What are the top AI inference chip startups by total funding raised?

Top startups in the AI inference chips market ranked by total funding raised

Here is an updated table that ranks the top startups in the AI inference chips market based on the total amount of funding they have raised to date.

The table also includes the total number of funding rounds, the date and size of the latest round, the financing type (e.g. Series A, equity financing), key investors, the startup’s current status (active, IPO, acquired, or shut down), and a confidence score based on the data collected (we excluded startups with very low data confidence, to make sure everything is reliable).

If you need to dig deeper and get more detailed data, please check our report on AI chip funding, valuations and who's ahead.

Which AI inference chip startups have raised the most funding?

Cerebras Systems, SambaNova Systems, Groq, Etched and Positron AI currently form the clear top tier of the AI inference chip funding race, with each having raised at least about $1.2 billion in the dataset we reviewed.

The gap becomes meaningful very quickly. Cerebras leads the ranking at roughly $2.8 billion of cumulative private funding, followed by SambaNova at about $2.5 billion and Groq at roughly $2.0 billion. Etched is already around $1.6 billion despite being much younger, while Positron has jumped to roughly $1.2 billion after one of the largest semiconductor startup rounds of the year.

Tenstorrent and Ayar Labs make up the rest of the billion-dollar group. Below them, Rebellions sits at roughly $850 million, while D-Robotics, MatX and several Chinese chip developers have moved into the $500 million-plus range.

The top of this market has become expensive fast. Seven companies in our dataset have reached roughly $1 billion or more, 13 are at $500 million or more, and 43 have crossed $100 million.

Rank	Company	Approx. cumulative funding	Main focus
1	Cerebras Systems	$2.8B	Wafer-scale AI compute
2	SambaNova Systems	$2.5B	Dataflow inference systems
3	Groq	$2.0B	LPU-based inference cloud
4	Etched	$1.6B	Transformer-specific ASICs
5	Positron AI	$1.2B	Memory-centric inference
6	Tenstorrent	$1.0B	AI accelerators and RISC-V
7	Ayar Labs	~$1.0B	Optical AI interconnects
8	Rebellions	~$851M	Datacenter inference accelerators
9	Sunrise	~$840M	Chinese inference GPUs
10	D-Robotics	~$770M	Robotics and edge AI chips

How much money has been invested in AI inference chip startups?

The 85 companies in our dataset have raised roughly $26 billion between them, and the real amount is somewhat higher because several Chinese and private-company totals are reported only as minimums or approximations.

That $26 billion figure is more useful as an order of magnitude than as an accounting-perfect market total. Some companies report individual rounds clearly, while others disclose only “more than” amounts, local-currency ranges or financing events with incomplete splits. A few companies also sit near the edges of the category, including optical-interconnect specialists and companies spanning training and inference.

Around $10.1 billion sits in just the first five companies. The top ten represent roughly $14.6 billion, while the top twenty account for about $19.4 billion.

So while AI inference chips look like a market with dozens of funded competitors, most of the capital is already controlled by a fairly small group. The bottom 65 companies combined account for only about one quarter of the capital represented in the dataset.

Is Cerebras still the most funded AI inference chip company?

Yes. Cerebras remains the best-funded company in this AI inference chip ranking, although calling it a startup now requires a qualification because Cerebras has already gone public.

Before its IPO, Cerebras closed a $1 billion Series H at a roughly $23 billion post-money valuation. The company subsequently completed its Nasdaq IPO and raised about $6.38 billion in gross public-market proceeds. A $23 billion private valuation followed by an IPO shows how fast worth can move; we cover what each company is worth, then and now, in our AI chips report.

We do not add those IPO proceeds to the private-startup funding ranking. That would turn the exercise into a comparison between venture funding and public equity issuance. The same applies to Cerebras' debt facilities.

Cerebras has also moved beyond the stage where funding is mainly financing chip development. Its first-quarter results showed $193.4 million of quarterly GAAP revenue, while the company has announced major infrastructure agreements including a multi-year OpenAI relationship and an AWS partnership.

It therefore remains number one on cumulative startup funding, but its position is no longer directly comparable with that of an earlier-stage chip company still financing its first large deployments.

Is SambaNova catching Cerebras in AI chip funding?

