What are the top Foundation Model startups by total funding raised?

Top startups in the Foundation Model Market ranked by total funding raised

Here is an updated table that ranks the top startups in the Foundation Model 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 covering the Foundation Model Market.

Which foundation model startups have raised the most funding?

OpenAI and Anthropic are in a completely different funding league from every other foundation model startup, with xAI a distant third and everyone else far behind.

Our reconstructed funding history puts OpenAI at roughly $181.9 billion in cumulative funding and Anthropic at about $123.3 billion. xAI follows at approximately $37.8 billion before its acquisition by SpaceX. Those three alone account for about $343 billion.

The gap immediately after them is enormous. Z.ai, also known as Zhipu AI, sits around the $8 billion level depending on how its public-market financing is treated. Safe Superintelligence is also around $8 billion if we include the approximately $5 billion NVIDIA investment reported during its latest strategic deal. DeepSeek follows at roughly $7.4 billion after completing its first major external financing.

That funding distribution tells us more than the ranking itself. Investors are backing dozens of foundation-model companies, but truly extraordinary amounts of capital have gone into a very small number of labs capable of competing at the frontier.

Company	Approx. cumulative funding	Latest major financing	Current situation
OpenAI	$181.9B	$122B	Private
Anthropic	$123.3B	$65B	Private
xAI	$37.8B	$20B	Acquired by SpaceX
Z.ai / Zhipu AI	~$8.2B	Major public-market financing	Public
Safe Superintelligence	~$8.0B	Reported ~$5B strategic investment	Private
DeepSeek	$7.4B	$7.4B	Private
Mistral AI	~$6.5B	€3B Series D	Private
Reflection AI	~$4.6B	$2.5B Series C	Private
MiniMax	~$3.8B	~$2B PIPE	Public
StepFun	$3.3B+	~$2.5B growth financing	Private

How far ahead are OpenAI and Anthropic in foundation model funding?

OpenAI and Anthropic have pulled so far ahead in foundation model funding that comparing them with ordinary venture-backed AI startups has become increasingly misleading.

OpenAI's latest round alone brought in $122 billion. Anthropic then raised $65 billion in its Series H. Either financing by itself is larger than the entire cumulative funding history of almost every other independent model developer in the world.

Within the ten largest companies in our ranking, OpenAI, Anthropic and xAI represent roughly $343 billion out of about $385 billion raised. In other words, close to nine dollars out of every ten represented by the top ten went to just three companies.

The market has an enormous frontier-compute tier at the top, followed by a second group operating at a completely different financial scale. It barely resembles a normal venture category anymore.

How much funding has OpenAI raised?

OpenAI has raised roughly $182 billion based on our reconstruction, making it the best-funded foundation model company by a wide margin.

The biggest change came with OpenAI's $122 billion financing, which valued the company at $852 billion post-money. OpenAI said the round included major commitments from Amazon, NVIDIA and SoftBank alongside institutional investors.

That one financing was several times larger than xAI's entire cumulative funding history and more than twenty times the total raised by Mistral AI.

OpenAI has explained the spending logic quite openly: more training and inference capacity requires enormous commitments to cloud providers, chips and data centers. Its infrastructure strategy now spans Microsoft, AWS, Oracle, Google Cloud and CoreWeave on the cloud side, plus NVIDIA, AMD, Trainium, Cerebras and custom silicon.

So the funding figure increasingly reflects more than model ambition. OpenAI is financing something that looks closer to a global computing platform.

Has Anthropic almost caught OpenAI in funding?

Anthropic has come remarkably close to OpenAI in cumulative foundation model funding, although OpenAI still leads by roughly $59 billion in our reconstruction.

Anthropic's $65 billion Series H changed the comparison dramatically. The company said that round valued it at $965 billion post-money and came only a few months after its previous financing.

Anthropic also disclosed that its annualized revenue had crossed $47 billion around the time of the round. Investors were no longer backing only a research trajectory. They were funding a company already producing revenue on a scale rarely seen in private technology markets.

The company said the money would go toward additional compute, Claude products, partnerships, safety work and interpretability research. That helps explain why Anthropic can attract capital approaching OpenAI's scale: it combines an expensive frontier-model program with a rapidly growing commercial platform.

