What are the top semiconductor startups by revenue today?

The ranking of top startups in the semiconductor industry 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 covering the Semiconductor Industry.

What actually counts as a semiconductor startup today?

The most useful definition today is a semiconductor company from the recent startup generation, even if commercial success has already taken it public.

A strict private-only definition creates a strange ranking. Astera Labs, Credo, Cambricon, Horizon Robotics, Ambiq and several Chinese GPU companies would disappear simply because they reached the public markets. Yet those listings give us much better financial data and tell us which startups actually became large businesses.

Age matters more here than legal status. CXMT and YMTC were created in the past decade. Cambricon dates from 2016, Astera Labs from 2017, Cerebras from 2016, Biren from 2019 and Moore Threads from 2020. We treat companies such as these as part of the recent semiconductor startup generation while excluding long-established groups such as Intel, Qualcomm, Texas Instruments and NVIDIA.

This also solves another problem: some of the largest private chip companies disclose almost no revenue. Restricting the ranking to private businesses would reward opacity and remove companies precisely when their financial performance becomes measurable.

Which semiconductor startups make the most revenue today?

CXMT is currently in a league of its own, followed by YMTC; among younger fabless companies, Credo, Cambricon and Astera Labs form the next major revenue group.

CXMT reported RMB150.3 billion, or about $22.4 billion, of revenue in the first half of 2026 alone. YMTC generated RMB63.2 billion in 2025, roughly $8.8 billion at a representative exchange rate.

There is then a sharp drop. Credo generated $1.335 billion in its latest fiscal year. Cambricon reported RMB6.5 billion, around $900 million, for 2025. Astera Labs reached $852.5 million.

Horizon Robotics recorded RMB3.76 billion, roughly $520 million, while Cerebras generated $510 million. Below them, Moore Threads, InnoScience, Biren Technology and several other Chinese chip companies have already reached significant nine-figure dollar scale.

That hierarchy looks very different from the semiconductor companies that dominate social media or fundraising headlines.

Company	Best recent revenue evidence	Period	Main business
CXMT	RMB150.3B (~$22.4B)	H1 2026	DRAM
YMTC	RMB63.2B (~$8.8B)	2025	NAND flash
Credo	$1.34B	FY2026	Data-center connectivity
Cambricon	RMB6.50B (~$900M)	2025	AI accelerators
Astera Labs	$853M	2025	Data-center connectivity
Horizon Robotics	RMB3.76B (~$520M)	2025	Automotive computing
Cerebras	$510M	2025	AI compute
Moore Threads	RMB1.74B in H1	H1 2026	GPUs
InnoScience	RMB1.21B (~$170M)	2025	GaN power
Biren Technology	RMB1.03B (~$150M)	2025	AI GPUs
Black Sesame	~RMB822M	2025	Automotive AI chips
Ambiq	$72.5M	2025	Ultra-low-power MCUs

Is CXMT really the biggest semiconductor startup by revenue?

Yes. CXMT's current revenue is so large that the word startup barely describes the business anymore.

Its first-half revenue reached RMB150.3 billion, about $22.4 billion, according to its first financial results after listing. That was up almost ninefold year over year and already exceeded twice its revenue for the whole previous year.

We do not double the six-month number to create a synthetic annual figure. There is no need. CXMT's reported half-year revenue alone is several times YMTC's annual sales and more than ten times Credo's latest full-year revenue.

The huge jump also needs context. Memory pricing has risen sharply during the AI infrastructure build-out, and CXMT benefited from both higher prices and stronger demand. So the latest number represents an unusually strong part of the memory cycle rather than a normal startup growth curve.

Still, the commercial scale is real. CXMT has moved far beyond the stage where funding, factory capacity or customer announcements are useful substitutes for revenue.

How big is YMTC now?

YMTC is currently an $8–9 billion annual-revenue business, which puts it far ahead of almost every younger AI-chip company.

The NAND manufacturer generated RMB63.2 billion in 2025, according to financial information disclosed around its listing preparations. That revenue increased by roughly 40%, while profit reached RMB14.2 billion.

YMTC's position becomes clearer when we compare orders of magnitude. Its annual revenue is around six times Credo's latest fiscal-year sales and roughly ten times Astera Labs' 2025 revenue.

The company has also become a meaningful NAND supplier rather than a small domestic challenger. Recent reporting based on its listing disclosures placed its share of global NAND shipments in the low teens.