SambaNova has nearly caught Cerebras on cumulative private funding, with roughly $2.5 billion raised versus about $2.8 billion for Cerebras in our reconstruction.

Its latest financing changed the comparison dramatically. SambaNova announced the first close of a $1 billion Series F at an $11 billion post-money valuation, led by General Atlantic with participation from investors including Seligman Ventures, T. Rowe Price, BlackRock-managed funds and Qatar Investment Authority.

That round is unusually large even by AI infrastructure standards. A single financing event added an amount equivalent to the entire cumulative funding of Tenstorrent.

SambaNova and Cerebras are now at different corporate stages, though. Cerebras has crossed into the public markets, while SambaNova remains private. SambaNova could therefore become the most heavily venture-funded private pure-play company in the group without necessarily becoming the largest company operationally.

How much funding has Groq actually raised?

Groq belongs in the roughly $2 billion funding tier, but its recent corporate restructuring makes the headline total less straightforward than it looks.

Groq raised $640 million in 2024 at a $2.8 billion valuation. Then, after the company's non-exclusive technology licensing agreement with NVIDIA and significant management changes, the current Groq entity announced another $650 million of growth capital in June 2026.

Only weeks later, Groq LLC closed a $350 million Series A led by Disruptive, with planned participation from NVIDIA. Groq explicitly described those two recent financings as $1 billion of recent funding.

That unusual “Series A” label does not mean Groq suddenly became an early-stage startup. It reflects the corporate structure that followed the NVIDIA licensing transaction. The business already operated 13 data centers and said it served more than six million developers by the time of the August financing.

We therefore keep Groq among the market's best-capitalized inference companies while making the restructuring visible. Reading “Series A” beside a $350 million round without that context would be pretty misleading.

Why has Etched raised so much money so quickly?

Etched has become one of the most aggressively financed AI chip startups because investors are funding it as a potential large-scale inference manufacturer rather than as a normal early-stage semiconductor company.

The company announced a $700 million financing at a $21 billion valuation after shipping its first rack to Jane Street. Etched said Jane Street tested the hardware before leading the round, giving the financing more commercial weight than a round based only on a future chip roadmap.

Etched also says it has raised about $800 million through four previously unannounced financings, on top of its disclosed rounds. Our reconstructed cumulative figure is around $1.6 billion.

The speed of the valuation change is striking. Etched was valued around $500 million in its 2024 Series A. Two years later, it was raising capital at $21 billion.

That does not prove Etched will become the next NVIDIA-scale chip company. It does show how aggressively investors are willing to fund specialized inference silicon once working hardware, customer testing and a plausible path to large deployments are visible.

Has Positron AI become one of the best-funded NVIDIA challengers?

Yes. Positron AI has moved from a relatively small semiconductor startup into the top five of our inference funding ranking after raising $875 million in a Series C.

The jump happened extremely quickly. Positron had a $50 million-plus Series A in 2025, followed by a $230 million Series B at a valuation above $1 billion. Its subsequent Series C valued the company at $5 billion. A $50 million Series A to a $5 billion Series C in a year is quick; we show who is growing fastest in our AI chips market report.

That progression brings the disclosed financing history to well above $1.1 billion even before smaller earlier funding is considered.

The company says more than 50 of its racks had already been deployed at Oracle before the Series C. That gives the funding a useful commercial context: investors are financing a scale-up from deployed systems toward Positron's next-generation Asimov silicon, rather than funding a chip design that exists only on a slide.

Positron has entered the same financial conversation as Cerebras, SambaNova, Groq and Etched much faster than its age would suggest.

What are the largest funding rounds in AI inference chips?

The largest recent AI inference chip rounds are now approaching or reaching $1 billion, a level that would have looked extreme for a semiconductor startup only a few years ago.

SambaNova and Cerebras each raised $1 billion private rounds. Positron followed with $875 million, while Etched raised $700 million. Tenstorrent's $693 million Series D, enormous when announced in 2024, now has several comparable transactions around it.

MatX is another revealing case. Its $500 million Series B is intended to take its LLM accelerator through final development and manufacturing scale, with tapeout still ahead when the financing was announced. Raising $500 million with tapeout still ahead is a bet on the future, so we weigh the likely future winners in our report on AI chips.

These rounds increasingly finance manufacturing capacity, racks, data centers, software, customer deployment and supply-chain commitments alongside chip design itself.

