What are the latest funding trends in Foundation Model?

All funding deals in the Foundation Model Market over the last years

Below is the table listing all the deals. 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 Foundation Model Market.

Is Foundation Model funding really booming right now?

Foundation Model funding is booming today, and this time the increase goes well beyond a few giant rounds: we count 63 financings and 51 funded companies in the current YTD period, almost double last year's levels.

Deal count is up 91% from 33, while the number of companies receiving funding climbed 89% from 27. Compared with the same period two years ago, activity is also comfortably higher: 63 deals versus 43 and 51 companies versus 38.

The dollars have moved on another scale entirely. We calculate about $251.8 billion of usable current-period capital, after removing two unusually difficult-to-compare transactions, versus $74.8 billion last year and $12.7 billion two years ago. The median usable financing rose from $110 million to about $263 million.

Even without trying to forecast the rest of the year, the current period has already produced more deals than the 45 recorded during all of 2025 and the 55 recorded during all of 2024. The acceleration we see today is hard to explain away as timing.

Metric	2026 YTD	2025 YTD	2024 YTD	Change vs 2025
Deals	63	33	43	+90.9%
Unique funded companies	51	27	38	+88.9%
Usable capital	$251.8B	$74.8B	$12.7B	+236%
Median usable round	$263.3M	$110.0M	$75.0M	+139%
Average usable round	$4.20B	$2.27B	$352.9M	+85%
Largest round	$122B	$40B	$6B	+205%
Companies raising more than once	8	5	4	+60%
Full prior-year deal count	45 in 2025	-	55 in 2024	2026 YTD already higher

Would Foundation Model funding still look strong without the megadeals?

Foundation Model funding would still look better than last year without the megadeals, but the spectacular $251.8 billion headline would disappear very quickly.

Once we remove every financing above $100 million, current-period capital falls to roughly $945 million. That remains 44% above last year's $654 million, yet it is about 8% below the comparable $1.03 billion recorded two years ago.

The picture gets even less dramatic below $50 million. We find about $286 million today, against $201 million last year and $400 million two years ago.

The smaller end has recovered from a weak 2025, but it has not broken out compared with 2024. Almost all of the extraordinary dollar growth is happening higher up the financing stack.

OpenAI makes the distortion obvious. Its $122 billion round alone represents around 48.5% of usable current-period capital. Anthropic's $65 billion Series H then pushes the market further into territory where normal venture comparisons become less useful. The top three rounds together account for roughly 86% of capital.

Funding test	2026 YTD	2025 YTD	2024 YTD
Total usable capital	$251.8B	$74.8B	$12.7B
Capital with >$100M rounds removed	$944.8M	$654.1M	$1.03B
Capital with >$50M rounds removed	$285.8M	$201.3M	$400.1M
Top 1 share of capital	48.5%	53.4%	47.2%
Top 3 share	86.2%	81.9%	63.0%
Top 5 share	90.4%	89.3%	73.6%
$100M+ usable rounds	40	18	12
Median usable round	$263.3M	$110M	$75M

Are more Foundation Model startups getting funded, or are rounds just getting bigger?

More Foundation Model startups are getting funded and rounds are getting bigger at the same time, which is why this year's acceleration looks much more convincing than a simple megaround spike.

The number of funded companies jumped from 27 to 51 in one year. Deal count moved at almost the same rate, from 33 to 63, so the extra activity is not mainly coming from repeatedly financing the same labs.

Meanwhile, the typical financing has changed dramatically. A usable median of roughly $263 million is more than twice last year's $110 million and more than three times the $75 million median from two years ago.

Only eight companies in our current-period dataset raised more than once, up from five last year. That increase is real, but far too small to explain the additional 24 funded companies.

Foundation Model funding today is broader and much deeper at the same time. More companies are getting financed, while the strongest ones have access to sums that were rare even recently.

Are $100 million Foundation Model rounds becoming normal?

Nine-figure Foundation Model rounds have become remarkably common: we count about 40 usable financings of at least $100 million in the current period, versus 18 last year and 12 two years ago.