Is xAI still one of the world's biggest foundation model startups?

xAI remains one of the largest foundation-model fundraising stories ever, but calling xAI an independent startup today is no longer accurate because SpaceX acquired the company.

Before the transaction, xAI raised $20 billion in its Series E, bringing our reconstructed cumulative figure to roughly $37.8 billion. Investors in that round included Valor Equity Partners, Fidelity, Qatar Investment Authority and MGX, while NVIDIA and Cisco participated strategically.

That Series E was largely aimed at infrastructure. xAI said it had ended the previous year with more than one million H100-equivalent GPUs across its Colossus computing systems.

SpaceX subsequently acquired xAI in a transaction between Elon Musk-controlled companies. SpaceX's own prospectus confirms that the merger became effective in early 2026.

We still include xAI's historical funding when looking at how much capital went into building major foundation-model competitors. For current-company comparisons, though, xAI now belongs inside a much larger SpaceX group.

Which foundation model companies come after OpenAI, Anthropic and xAI?

There is a real second tier of foundation model companies, but the largest members of that group have raised roughly $3 billion to $8 billion rather than tens or hundreds of billions.

Z.ai, Safe Superintelligence, DeepSeek and Mistral AI currently form the strongest cluster behind the three giants. Reflection AI has also moved rapidly upward after two enormous rounds.

The group is more varied than the numbers suggest. DeepSeek emphasizes efficient open-weight reasoning models. Mistral combines open models with enterprise and sovereign AI. SSI is pursuing superintelligence research without the usual product cycle. Reflection is building open frontier intelligence.

There is plenty of capital behind alternatives to OpenAI and Anthropic, but investors are spreading it across several competing philosophies rather than backing one obvious fourth global leader.

Company	Approx. funding	Main focus
Z.ai / Zhipu AI	~$8.2B	Chinese GLM foundation models
Safe Superintelligence	~$8.0B	Superintelligence research
DeepSeek	$7.4B	Open-weight reasoning models
Mistral AI	~$6.5B	Open-weight and enterprise models
Reflection AI	~$4.6B	Open frontier models
MiniMax	~$3.8B	Multimodal models
StepFun	$3.3B+	Multimodal models and agents
Moonshot AI	~$3.0B	Kimi language and reasoning models
Kling AI	~$3.0B	Generative video models
Physical Intelligence	~$2.1B	Robotic foundation models

Which Chinese foundation model companies have raised the most money?

DeepSeek, Z.ai, MiniMax, StepFun and Moonshot AI show that China's foundation-model ecosystem has moved well beyond small venture rounds and into multi-billion-dollar financing.

DeepSeek produced the biggest private financing shock. The Information reported that the company closed its first external round at more than RMB50 billion, or roughly $7.4 billion. The structure was unusual, with much of the capital reportedly going through a limited partnership controlled by founder Liang Wenfeng.

Z.ai has moved into public markets, which makes its cumulative number harder to compare directly with private labs. A recent transaction advised by Linklaters combined a new H-share placement with convertible bonds for the equivalent of $5.02 billion.

MiniMax is already listed in Hong Kong as well. StepFun, meanwhile, was reported by Chinese financial outlet Yicai to be completing a financing of nearly $2.5 billion as it prepared for a potential Hong Kong listing. Moonshot AI has continued raising from investors including Alibaba, Tencent and 5Y Capital.

China now has several heavily financed model developers. What it still lacks is one private company funded on anything close to OpenAI or Anthropic's scale.

Is Mistral AI really Europe's best-funded foundation model challenger?

Mistral AI is comfortably Europe's strongest independently funded frontier-model company and has now reached a scale that makes the comparison with U.S. labs more serious.

Mistral recently announced a €3 billion Series D at a post-money valuation above €21 billion. Samsung led the round, while EQT's Scaleup Europe Fund and PSG Equity co-led.

The financing takes Mistral's cumulative funding into roughly the $6.5 billion range in our reconstruction. That still leaves a huge gap with OpenAI and Anthropic, but it puts Mistral above most specialized American foundation-model companies.