Company	Recent major round	Round type	What it was funding
Cerebras	$1.0B	Series H	AI infrastructure expansion
SambaNova	$1.0B	Series F	Enterprise inference scale
Positron AI	$875M	Series C	Asimov silicon and Titan production
Etched	$700M	Series D	Production and large-scale deployment
Tenstorrent	$693M+	Series D	AI compute and silicon roadmap
MatX	$500M	Series B	Chip development and manufacturing
Unconventional AI	$475M	Seed	New AI computing architecture
Rebellions	$400M	Pre-IPO	Production and global expansion
D-Robotics	$400M	Series C	Robotics chip and software platform
EnCharge AI	$310M	Series C	Analog in-memory AI compute

How many AI inference chip startups have raised more than $1 billion?

Only seven companies in our dataset are currently at roughly $1 billion or more, making the billion-dollar funding club much smaller than the overall startup count suggests.

Those seven companies together account for approximately $12.1 billion, close to half of all funding represented by the 85 companies in the dataset.

Most of the group is focused on datacenter-scale infrastructure rather than tiny embedded inference devices. Ayar Labs is the exception: it is primarily an optical interconnect company rather than an inference accelerator vendor in the narrowest sense.

Reaching $1 billion is closely associated with problems that require enormous physical deployment: datacenters, racks, advanced manufacturing, packaging and infrastructure-scale systems.

Company	Approx. funding	Core approach
Cerebras Systems	$2.8B	Wafer-scale systems
SambaNova Systems	$2.5B	Dataflow processors
Groq	$2.0B	LPU inference infrastructure
Etched	$1.6B	Transformer ASICs
Positron AI	$1.2B	Memory-centric accelerators
Tenstorrent	$1.0B	AI accelerators / RISC-V
Ayar Labs	~$1.0B	Optical interconnect infrastructure

Is AI inference chip funding concentrated in just a few companies?

Very much so. Roughly 56% of the capital in our dataset sits in the top ten companies, even though those companies represent less than 12% of the names we tracked.

The top five hold about 39% of the funding. The top twenty control roughly 74%. That leaves 65 companies sharing the remaining quarter.

So this is not one broad, evenly distributed AI-chip funding boom. Investors are backing plenty of architectures, but the largest checks are going to a narrow set of companies considered capable of financing fabrication, systems deployment and major customer installations.

There is still plenty of experimentation lower down the market. Startups such as SEMRON, GEMESYS, Mosaic SoC, Sensesemi and optoML are trying architectures ranging from memcapacitors to neuromorphic and optical computing. Financially, though, they are playing a very different game from the companies building billion-dollar capital bases.

Group	Approx. funding	Share of dataset funding
Top 5	$10.1B	39%
Top 10	$14.6B	56%
Top 20	$19.4B	74%
Top 40	~$24.4B	~94%
Companies 41–85	~$1.6B	~6%
Companies with ≥$1B	7	-
Companies with ≥$500M	13	-
Companies with ≥$100M	43	-

Are datacenter AI chips attracting more money than edge AI chips?

Yes. The biggest pools of funding currently sit around datacenter inference, where startups need enough capital to finance advanced silicon, memory, packaging, complete servers and sometimes their own cloud infrastructure.

Cerebras, SambaNova, Groq, Etched, Positron, Tenstorrent, Rebellions, MatX, d-Matrix and Fractile all sit toward the upper part of our ranking. Their products differ substantially, but each is chasing high-value workloads where large language models or other demanding AI systems run at scale.

Edge specialists generally sit lower. SiMa.ai has raised $355 million, Hailo raised about $341 million before its acquisition, Axelera has substantial funding across equity and other capital sources, and companies such as DEEPX, Quadric, Mobilint, MemryX and Expedera are mostly below the datacenter leaders.

The difference makes economic sense. An edge startup can build a meaningful semiconductor business without financing hundreds of megawatts of compute capacity. A company trying to serve frontier-model inference may need billions before it can compete at the system level.

The funding gap therefore reflects the business model as much as investor enthusiasm. Datacenter inference consumes more capital because the product increasingly extends beyond the chip itself.

Are photonic AI chips finally attracting serious funding?