That is probably the cleanest structural change in the whole dataset because it does not depend on whether OpenAI raised $122 billion or $100 billion. The number of companies crossing nine figures has itself surged.

The change now shows up across very different businesses. OpenAI and Anthropic sit at the extreme frontier. Mistral raised €3 billion. ElevenLabs raised $500 million earlier this year. General Intuition raised $320 million and then another $220 million. Mirendil launched with a $200 million seed. Arcee's recent Series B was reported at at least $150 million. TwelveLabs raised $100 million.

Those companies are building different types of models for different customers, yet investors are repeatedly accepting capital requirements that would have looked very aggressive for an AI startup a few years ago.

These days, $100 million increasingly looks like a normal large round in Foundation Models rather than a once-in-a-cycle event.

Can new Foundation Model startups still get funded?

New Foundation Model startups can definitely still raise money, and some are starting with amounts that used to belong to mature companies.

We count 12 first financings in the current period, up from seven last year and ten two years ago. Their share of total deals slipped to 19% because follow-on funding grew even faster, but the absolute number of new entrants increased.

The surprising part is the size. Median first financing reached roughly $340 million, compared with just $15 million last year.

Mirendil is a good example of how strange the new baseline can look. The company formally launched this year with a 20-person founding team drawn from places including Anthropic, xAI, Google DeepMind and OpenAI, and announced a $200 million seed led by Andreessen Horowitz and Kleiner Perkins, with NVIDIA also investing.

Other unusually well-capitalized entrants push the median still higher. A Foundation Model startup can now arrive with an elite research team, an expensive compute plan and hundreds of millions of dollars before following anything resembling the traditional Seed-A-B progression.

New-company funding	2026	2025	2024
First financings	12	7	10
First-financing share	19.0%	21.2%	23.3%
Follow-on financings	50	26	33
Follow-on share	79.4%	78.8%	76.7%
Capital in first financings	$13.55B	$2.23B	$1.29B
Capital in follow-ons	~$238.1B	$72.61B	$11.42B
Median first financing	~$340M	$15M	$30.5M
Median follow-on financing	High-$200Ms	$138.6M	$80M

Are early-stage Foundation Model startups getting squeezed out?

Early-stage Foundation Model startups are still getting funded today, although the market is clearly more mature than it was two years ago.

Seed and Series A rounds account for roughly one-third of current activity. That is close to last year's 33.3% share, so there is little evidence of a fresh early-stage collapse this year.

The bigger shift happened between 2024 and 2025. Two years ago, early-stage rounds represented 51.2% of deals. The market has since settled at a much lower share.

Company age tells a similar story. The median funded company is around 3.2 years old now, barely different from 3.1 years last year but well above roughly 1.4 years two years ago.

Where 2026 looks extreme is the amount of money attached to mature companies. Series D+ and growth financings absorb roughly nine-tenths of capital, even though those rounds make up a much smaller share of transactions.

The current Foundation Model market looks mature and very capital-intensive. It does not look newly hostile to young companies.

Which Foundation Model categories are actually getting hotter?

Language models and world/robotics models are the two Foundation Model categories where we see the clearest real momentum today.

Language-model deal share jumped from 18.2% last year to 31.7%, while the category's capital share rose to about 45.7%. Both the number of financings and the dollars are moving in the same direction.

Recent rounds keep adding to that impression. Mistral raised €3 billion to expand frontier research, compute and international deployment. Arcee announced a new Series B to continue building open-weight models in the U.S. Mirendil launched with $200 million for a frontier research lab focused on science.

World and robotics models have a different but equally interesting trajectory. Deal count rose from six to 14, and the median round climbed from about $98 million to $305 million.

Image-generation companies also became more active, rising from two deals to seven, helped by large 3D-generation financings such as Tripo AI. Audio and speech remain busy in company terms, although their share of total dollars is small.