Mistral is also building around a distinct commercial position. Its pitch combines open-weight models, enterprise deployments and sovereign AI infrastructure, with more than 125 large enterprise customers cited by the company.

Europe does have a well-capitalized foundation-model champion. The financial difference with the two U.S. leaders is still measured in tens of billions of dollars.

Why are foundation model startups raising such enormous rounds?

Foundation model startups are raising enormous rounds because frontier AI has become extraordinarily compute-intensive, and the largest labs increasingly need infrastructure budgets that look more like hyperscaler budgets than traditional startup budgets.

OpenAI explicitly ties its capital requirements to compute. xAI used its Series E to expand Colossus. Anthropic says additional funding is going into compute capacity. Mistral's latest financing also includes scaling the infrastructure required for model training.

The same pattern appears lower in the ranking. Reflection has signed multi-billion-dollar compute agreements. Thinking Machines announced a partnership with NVIDIA built around at least one gigawatt of Vera Rubin systems. SSI said its NVIDIA partnership would increase available compute by an order of magnitude.

And that is before salaries, data, product development or international expansion. The biggest fundraising rounds increasingly function as financing for AI factories.

Who are the biggest investors in foundation model startups?

NVIDIA has become the most visible strategic investor across the foundation-model market, while Sequoia, Andreessen Horowitz, Fidelity, SoftBank, major sovereign funds and hyperscalers repeatedly appear around the largest companies.

NVIDIA has invested or participated in financings involving companies including OpenAI, xAI, SSI, Reflection, ElevenLabs, World Labs and Skild AI. It has also formed technical partnerships with several of them.

Sequoia appears across Anthropic, SSI, ElevenLabs and other major AI companies. Fidelity has backed several large model developers. SoftBank has played an important role in OpenAI and Skild AI. Amazon's exposure to Anthropic and OpenAI shows that hyperscalers are willing to back more than one major model platform.

The investor map is becoming increasingly interconnected. The same infrastructure companies, venture firms and institutional investors hold positions across several potential winners rather than choosing one model lab and hoping for the best.

Is NVIDIA becoming the kingmaker of the foundation model market?

NVIDIA now has influence across so many foundation-model companies that its role goes far beyond simply selling GPUs.

The clearest examples combine financing with infrastructure. NVIDIA participated in xAI's Series E while xAI expanded its giant GPU clusters. Its SSI partnership included both investment and access to Vera Rubin systems. NVIDIA also made a strategic investment in Thinking Machines alongside a gigawatt-scale compute partnership.

Poolside offers an even more unusual example. Bloomberg reported that NVIDIA agreed to invest another $1 billion in the company while separately paying $6 billion for technology licensing and bringing more than 100 employees into NVIDIA.

NVIDIA is in a rare position: it can profit whichever model company wins because many competing labs need NVIDIA hardware, NVIDIA capital, or both.

Are open-weight foundation model startups still attracting serious funding?

Open-weight foundation models are attracting billions of dollars, so investors clearly do not believe proprietary APIs will be the only viable model business.

Mistral is the biggest example, with cumulative financing around $6.5 billion. Reflection has raised roughly $4.6 billion while explicitly positioning itself around frontier open intelligence. Arcee AI recently raised a $150 million Series B and is building open American models.

NVIDIA has reinforced the trend through its Nemotron Coalition, which includes model builders such as Mistral, Reflection, Arcee, Sarvam and Thinking Machines.

The commercial bet is that open models can create value through enterprise deployment, sovereign infrastructure, customization and specialized services even when model weights themselves are available more broadly.

The funding totals are now too large to dismiss that thesis as a niche open-source movement.

Which specialized foundation model startups have raised the most money?

Specialized foundation models have become a major funding category of their own, particularly in robotics, video, audio, coding and world models.

Physical Intelligence and Skild AI have each raised close to or above $2 billion for robotic intelligence. Poolside has accumulated about $1.6 billion in our reconstruction around coding and model-development infrastructure. World Labs raised roughly $1.2 billion before agreeing to be acquired by AMD.