Although YMTC is young by semiconductor-industry standards, its commercial profile now looks much closer to an established memory manufacturer than to a conventional venture-backed chip startup.

Has Credo become a billion-dollar semiconductor company?

Yes. Credo has crossed $1 billion in annual revenue and is currently one of the largest younger fabless semiconductor companies we can measure cleanly.

Credo generated $1.335 billion in its latest fiscal year, up 206% from $436.8 million one year earlier. Two years before that, revenue was only $193 million.

Few semiconductor scale-ups have moved through those revenue brackets so quickly.

Credo's business also tells us where AI spending is spreading. The company sells high-speed connectivity products used inside large data-center systems. As AI clusters pack in more accelerators, moving data between processors, memory and networking equipment becomes a bigger technical problem and a bigger source of semiconductor spending.

That has produced a large business without Credo needing to build a general-purpose GPU. Some of the fastest-growing semiconductor companies are selling the plumbing around AI accelerators rather than the accelerators themselves.

Is Cambricon now bigger than Astera Labs?

Yes, but only slightly on the latest full-year revenue figures.

Cambricon reported RMB6.5 billion of 2025 revenue, around $900 million, while Astera Labs generated $852.5 million.

The more striking number is Cambricon's growth. Revenue increased by more than five times from RMB1.17 billion the previous year, and the company reached its first annual profit. More recently, Cambricon reported RMB6.0 billion of revenue in the first half of 2026 alone, showing that the 2025 jump was followed by another large commercial step-up.

Astera Labs has been moving almost as fast. Revenue rose from $396.3 million in 2024 to $852.5 million in 2025, a 115% increase.

This comparison could move again quickly. For now, Cambricon has the larger reported annual figure, while both companies clearly sit in the same broad revenue class.

How big is Astera Labs today?

Astera Labs is already an $850 million annual-revenue semiconductor company.

Revenue climbed from $115.8 million in 2023 to $396.3 million in 2024 and then to $852.5 million in 2025. The latest year alone added more revenue than the company generated during all of 2024.

Astera sells connectivity silicon and related products for rack-scale AI infrastructure. That sounds less glamorous than building a GPU, but the economics have been excellent. Larger AI clusters require more links between accelerators, CPUs, memory and other components, and Astera has become one of the companies monetizing that complexity.

Profitability has arrived as well. The company reported $173.4 million of GAAP operating income for 2025 alongside a gross margin above 75%.

Astera is one of the clearest examples of a semiconductor startup becoming large through AI infrastructure without trying to replace NVIDIA directly.

Is Cerebras already a large semiconductor company?

Yes. Cerebras generated $510 million in 2025, although calling the whole business semiconductor revenue would now be too simplistic.

Its revenue increased from $78.7 million in 2023 to $290.3 million in 2024 and $510 million in 2025. Hardware accounted for $358.4 million of the latest annual figure, while cloud and other services generated another $151.6 million.

That mix is changing quickly. Cerebras increasingly sells access to computing infrastructure alongside its wafer-scale systems, so its economics now span hardware and cloud services.

This gives Cerebras a different revenue profile from a pure fabless chip designer such as Astera Labs. The total revenue is still useful for measuring company scale, but readers should not interpret the entire $510 million as semiconductor sales.

Cerebras has crossed the half-billion-dollar threshold and is commercially much larger than the typical private AI-chip startup.

How big is Horizon Robotics now?

Horizon Robotics generated RMB3.76 billion of revenue in 2025, making it one of the largest younger automotive-computing companies in the ranking.

The figure grew 57.7% from RMB2.38 billion the previous year. Horizon's own investor materials now highlight roughly RMB3.8 billion of 2025 revenue alongside more than 15 million cumulative Journey-series processing units.

The revenue mix deserves attention. Horizon earns money from automotive product solutions as well as licensing and services, so the full company figure cannot be treated as pure chip sales.

Still, the commercial footprint is large. Journey hardware has moved into hundreds of vehicle models, and Horizon says its systems now serve millions of car owners.

The comparison with pure AI-chip startups is imperfect, but Horizon's overall commercial scale clearly belongs in the upper part of this ranking.

How much revenue does Moore Threads make now?

Moore Threads is already generating billions of renminbi from GPUs, and its latest growth has been very fast.