Foundation Model category	2026 deals	Deal share	2026 capital	Capital share	Median round	Deal-share change vs 2025
Language Models	20	31.7%	~$115.1B	45.7%	~$700M	+13.6 pp
World & Robotics Models	14	22.2%	$6.77B	2.7%	$305M	+4.0 pp
Video Generation	11	17.5%	$4.47B	1.8%	$86M	-0.7 pp
Image Generation	7	11.1%	$1.49B	0.6%	$200M	+5.1 pp
Audio & Speech	7	11.1%	$1.03B	0.4%	$30M	-4.0 pp
Multimodal Models	4	6.3%	$122.9B	48.8%	$409M	-17.9 pp
Total	63	100%	~$251.8B	100%	$263.3M overall	-

Is multimodal AI really winning half of all Foundation Model funding?

Multimodal AI's near-50% share of Foundation Model funding is mostly an OpenAI effect, so it tells us very little about how broad investor interest in the category really is.

Multimodal companies account for only four current-period deals, or 6.3% of activity. Last year, they represented 24.2%.

OpenAI happens to sit in that bucket, and a $122 billion transaction is large enough to overwhelm category statistics. That pushes multimodal models to about 48.8% of total capital even while the number of deals falls sharply.

Language models give us a much better example of broad category momentum because their deal share and capital share both increased.

Any claim that "multimodal AI captured half the market" therefore needs a large asterisk. Half the dollars went into the category, but investor activity across multimodal companies was actually much narrower.

Why are world models suddenly attracting so much money?

World models are attracting much more money because investors are now placing serious bets on AI that learns how environments change, rather than focusing only on language.

We count 14 world and robotics model deals in the current period, more than twice last year's six, and the median financing has roughly tripled.

General Intuition gives us a particularly clear example. The company trains models on action-labelled gameplay data so they can learn how actions change an environment, with the aim of transferring those capabilities into simulation and robotics. General Intuition now says it has raised more than $650 million over the past year.

The company is already working with partners across games, simulation and robotics. That makes the thesis more concrete than the old idea that "world models" are simply research projects waiting for a use case.

TwelveLabs points in a related direction from video understanding. Its recent $100 million Series B is funding models designed to understand long-form video as a record of how the physical world unfolds, while the company has lately been expanding its work toward physical intelligence.

We still do not know which architecture will win here. What is already clear is that world understanding has moved from a niche research theme to one of the best-funded new parts of the Foundation Model market.

Is generative video still hot with investors?

Generative video is still getting funded heavily, but it no longer stands out as the fastest-moving part of the Foundation Model market.

We count 11 video-generation financings in the current period. That is enough to make video one of the market's busiest categories, yet its deal share is basically flat versus last year.

The category is also becoming more diverse geographically. Chinese companies such as Kling AI and ShengShu Technology now sit alongside North American companies that dominated much of the earlier funding cycle.

The competitive set around video has become much wider. Investors can now put meaningful sums into language models, world models, robotics intelligence, 3D generation, speech and specialized open models.

Video has kept its place at the table. It just no longer owns the most interesting funding narrative by itself.

Why is China suddenly everywhere in Foundation Model funding?

China has become one of the biggest sources of new Foundation Model deal activity, with 18 current-period financings versus only four in the comparable period last year.

This is broader than a single Chinese champion. Moonshot AI is raising around large language models. Kling AI is pursuing generative video. Tripo AI is building 3D-generation models. HiDream focuses on visual models. ShengShu develops video and world-model technology.

Tripo AI is especially useful for seeing how quickly capital can stack up around a promising Chinese model company. We recorded multiple distinct financings during the year, including a $150 million Series A3 backed by strategic investors from automotive and gaming, followed later by a much larger Series B/B+ financing.

DeepSeek's financing shows another side of the Chinese market. The roughly RMB51 billion transaction involved its founder alongside large corporates, financial investors and national AI capital. We treat that capital carefully because the structure is different from a conventional outside VC round.

China is currently adding many more funded companies than it did last year, even though the United States still dominates global Foundation Model dollars.