Creative AI has produced another group of heavily funded specialists. ElevenLabs has raised $781 million for audio models, Suno about $775 million for music, Runway roughly $860 million for video and world models, and Higgsfield more than $500 million for visual generation.

The foundation-model market is becoming much less synonymous with general-purpose chatbots. Investors increasingly treat narrower modalities as potential platform businesses in their own right.

Company	Approx. funding	Foundation-model focus
Physical Intelligence	$2.1B	Robotics
Skild AI	$1.9B	Robotics
Poolside	$1.6B	Software development
World Labs	$1.2B	Spatial intelligence
Luma AI	$1.1B	Video and world models
Runway	$860M	Video and world models
ElevenLabs	$781M	Voice and audio
Suno	$775M	Music
Generalist AI	$738M	Robotics
Higgsfield AI	$538M	Video and image generation

Are robot foundation models becoming a serious investment category?

Robot foundation models have clearly become a serious funding category, with multiple startups now raising hundreds of millions or more than $1 billion.

Skild AI's $1.4 billion Series C is the clearest example. SoftBank led the round, with NVIDIA's venture arm, Macquarie and Bezos Expeditions among the investors, giving Skild a valuation above $14 billion.

Physical Intelligence has also attracted roughly $2 billion across its financing history in our dataset. Generalist AI is around $738 million, while FieldAI is above $400 million. Smaller companies such as Flexion Robotics and RLWRLD are pursuing the same broad idea at earlier stages.

Investors are betting that the foundation-model approach used for language, images and video can also produce reusable intelligence for machines operating in the physical world.

The thesis is still less commercially proven than language models, but the capital committed to it is already far too large to treat robotics foundation models as an experimental side market.

Are world models becoming the next big foundation-model funding theme?

World models have become one of the fastest-moving parts of foundation-model funding, particularly because robotics, simulation and spatial intelligence need models that understand environments rather than only text.

World Labs raised $1 billion in new financing after initially launching with roughly $230 million. AMD subsequently agreed to acquire the company for approximately $8.2 billion, subject to closing conditions.

Runway is increasingly describing its work through world simulation rather than only video generation and raised another $315 million for that effort. Decart has accumulated more than $450 million for real-time generative worlds. Odyssey raised $310 million in a single Series B according to our dataset.

The boundary between video generation and world modeling is still fuzzy in places. Investor behavior is clearer: capital is moving toward models designed to represent persistent spaces, physical relationships and interactive environments.

Is generative video still attracting big funding rounds?

Generative video remains one of the best-funded specialized foundation-model categories, and recent rounds suggest investor interest has increased rather than faded.

Runway raised $315 million in Series E financing. Suno's neighboring creative-AI category attracted more than $400 million for music, while Higgsfield raised $400 million in Series B funding at a $5.4 billion valuation.

Our broader dataset also includes Luma AI at roughly $1.1 billion in cumulative funding, ShengShu Technology near $942 million, PixVerse above $580 million and several smaller video-model companies.

The funding pattern has changed too. Early video companies were often financed like creative tools. Today investors increasingly back them as model platforms capable of serving advertising, entertainment, simulation and eventually world-model applications.

Is voice AI becoming a major foundation-model category too?

Voice foundation models are now producing genuinely large companies, with ElevenLabs providing the clearest evidence that specialized audio AI can support both major funding rounds and substantial revenue.

ElevenLabs raised $500 million in its Series D, taking total funding to $781 million. The company said it ended the previous year above $330 million in annual recurring revenue and crossed $500 million only a few months later.

Its voice-agent product has also moved beyond experimental usage. ElevenLabs says its agents now handle millions of conversations each week across sales, support and operational workflows.

Other companies remain much smaller financially. Cartesia has raised roughly $191 million, Hume AI about $73 million, Fish Audio $52 million and WaveForms AI $40 million in our ranking.

Voice funding is less concentrated than frontier LLM funding, but ElevenLabs has already shown that a specialized foundation-model company can grow into a multi-billion-dollar business without competing head-on with ChatGPT.

What are the biggest foundation model funding rounds outside OpenAI, Anthropic and xAI?