Photonics is attracting serious money now, although most photonic AI companies still sit well below the largest electronic-compute startups.

Ayar Labs is the big exception. Its $500 million Series E brought cumulative funding to $870 million, and an additional $150 million financing later expanded the capital raised during 2026. The company focuses on co-packaged optical connectivity rather than replacing a GPU's compute cores, but its presence in the ranking shows how closely inference economics are becoming tied to data movement.

Processor-focused photonics companies remain smaller. Neurophos announced a $110 million Series A. Arago raised $26 million in seed funding. Lumai, CogniFiber and Opticore remain in the low tens of millions in our dataset. With photonic processor makers still raising far less than the leaders, we watch which technologies are gaining traction in our report covering AI chips.

Investors increasingly believe optical technology will matter to AI compute, but they have not yet converged on a photonic processor architecture worth funding at the scale of Etched or Positron.

Ayar Labs also shows why the category boundary gets messy. AI inference funding is no longer only about the arithmetic engine. Memory bandwidth, optical links and how thousands of processors communicate can be just as important to the economics of running large models.

Who are the biggest investors in AI inference chip startups?

There is no single dominant investor in AI inference chips. The strongest pattern is the overlap between large financial investors and strategic semiconductor companies.

NVIDIA appears around companies such as Ayar Labs and Groq. AMD has invested around the optical-interconnect ecosystem. Samsung-related entities appear around Tenstorrent, Rebellions, Axelera and other semiconductor companies. Arm has backed companies including Rebellions and Positron.

Large financial investors repeat too. Tiger Global appears in the Cerebras and Etched investor groups. QIA has backed SambaNova, Positron, Ayar Labs and d-Matrix-related financing. Fidelity appears around multiple late-stage AI infrastructure companies. Jane Street has gone further by leading financings for both Etched and MatX.

Jane Street is an interesting case because it is not a semiconductor manufacturer but is a sophisticated buyer of extremely low-latency computing. Its role as both an investor and early Etched customer shows why the investor map now includes organizations that understand the economics of computation directly.

The financing ecosystem has become unusually mixed: venture funds still matter, but sovereign funds, chipmakers, asset managers, trading firms and prospective customers increasingly write some of the biggest checks.

Is China producing heavily funded AI inference chip startups too?

Yes. Chinese AI chip companies collectively represent several billion dollars of funding in our dataset, although public disclosure is much less uniform than for the leading U.S. companies.

Using the available financing estimates, the Chinese companies we identified account for roughly $4.5 billion. That figure should be treated as approximate because several rounds are disclosed in renminbi ranges, some totals are only minimum values, and many companies have weaker public financing histories than U.S. peers. With Chinese totals resting on renminbi ranges and minimum values, we handle who is ahead and who is falling behind carefully in our AI chips market report.

The largest names are still substantial. Sunrise is around $840 million in our reconstruction, EVAS Intelligence above $575 million, NanoCoreChip above $475 million, Kunlunxin above $340 million, TsingMicro above $312 million and Axera Semiconductor around $295 million.

The funding pattern is different too. Chinese rounds frequently involve local government-linked funds, industrial funds, securities-company investment arms and strategic investors, so simply looking for familiar Silicon Valley venture firms misses a large part of the market.

China has real depth in inference silicon. What our public dataset does not yet show is a Chinese company with the clearly documented $1 billion-plus private-funding profile of Cerebras, SambaNova, Groq or Etched.

Is South Korea becoming a serious AI inference chip funding hub?

South Korea has already become one of the most important countries outside the United States for funded AI inference chip startups.

Rebellions is the standout. Its $400 million pre-IPO round followed a $250 million Series C and brought cumulative funding to roughly $850 million. The company said more than three quarters of its capital had been raised in just six months around those transactions.

FuriosaAI has raised $246 million after its $125 million Series C bridge. DEEPX sits around $128 million, BOS Semiconductors around $84 million, Mobilint around $72 million, iHW around $46 million and HyperAccel around $36 million in our dataset.

These companies are not all chasing the same customer. Rebellions and Furiosa are pushing further into datacenter inference, while DEEPX and Mobilint have stronger edge positions.

Korea has produced several independently financed AI processor companies rather than one national champion surrounded by tiny startups. Rebellions is clearly ahead financially, but the second tier is large enough to make the ecosystem meaningful on its own.