Is Foundation Model funding becoming less U.S.-centric?

Foundation Model funding is becoming much less U.S.-centric by deal count, but the global capital map still revolves around American frontier labs.

Asia-Pacific now accounts for 36.5% of current-period deals, up more than 21 percentage points from last year. North America's share dropped from 69.7% to 52.4%, even though the absolute number of North American deals still increased.

Europe remains smaller, with seven deals, but several European companies are raising serious amounts. ElevenLabs closed a $500 million Series D. Stability AI recently added $76 million from a group including Electronic Arts, Sony Music Group, Universal Music Group and Warner Music Group. Mistral's €3 billion round is on another scale again.

Mistral is particularly relevant because its financing is tied to a distinctly European pitch: more frontier research, more compute capacity and sovereign AI infrastructure. That gives European Foundation Model funding a different flavor from simply trying to reproduce Silicon Valley.

Still, North America takes close to 90% of current capital. Geographic diversification is happening much faster in company count than in money.

Geography	Current deal count	Current deal share	Capital share	Deal-share change vs 2025
North America	33	52.4%	~89%	-17.3 pp
Asia-Pacific	23	36.5%	~8%	+21.4 pp
Europe	7	11.1%	~2.5%	-1.0 pp
Middle East	0	0%	0%	-3.0 pp
United States	32	-	Dominant	Higher absolute count
China	18	-	Much smaller than U.S.	Up from 4 deals
France	Includes Mistral	-	Lifted by €3B Mistral round	Large capital gain
United Kingdom	Includes ElevenLabs, Stability AI	-	Several sizeable rounds	Active model cluster

Why are so many strategic companies investing in Foundation Model startups now?

Strategic investors are much more involved in Foundation Model funding now because owning part of a model company can also strengthen a company's position in compute, cloud, distribution, media, enterprise software or industrial AI.

Strategic participation appears in 68.3% of current-period deals, compared with 51.5% last year. That is a big enough move to change the investor base.

Compute companies are an obvious part of the story. NVIDIA and AMD repeatedly invest around businesses that could become major accelerator customers. Cloud and platform groups such as Amazon, Google, Microsoft, Tencent and Alibaba have similar incentives around distribution and infrastructure.

Lately, the pattern has spread well beyond chips and cloud. Stability AI's latest round brought in Electronic Arts, Sony Music Group, Universal Music Group and Warner Music Group. These companies operate directly in the creative markets affected by generative image, audio and video models.

Tripo AI's investor base includes gaming companies and Geely Capital, which makes sense for a company generating 3D assets used in games, digital content and potentially automotive design.

Strategic capital is increasingly following the future customer and supplier map of the Foundation Model industry.

Has compute become part of the funding deal itself?

Compute is now so important to Foundation Model companies that financing and infrastructure strategy increasingly overlap.

OpenAI made this explicit when announcing its giant round, calling durable compute access a strategic advantage and linking compute with research, products, developer usage and enterprise deployment. The same announcement said OpenAI was generating around $2 billion of monthly revenue, which helps explain why investors are willing to underwrite infrastructure spending on such an unusual scale.

Anthropic's recent numbers point in the same direction. When the company announced its $65 billion Series H, it said its revenue run rate had crossed $47 billion and that the new capital would help expand compute as well as products and partnerships.

AMD's separate agreement with Anthropic makes the connection unusually visible. AMD said Anthropic plans to deploy up to two gigawatts of its Instinct MI450 infrastructure, while AMD committed to a strategic equity investment of up to $5 billion. We exclude that full $5 billion commitment from our ordinary capital totals because the timing of the equity tranches is unclear, but the structure itself is revealing.

Mistral is also using fresh capital to expand its own compute capacity. The company says the €3 billion Series D will help scale infrastructure alongside frontier-model development.

Today, a Foundation Model funding round can be partly about securing the physical capacity required to keep competing.

Are new investors entering Foundation Model funding?

Foundation Model investing looks much more crowded today, with far more disclosed lead and co-lead investors appearing across the market.