Once OpenAI, Anthropic and xAI are removed, foundation-model startups are still producing rounds of a size that would have looked extraordinary for software companies only a few years ago.

DeepSeek's reported $7.4 billion first external financing sits near the top. SSI's latest NVIDIA investment was officially described only as significant, although Bloomberg reporting put it at around $5 billion. Mistral raised €3 billion, while Reflection closed a $2.5 billion Series C.

StepFun has been reported around $2.5 billion for its latest financing. Skild AI raised $1.4 billion. World Labs raised $1 billion, and Poolside reportedly received another $1 billion investment from NVIDIA.

The billion-dollar round has stopped being an anomaly in foundation models. Several independent labs now need that level of financing simply to remain credible participants in compute-heavy areas.

Company	Recent major round or investment	Area
DeepSeek	~$7.4B	General reasoning models
Safe Superintelligence	Reported ~$5B	Superintelligence
Mistral AI	€3B	Open and enterprise models
Reflection AI	$2.5B	Open frontier models
StepFun	~$2.5B	Multimodal models
Skild AI	$1.4B	Robotics
Physical Intelligence	~$1.1B	Robotics
World Labs	$1.0B	Spatial intelligence
Poolside	$1.0B reported investment	Coding models
ElevenLabs	$500M	Voice and audio

Can a foundation model startup really raise billions before proving a business?

Safe Superintelligence shows that elite AI teams can raise billions before proving a conventional commercial business, although very few founders could reproduce that result.

SSI initially raised $1 billion only months after being founded. A second financing added roughly $2 billion even though the company still had no public commercial product.

The latest NVIDIA partnership pushed the experiment further. NVIDIA officially confirmed making a substantial investment and providing access to Vera Rubin systems that SSI said could multiply its compute capacity. Reporting placed the investment at several billion dollars.

Thinking Machines offers another version of the same phenomenon. Mira Murati's company raised heavily before building a mature commercial business and later announced a major strategic compute partnership with NVIDIA.

In these cases, investors are financing people, research capability and access to scarce technical talent long before normal startup metrics can justify the valuation.

Does raising more money mean a foundation model company has better AI?

Funding is a useful measure of how much financial firepower a foundation-model company has, but it tells us surprisingly little about which company has the best model.

DeepSeek is the clearest warning against using fundraising as a technical leaderboard. Before its enormous external financing, the company became globally prominent through models that competed with much more heavily funded Western labs.

The reverse is also possible. SSI has raised billions despite releasing little publicly. Large financing there reflects investor conviction about Ilya Sutskever and future research rather than measurable current market leadership.

Commercial performance adds another layer. ElevenLabs has raised less than $1 billion yet has reported hundreds of millions of dollars in annualized revenue. Several research-heavy labs above it in the funding ranking have disclosed far less commercial evidence.

Funding measures capacity to hire, train models and buy compute. It should not be confused with a clean ranking of model quality or business strength.

Are foundation model startups becoming too expensive for normal venture capital?

The largest foundation-model companies have already outgrown traditional venture financing and now depend on sovereign capital, strategic corporations and giant institutional investors.

A $65 billion Anthropic round or $122 billion OpenAI financing cannot be funded by ordinary venture partnerships alone. These transactions need large asset managers, technology companies, sovereign investors and other institutions capable of writing multi-billion-dollar checks.

The same shift appears further down. Samsung led Mistral's latest round. SoftBank led Skild's Series C. NVIDIA is investing directly across the model ecosystem. Hong Kong's public markets are becoming a source of capital for Chinese AI companies.

Foundation-model financing increasingly resembles infrastructure finance mixed with late-stage private equity rather than classic startup venture capital.

That favors companies that already have access to major strategic partners and makes it progressively harder for new frontier labs to compete through normal VC fundraising alone.

Are foundation model IPOs changing the funding ranking?

Public-market financing is making foundation-model funding rankings harder to interpret because companies such as Z.ai and MiniMax can keep raising equity after their IPOs while private competitors rely on venture rounds.

MiniMax is already listed in Hong Kong and has subsequently raised additional public equity. Z.ai is also public and has completed major capital-market transactions following its listing.