The company generated about RMB1.51 billion in 2025. More recent disclosures then put first-half 2026 revenue around RMB1.74 billion, roughly 147% higher than the same period a year earlier.

We keep that half-year figure as a half-year figure rather than doubling it.

For a company founded in 2020, the ramp is notable. Moore Threads has moved from building and demonstrating domestic GPUs to selling them at meaningful scale in only a few years.

Cambricon remains significantly larger, but Moore Threads now sits well above the level where it makes sense to describe the business mainly through fundraising or product announcements.

Are Biren Technology and the other Chinese GPU startups making real revenue?

Yes. Biren and several other Chinese GPU companies have crossed into genuine commercial scale, although they remain much smaller than Cambricon.

Biren reported RMB1.035 billion of 2025 revenue, up 207%. That puts annual sales around $150 million. The company also reported RMB557 million of gross profit but remained loss-making after heavy research and development spending.

Moore Threads is already larger, while Cambricon has moved into a different bracket entirely. Other companies in the Chinese GPU cohort are also starting to disclose meaningful sales as public-market filings become available.

This gives us a better picture than the fundraising headlines from a few years ago. The Chinese accelerator market now contains several companies with actual nine-figure dollar revenue, rather than a long list of similarly valued pre-commercial startups.

The remaining gap is profitability. Developing new accelerator architectures and their software ecosystems still consumes enormous amounts of capital.

Is InnoScience turning GaN chips into a real business?

Yes. InnoScience reached RMB1.21 billion of 2025 revenue, while its manufacturing economics improved sharply.

Revenue grew 46.4%. More importantly, gross margin moved from negative 19.5% to positive 7.3%, and adjusted EBITDA turned positive.

That combination tells us more than revenue growth alone. A semiconductor manufacturer can increase sales aggressively while losing money on every additional unit. InnoScience's latest figures show that higher volume is starting to improve the underlying business as well.

This makes InnoScience one of the more interesting power-semiconductor companies in the younger cohort. GaN has been discussed for years as a future technology for power conversion, chargers, data centers and vehicles. We now have evidence of a specialist generating roughly $170 million of annual sales from that market.

How big is Black Sesame's automotive chip business?

Black Sesame is now around the $100 million-plus annual-revenue level, clearly commercial but still much smaller than Horizon Robotics.

Its latest annual results put company revenue around RMB822 million. The business centers on automotive computing chips and platforms, with newer work extending into robotics and embodied-AI applications.

Horizon generated more than four times as much company revenue during the same broad period.

That gap is useful because both companies often appear in the same conversation about Chinese autonomous-driving silicon. They have reached very different levels of commercial scale.

Black Sesame has enough sales to belong in a serious revenue ranking, but Horizon is currently the much larger business.

Is Ambiq still a small semiconductor startup?

Yes by semiconductor standards. Ambiq generated $72.5 million of sales in 2025, which puts it well below the leading startup-origin chip companies.

Revenue was actually slightly lower than the $76.1 million reported in 2024, although sales improved sequentially through 2025 and reached $20.7 million in the fourth quarter.

The company still has a real commercial business. Ambiq sells ultra-low-power semiconductor products used in wearables, health devices, industrial products and smart-home applications, and it says more than 80% of units shipped in 2025 ran AI algorithms.

The revenue comparison is nevertheless stark. Credo generated more than eighteen times Ambiq's annual sales. Astera generated nearly twelve times as much.

Revenue is useful in this market for exactly that reason: companies that can all plausibly be called semiconductor scale-ups may sit several orders of magnitude apart commercially.

Why isn't Groq near the top of this semiconductor startup ranking?

Groq does not have a clean enough current revenue disclosure for us to rank it alongside companies publishing audited or regulatory financials.

That is a meaningful omission because Groq is one of the best-known AI-inference companies. Its visibility, funding and technical reputation can easily create an assumption that revenue must also rank near the top.

Public evidence does not let us make that jump confidently.

Groq's corporate perimeter has also changed after the transaction involving NVIDIA and parts of Groq's technology and team. Older revenue estimates may describe a business that is no longer directly comparable with the company operating now.

So we keep Groq out of the numerical leaderboard rather than fill the gap with a weak estimate. The company may be commercially significant, but a precise ranking would imply more certainty than the public evidence supports.

Where do Tenstorrent, SiFive and other famous private chip startups rank?