Are new AI chip startups raising huge rounds before they even reach scale?

Yes. One of the stranger features of today's inference market is how much capital some companies can raise before reaching mature commercial scale.

MatX raised a $500 million Series B while saying it would use the money to complete development and move toward tapeout. EUCLYD raised more than €200 million in a Series A for an AI infrastructure architecture whose commercial roadmap extends years into the future. Neurophos raised $110 million in a Series A as it moved toward commercialization.

Unconventional AI is the most extreme example in the dataset: roughly $475 million in a seed round.

Advanced AI hardware often requires the reverse sequence of a conventional software startup. A software company can sometimes find product-market fit cheaply and scale later. A chip company may need hundreds of millions of dollars for tapeout, manufacturing capacity, software and complete systems before customers can properly evaluate the finished product.

That puts a remarkable amount of technological risk inside early financing rounds. Investors are making scale-sized bets before scale has been demonstrated.

Does raising more money mean an AI chip startup is actually winning?

No. Funding tells us which AI chip startups have convinced investors to finance their plans, but deployment and commercial adoption tell us much more about who is actually winning.

Cerebras offers the clearest example because it has already converted years of financing into meaningful reported revenue and public-company scale. Groq says it operates 13 data centers and serves millions of developers. Positron says more than 50 racks were deployed at Oracle before its latest major raise.

Etched is at a different stage. Shipping its first rack to Jane Street was an important commercial milestone, but one rack does not yet make Etched operationally comparable with companies running production infrastructure across several regions. A single delivered rack is a long way from a production fleet, and we dig into the biggest unsolved problems in our AI chips market report.

MatX provides an even sharper contrast. A $500 million Series B puts it among the best-funded startups in the market, while the company was still working toward tapeout when the financing was announced.

Funding is best read as capacity to compete. It buys engineering talent, wafers, packaging, datacenters, time and credibility with customers. Turning those resources into a large business is the harder part.

Which AI inference chip companies have already gone public or been acquired?

Only a small minority of the companies in this ranking have reached an IPO or acquisition, so most of the market remains privately financed.

Cerebras is now publicly traded on Nasdaq after completing its IPO. Blaize also trades publicly following its business combination with BurTech, which closed in early 2025.

Hailo moved in the opposite direction. Microchip Technology completed its acquisition of the edge AI processor company after announcing the transaction earlier in 2026. That makes Hailo's roughly $341 million historical venture funding useful for understanding how much capital went into the business before its exit, but Hailo is no longer an independent startup.

These outcomes show the different routes opening up. Some inference companies may stay private for a long time because private investors are willing to provide enormous amounts of capital. Others can use the public markets to finance infrastructure at a scale venture capital struggles to match. Strategically valuable edge-chip companies may simply end up inside larger semiconductor groups.

That is why company status has to sit beside funding totals in this ranking.

Are AI inference chip funding rounds getting much bigger?

Yes. The upper end of AI inference financing has moved into a very different range from just a few years ago.

Tenstorrent's $100 million strategic round in 2023 was already substantial. Its $693 million Series D in 2024 looked exceptional. Today, that $693 million financing would sit behind $1 billion rounds from Cerebras and SambaNova and Positron's $875 million Series C, while landing close to Etched's $700 million round.

The change also appears among younger companies. Fractile raised $220 million in a Series B. MatX raised $500 million in a Series B. EUCLYD passed €200 million in a Series A. Neurophos raised $110 million at Series A.

Most AI chip startups still cannot raise hundreds of millions. The top ten alone hold more than half of all capital in the dataset.

What changed is the ceiling. Investors have shown they will write software-scale mega-rounds into semiconductor companies when they think the startup could capture a meaningful piece of AI inference infrastructure.

Why are investors putting so much money into AI inference instead of only AI training?

Investors are increasingly financing inference because running models is becoming a massive recurring compute problem rather than a secondary workload that happens after training.

Training a model is expensive but episodic. A successful model may then answer requests, run agents, generate video, process enterprise data or execute code continuously. Every additional user and every longer reasoning chain creates more inference demand. Since inference demand never switches off, chip startups have something to aim at, and we look at where the biggest opportunities are in our report covering AI chips.