For a cumulative-equity ranking, those transactions can still represent real equity capital raised by the company. They are nevertheless different from venture financing and should be identified clearly rather than blended invisibly with Series A, B or C rounds.

This distinction will become more important as additional foundation-model companies list publicly. A company that completes an IPO and several follow-on offerings could eventually rank above a private competitor even if private investors originally funded the competitor more aggressively.

Are acquisitions already reshaping the foundation model market?

Acquisitions are now removing important independent foundation-model companies from the startup landscape, with xAI and World Labs providing two very different examples.

SpaceX completed its acquisition of xAI, meaning one of the three best-funded AI labs no longer exists as an independent startup.

AMD has meanwhile signed a definitive agreement to acquire World Labs for approximately $8.2 billion in stock. The deal has not yet closed, so World Labs should currently be described as pending acquisition rather than already fully absorbed into AMD.

Both transactions involve strategic combinations with companies controlling enormous computing resources. SpaceX brings infrastructure, capital and distribution to xAI. AMD gains a leading spatial-intelligence research team that can help shape future hardware and software.

As model development becomes more expensive, acquisition by infrastructure providers may become a realistic alternative to raising another multi-billion-dollar private round.

Which foundation model startups look surprisingly lightly funded?

Some technically visible foundation-model companies remain surprisingly small by the standards of this market, showing that billions of dollars are not required to build something relevant.

Liquid AI has raised roughly $293 million while developing non-transformer model architectures. H has raised about $220 million around agentic computer-use models. Reka AI is around $168 million, and Stability AI is roughly $257 million in our reconstruction despite being one of the names that helped popularize open generative image models.

At the smaller end, Ideogram is below $100 million, Hume AI around $73 million and Deep Cogito about $56 million.

Those numbers would be large in many software categories. In foundation models they look almost modest.

The long tail still matters because model efficiency keeps improving. A smaller company does not need to reproduce OpenAI's infrastructure footprint if it can specialize aggressively, use open components or develop models that require less compute.

What types of foundation model startups attract the most funding today?

General-purpose frontier labs still attract the most capital by far, but robotics, world models, video, coding and voice have all become substantial funding markets of their own.

The generalist group dominates because OpenAI, Anthropic, xAI, DeepSeek, Mistral and other large labs need huge amounts of training and inference capacity.

Robotics forms the clearest specialist cluster, with Physical Intelligence, Skild, Generalist AI and FieldAI collectively raising billions. World-model and spatial-intelligence companies such as World Labs, Runway and Decart form another cluster.

Creative models have also matured financially. Runway, Luma, Higgsfield, Suno and ElevenLabs show that video, music and audio can support companies raising hundreds of millions of dollars each.

The market is branching rather than converging around one universal model. Investors increasingly seem willing to finance several different kinds of foundational intelligence.

So what does the foundation model funding ranking really tell us today?

The foundation model market is massively funded, brutally concentrated at the top and broadening quickly underneath, with OpenAI and Anthropic occupying a financial tier that no ordinary AI startup currently approaches.

The headline ranking can make the sector look like one giant race, but our research points to several races happening simultaneously. OpenAI and Anthropic are financing global AI platforms. DeepSeek, Mistral and Reflection are challenging the assumption that frontier models must remain closed. SSI and Thinking Machines represent huge bets on elite research teams. Skild and Physical Intelligence are taking the foundation-model idea into robotics. Runway, ElevenLabs, Suno and others are building modality-specific platforms.

The funding gap is still staggering. OpenAI and Anthropic together have raised more than $300 billion in our reconstruction, while most serious competitors remain below $10 billion.

Yet capital has not eliminated competition. DeepSeek showed how quickly technical perceptions can move. ElevenLabs demonstrated that a much smaller specialist can build a fast-growing commercial business. Mistral has assembled a credible European position without trying to replicate OpenAI dollar for dollar.

Our conclusion is clear: OpenAI and Anthropic dominate foundation-model fundraising today, but the market underneath them is becoming more diverse rather than consolidating around a single challenger. The next important winner may emerge from a company that raises another $50 billion, but it could just as easily come from a specialist that uses far less capital much more efficiently.