Tenstorrent, SiFive and several other well-known private semiconductor companies remain difficult to rank because current company-wide revenue is rarely disclosed.

That does not make them small. It simply gives us weaker evidence.

Tenstorrent has raised large amounts of capital and built processors around its own architecture and RISC-V technologies. SiFive has become an important commercial company in the RISC-V IP ecosystem. Other private startups can point to customer wins, production programs, bookings or partnerships.

None of those automatically equals recognized annual revenue.

Cerebras shows the difference. Once a company provides filed financial statements, we can see exactly how much revenue came from hardware and how much came from cloud services. For a private company without comparable disclosure, trying to reconstruct revenue from customers, funding or pricing would make the ranking look more precise while making it less trustworthy.

For now, we treat these companies as important but unranked where the evidence is not strong enough.

Are Chinese semiconductor startups dominating the revenue ranking?

Yes. Chinese companies currently occupy most of the largest revenue positions in the recent semiconductor generation, especially in memory and AI computing.

CXMT and YMTC create much of that lead because memory manufacturing can generate huge revenue once production reaches scale. Together, their reported sales dwarf the younger U.S. fabless companies in this ranking.

China also has more depth below them. Cambricon, Horizon Robotics, Moore Threads, Biren, InnoScience and Black Sesame all have substantial disclosed revenue.

The U.S. cohort is particularly strong in AI infrastructure. Credo and Astera have built very large connectivity businesses, while Cerebras has crossed $500 million by combining hardware with cloud services.

So the difference is partly geographic and partly structural. China's largest recent semiconductor companies include capital-intensive memory manufacturers, while the strongest younger U.S. companies here are mostly fabless infrastructure specialists.

Company	Geography	Recent revenue scale	Category
CXMT	China	~$22.4B in H1	DRAM
YMTC	China	~$8.8B annual	NAND
Credo	U.S.-origin	$1.34B annual	Connectivity
Cambricon	China	~$900M annual	AI accelerators
Astera Labs	U.S.	$853M annual	Connectivity
Horizon Robotics	China	~$520M annual	Automotive compute
Cerebras	U.S.	$510M annual	AI compute
Moore Threads	China	~$259M in H1	GPUs
InnoScience	China	~$170M annual	GaN power
Biren	China	~$150M annual	GPUs
Black Sesame	China	~$100M+ annual	Automotive compute
Ambiq	U.S.	$72.5M annual	Edge MCUs

Which type of semiconductor startup makes the most money?

Memory currently produces the biggest revenue numbers, while AI-connectivity companies have created some of the fastest-growing fabless businesses.

CXMT and YMTC sit at the top because memory is a huge-volume market. Once a manufacturer reaches meaningful global production, annual sales can move into billions of dollars.

Credo and Astera use a very different model. Both benefit from the increasingly difficult job of moving data around large AI systems. Their latest results show that connectivity can become a billion-dollar semiconductor category surprisingly quickly.

AI accelerators form a more uneven group. Cambricon is already approaching a billion dollars of annual revenue, Cerebras has crossed $500 million, and several younger Chinese GPU companies are in the $100 million-plus range.

Power semiconductors and automotive compute are producing meaningful businesses too, but currently at lower revenue levels.

Semiconductor model	Examples	What revenue looks like now
Memory	CXMT, YMTC	Multi-billion-dollar scale
AI connectivity	Credo, Astera Labs	~$850M–$1.3B+
AI accelerators	Cambricon, Cerebras	~$500M–$900M leaders
Chinese GPUs	Moore Threads, Biren	~$150M–$260M disclosed scale
Automotive computing	Horizon, Black Sesame	~$100M–$500M+
GaN power	InnoScience	~$170M
Ultra-low-power MCU	Ambiq	~$70M

Are AI-chip startups actually the biggest semiconductor startups?

No. The current revenue leaders come from memory, while connectivity companies also outperform most accelerator startups.

AI accelerators dominate a lot of the conversation because the potential market is enormous and NVIDIA has become the industry's central company. That attention can blur the difference between future opportunity and current sales.

CXMT's half-year revenue alone is many times larger than Cambricon's latest annual revenue. YMTC also remains several times larger than any younger U.S. AI-infrastructure company in this analysis.

Even within the AI boom, some of the best businesses sit beside the GPU. Credo and Astera are monetizing the data movement required by larger accelerator clusters, and both have reached impressive scale.