Recent financing announcements repeatedly point to that economics. SambaNova has centered its pitch around lower-cost enterprise inference. Groq is building a globally distributed inference cloud. Positron is targeting memory and power constraints. Etched is designing hardware specifically around transformer workloads. d-Matrix is optimizing memory movement for inference.

The architectures differ, but they all attack the same basic pressure: inference has to become faster, cheaper and less power-hungry if AI usage keeps growing.

That is also why investors can simultaneously finance very different technologies. Nobody yet knows whether the eventual winners will use wafer-scale compute, LPUs, transformer ASICs, in-memory processing, photonics or some combination of them.

Is edge AI still attractive when datacenter inference gets most of the huge rounds?

Yes. Edge AI remains a real investment category, but its economics produce smaller funding totals and a more fragmented competitive field.

SiMa.ai has raised $355 million, while Hailo reached roughly $341 million before being acquired. Axelera has also secured substantial capital and says it has shipped to hundreds of customers. Companies including DEEPX, Quadric, BOS Semiconductors, Mobilint, MemryX and Expedera add another layer of financed competitors.

What changes at the edge is the job the chip must do. Datacenter startups can optimize around giant language models and expensive racks. Edge chips may have to fit inside cars, cameras, industrial machines, robots or low-power embedded systems, where power, latency, price and physical size matter differently.

That leaves more room for specialization. An automotive inference processor does not need to beat Cerebras on frontier-model tokens per second, and an industrial vision accelerator does not need to recreate Groq's global cloud.

The smaller round sizes should not be read as weak demand. They partly reflect businesses where individual deployments require far less infrastructure capital.

Why are memory and interconnect startups showing up in an AI inference chip ranking?

Memory movement and chip-to-chip communication now affect AI inference economics so much that the boundary around an “AI chip company” has become harder to draw cleanly.

Ayar Labs is the clearest example. It does not build a conventional inference accelerator, yet investors have financed it at a scale comparable to major AI processor startups because optical interconnects can determine how efficiently thousands of accelerators work together.

Positron takes another route by emphasizing a memory-centric inference architecture. d-Matrix moves computation closer to memory through digital in-memory compute. Several analog compute-in-memory companies in the dataset are also trying to reduce the energy spent moving data between conventional memory and processing units.

We therefore use a market-centered definition rather than limiting the ranking to direct GPU replacements. If a company's core technology directly changes the economics of AI inference compute, it can belong in the wider competitive landscape even when its product is not a standalone accelerator.

The distinction still matters. Ayar Labs has not raised close to $1 billion to build “another GPU.” Its role in AI infrastructure is materially different.

Which AI inference chip startups could raise another major round or go public next?

Rebellions currently looks like the clearest near-term IPO candidate in this group because its latest $400 million financing was explicitly described as a pre-IPO round.

The company has already moved through the funding scale that often precedes a public offering: approximately $850 million raised, a multibillion-dollar valuation, production hardware and international expansion.

Ayar Labs is another company to watch. Its 2026 financing has pushed capital well beyond the level normally associated with a small semiconductor startup, while the company is moving from technology development toward high-volume manufacturing.

FuriosaAI has also publicly been associated with plans for a much larger financing ahead of a possible IPO. Its position is earlier than Rebellions financially, but commercial shipments and large strategic customers could change its funding profile quickly.

Etched and Positron are harder to call. Both are extremely well funded, but private investors are currently willing to provide them with huge amounts of capital, which reduces the immediate pressure to list.

So which AI inference chip startups are actually best funded right now?

Cerebras, SambaNova, Groq, Etched and Positron AI are the five companies that currently define the top of the AI inference chip funding market.

Our dataset contains 85 companies and roughly $26 billion of cumulative financing, yet about 39% of that money belongs to the first five names. The top ten hold around 56%, and the top twenty roughly 74%.

The hierarchy can still change surprisingly fast. Etched and Positron moved into the top five through enormous recent rounds. SambaNova closed much of its gap with Cerebras through a single $1 billion financing. Rebellions raised more than three quarters of its cumulative capital within about half a year. Totals that jump after one big round are why we look at who has raised the most, and from whom, in our AI chips report.

Funding has narrowed the field of companies financially capable of competing at global scale. It has not told us which architecture wins. From here, production volumes, deployed systems, customer spending and inference economics matter more than simply announcing the biggest venture round.