates and LPDDR5 rather than depending on the same HBM and advanced CoWoS packaging stack that constrains many high-end accelerators.
Tenstorrent already says Blackhole is shipping in volume. FuriosaAI is in mass production. Rebellions has already completed one commercial chip generation and is moving into another.
Etched is staffing manufacturing and testing operations in Taiwan and has enough capital to attack the problem aggressively. Still, this is the point where the company's growth story becomes much harder.
Signing customers can happen quickly. Supplying thousands of reliable accelerators quarter after quarter is where semiconductor companies show whether the growth is durable.
What could stop Etched from staying the fastest-growing AI chip startup?
Etched could lose its lead surprisingly quickly if production slips, customers fail to reorder or competing inference chips improve faster than Etched can scale.
The competitive bar keeps rising. Nvidia's latest quarterly Data Center revenue reached $89 billion, up 117% from a year earlier. An incumbent already operating at enormous scale is therefore growing faster than many startups.
Custom chips are accelerating too. Google, AWS, Meta and other hyperscalers are putting more money into their own accelerators, while Broadcom's custom-chip business has exploded. Etched is entering a market where customers have more alternatives every quarter.
Software could become just as important as raw chip speed. Nvidia customers inherit years of CUDA tools, libraries and engineering experience. An Etched system has to save enough money or deliver enough additional performance to justify moving important workloads onto a younger stack.
Customer concentration is another unknown. Etched has disclosed a large contract base but has not shown how evenly that demand is spread. A handful of very large buyers would make growth look spectacular while increasing the damage if one customer delays a deployment.
Model architecture presents a subtler risk. Etched originally became famous for building hardware tightly around transformers. The company has since broadened its approach to cover mixture-of-experts models and architectures such as Mamba, which reduces that exposure, but specialized hardware always carries some risk when AI models change quickly.
Etched therefore has the fastest overall growth story today and one of the hardest execution jobs in the sector. The next proof point is repeat buying from several unrelated customers, not another valuation increase.
If you want more recent data on this point, please see our latest AI chip market report.
This chart, featured in our AI chip market deck, shows how AI accelerator chip technology has evolved over time
Which AI chip startup is growing the fastest right now?
Etched is currently the fastest-growing AI chip startup overall, while DeepX, FuriosaAI and Rebellions each beat Etched on narrower definitions of growth.
Our judgment comes from combining several things rather than chasing the largest percentage.
Etched has moved from early silicon development into customer production unusually quickly. Its organization has expanded by more than ten times from the small team it had around its Series A. As seen above, signed commercial demand has already moved beyond $1 billion even though broad production is only beginning.
That gives Etched the strongest combination of absolute commercial expansion and operating growth we can find among independent private AI chip companies today.
DeepX looks more explosive when we focus on the pace at which small initial orders are multiplying. More than 60% of its first year's commercial purchase orders arrived during the latest four months, suggesting adoption is accelerating rather than simply accumulating.
FuriosaAI has the clearest chance of winning on annual revenue percentage. Moving from KRW 5.7 billion toward its KRW 80–100 billion target would mean roughly fourteenfold to seventeen-and-a-half-fold growth.
Rebellions wins a different argument. Revenue increased from roughly KRW 2.7 billion to KRW 32 billion in two years, backed by thousands of chips already sold. We have much less guessing to do about whether customers will actually pay for Rebellions hardware.
Those differences are exactly why the answer cannot come from one headline number.
If we ask which startup has the fastest-growing proven revenue base, Rebellions has the best case. If we ask which could post the largest revenue multiple this year, FuriosaAI could take it. DeepX is one of the strongest edge-AI growth stories.
But if we ask the broader and more useful question — which private AI chip company is expanding its real commercial position fastest right now — Etched is ahead.
The lead is real, although it is still young. Etched has already shown that customers want the hardware. What it has to prove next is that it can turn an extraordinary early order book into an equally extraordinary semiconductor business.
OUR METHODOLOGY
There is no single public metric that tells us which private AI chip startup is growing fastest. Revenue disclosure is uneven, companies are at different stages of production, and commercial progress can show up as recognized sales, signed orders, shipments or live customer deployments, so we treated the question as a multi-dimensional comparison rather than a one-number ranking.
We broke growth into the measures that most directly show a semiconductor company moving from technical promise to a larger commercial business: realized revenue and its trajectory, customer orders and contract momentum, production and shipment progress, live deployments, and the scale at which the company is building to support future deliveries.
We did not give every measure the same weight. Recognized revenue and hardware already operating with customers count more than funding, valuation or future revenue targets. Signed contracts and commercial purchase orders are useful evidence of demand, but we do not treat them as revenue before the products are delivered and the sales are recognized.
We also compared percentage growth with absolute commercial expansion. A company growing tenfold from a tiny base should not automatically outrank a company adding much more business at scale, while a larger company does not win simply because it is larger.
The comparison is limited to private companies whose core business still involves commercializing differentiated AI processor hardware. Publicly traded chip companies were excluded, and businesses that have shifted toward operating AI cloud infrastructure were treated separately when that made the chip-to-chip comparison less clean.
For each company, we prioritized fresh first-hand disclosures and then looked for customer, infrastructure-partner, investor or industry-research confirmation where it materially strengthened the claim. Older figures were mainly used as baselines so that recent growth numbers had something meaningful to be compared with.
Key sources include Gartner on AI-optimized IaaS spending and the inference/training split, TrendForce on AI-server shipment growth and hyperscaler capital expenditure, Broadcom on AI semiconductor revenue, and NVIDIA on Data Center revenue growth.
For Etched, the main first-hand sources are Etched on its production ramp, team size, TSMC N4P silicon and more than $1 billion of signed customer contracts and Etched on its first production rack at Jane Street and latest financing. For the Korean challengers, we used Seoul Economic Daily on DeepX's commercial order acceleration, Samsung SDS on commercial RNGD availability, FuriosaAI on its Equinix deployment, and Rebellions on Rebel100 production and rack-scale systems.
We also used SambaNova on its financing and JPMorganChase deployment, Groq on its inference-cloud footprint and developer base, Tenstorrent on Blackhole Galaxy production and volume shipping, Cirrascale on commercial Blackhole cloud availability, d-Matrix on Corsair deployment at Parasail, and Positron on its production-rack timeline and commercial workloads.
The final ranking comes from the combined evidence rather than the winner of any single metric. That is why Etched ranks first overall, while Rebellions, FuriosaAI and DeepX can still lead on narrower definitions such as proven revenue growth, potential annual revenue multiple or edge-AI order acceleration.
In our AI chip market deck, we identify pain points entrepreneurs should prioritize
Related blog posts
- Which AI chip startups are generating the most revenue today?