The revenue ranking tells a different story from the attention ranking: memory comes first, AI connectivity has become surprisingly large, and accelerator startups occupy a broad range from tens of millions to nearly a billion dollars.

Are the fastest-growing semiconductor startups profitable yet?

Some are, but high revenue growth still does not guarantee an economically mature semiconductor business.

Credo provides one of the clearest examples of operating leverage. Its annual revenue jumped to $1.335 billion, while GAAP operating income reached roughly $445 million.

Astera Labs also produced $173.4 million of GAAP operating income on $852.5 million of 2025 revenue.

Cambricon's change has been even more dramatic. After years of losses, the company reported its first annual profit as revenue surged to RMB6.5 billion.

Other businesses are much earlier in that transition. Biren generated more than RMB1 billion of revenue while continuing to spend heavily on next-generation products. Ambiq lost $36.5 million on $72.5 million of sales. InnoScience is somewhere in between: its gross margin and adjusted EBITDA recently turned positive, suggesting that scaling production is improving the economics.

Cerebras needs extra care because accounting items created a large GAAP net income figure while its non-GAAP result remained negative. The operating picture is less flattering than the headline profit number suggests.

Company	Recent revenue	Current profitability signal
Credo	$1.34B	Strong GAAP operating profit
Cambricon	RMB6.50B	First annual profit
Astera Labs	$853M	Positive GAAP operating income
Cerebras	$510M	Non-GAAP loss despite GAAP net income
InnoScience	RMB1.21B	Gross margin and adjusted EBITDA turned positive
Biren	RMB1.03B	Still loss-making after heavy R&D
Ambiq	$72.5M	$36.5M net loss

How much can we trust semiconductor startup revenue numbers?

We can trust the top of this ranking much more than the private-company tail.

Credo, Astera Labs, Cerebras and Ambiq publish detailed U.S. filings. Horizon, Biren, InnoScience and Black Sesame disclose financials through Hong Kong's reporting system. Cambricon and Moore Threads now provide public-market disclosures in China. These figures are far stronger than estimates based on employees, web traffic or fundraising.

The tricky cases are still-private companies and businesses with unusual revenue mixes. Horizon includes licensing and services. Cerebras combines hardware and cloud revenue. Half-year figures such as Moore Threads' latest sales are fresher than older annual revenue but cover a shorter period.

We preserve those differences rather than forcing every number into the same shape.

That approach sometimes leaves famous companies without a numerical position. It also avoids giving a weak estimate the same visual authority as an audited annual revenue figure.

Evidence	Typical example	How we use it
Audited or filed annual revenue	Credo, Astera, Ambiq	Strongest evidence
Regulatory financial disclosure	Cambricon, Horizon, Biren	Strong evidence
Filed interim revenue	Moore Threads, CXMT	Strong but shorter period
Mixed hardware/services revenue	Horizon, Cerebras	Reliable total, less directly comparable
Direct company disclosure	Useful when filings are unavailable	Usually strong
Tier-1 reported financial figure	Used selectively	Moderate to strong
Third-party revenue estimate	Private companies	Weak evidence
Funding, valuation or users converted into revenue	Any company	Excluded

What does the semiconductor startup revenue ranking actually tell us?

The clearest takeaway today is that a substantial middle generation of semiconductor companies has already grown far beyond the classic startup stage.

CXMT and YMTC have reached multi-billion-dollar memory scale. Credo is above $1.3 billion. Cambricon and Astera Labs are around the billion-dollar threshold. Horizon Robotics and Cerebras are around half a billion dollars. Several newer Chinese GPU companies are already generating more than $100 million.

That makes the usual picture of semiconductors — giant incumbents on one side and speculative startups on the other — increasingly outdated.

The ranking also shows where commercial success is happening. Memory remains enormous. AI connectivity has become a major category. Automotive computing has produced real revenue. GaN is moving beyond pilot adoption. AI accelerators are scaling, although the gap between the leaders and the rest remains wide.

And visibility is a poor shortcut for revenue. Several of the most talked-about private AI-chip companies cannot yet be ranked with confidence, while companies such as Credo and Astera have quietly built businesses approaching or exceeding $1 billion.

For now, CXMT is the clear revenue outlier among the recent semiconductor generation. YMTC follows at multi-billion-dollar scale, while Credo, Cambricon and Astera Labs form the most important next cluster with clean, substantial revenue evidence.