We identify roughly 79 distinct named lead or co-lead investors in the current period, versus 32 last year.

That expansion comes despite worse disclosure. A lead investor is identifiable for about 68% of current deals, down from roughly 79% last year. We are therefore seeing much greater investor breadth even with more missing names.

Sequoia appears four times as a disclosed lead or co-lead, while Radical Ventures appears three times. GIC, Coatue, Dragoneer, MGX, Greycroft, Alibaba Cloud and several others appear more than once.

Arcee's latest Series B gives a good picture of the mix now entering the market: Vista Equity Partners, Cambium Capital and Emergence Capital led, while Microsoft-linked M12, Hitachi Ventures and Wipro also participated.

The Foundation Model investor base now stretches well beyond specialist venture firms into large growth investors, sovereign capital, technology companies and industrial strategics.

Are Foundation Model investors becoming more selective even while funding more companies?

Foundation Model investors are funding many more companies, but they are concentrating serious money around a much smaller group of companies they believe can reach global scale.

That split is visible across almost every layer of the market. Small-round capital has not beaten its 2024 level, while large financings have exploded. Follow-on rounds account for about four-fifths of deals. Strategic participation is rising. Mature rounds dominate capital.

At the same time, 2026 is not a closed club. First-financing count increased, funded-company count almost doubled and new investor names continue to appear.

Getting some capital and getting frontier-scale capital have become two very different achievements.

For a company with the right team, compute access and technical story, hundreds of millions can now arrive remarkably early. Most other companies are operating in a funding market that looks much closer to historical levels.

What should we be most careful about in the 2026 Foundation Model funding numbers?

The headline Foundation Model funding total is useful, but we would trust deal count and company breadth more than we would trust every last decimal point of capital or stage mix.

Several transactions have unusual structures. Anthropic's AMD-linked investment is worth up to $5 billion, but the individual equity tranches and timing were not disclosed, so we keep it out of ordinary capital calculations. Poolside's reported $1 billion transaction is tied to a broader NVIDIA arrangement and may overlap an earlier commitment, which makes it poor material for a clean year-over-year comparison.

DeepSeek's financing includes major founder participation. Some Chinese amounts are lower bounds or currency conversions. General Intuition's June financing has conflicting stage reporting. Lead-investor disclosure is also less complete than last year.

Stage analysis has its own limits because only around four-fifths of current deals have a stage we can use confidently. Founding dates are available for 54 of 63 current-period deals rather than the full sample.

None of those issues seriously threatens the core conclusion that the market has become much busier. They do mean that a precise statement such as "Series A capital rose exactly X%" deserves much less confidence than the observation that nine-figure rounds have become dramatically more common.

What is actually changing in Foundation Model funding?

Foundation Model funding is genuinely stronger today. We found far more deals and far more funded companies than last year, so the market's expansion cannot be reduced to OpenAI and Anthropic.

The biggest change sits in financing size. Nine-figure rounds have become common across a much wider group of companies, while frontier labs can now raise sums that look more like infrastructure financing than traditional venture capital.

At the lower end, the story is far calmer. Funding below $100 million has recovered from last year but remains slightly below the comparable 2024 level. The boom becomes much less dramatic once the very large rounds disappear.

Investor attention is moving as well. Language models currently show the strongest combination of rising deal breadth and rising capital, while world and robotics models have moved from a smaller research-heavy niche into a serious funding category.

Geographically, China is producing a much larger share of new deals and Europe has built several well-funded model companies, but U.S. frontier labs still control the overwhelming majority of dollars.

The boundary between venture capital and industrial strategy keeps getting blurrier. Chipmakers, cloud providers, enterprise software companies, media groups and other operating companies are showing up more often because Foundation Models increasingly sit inside their own future infrastructure, distribution or content strategies.

The market now runs at two speeds. The overall ecosystem is getting broader, while a relatively small group of frontier companies is pulling farther away from everyone else in capital, compute and financial scale.