What are the fundraising trends in the foundation model market?
In our updated market reports, you will find everything you need
SUMMARY
We analyzed publicly disclosed equity funding rounds in the foundation model market across 2024, 2025, and year-to-date 2026. The dataset keeps only disclosed equity rounds of $300K or more from pure-play foundation model companies, including language, multimodal, vision, audio, code, scientific foundation model, model-hosting, and fine-tuning infrastructure companies.
The foundation model market has expanded dramatically by capital. Full-year funding rose from $28.7B in 2024 to $71.5B in 2025, and year-to-date 2026 already reached $176.7B through early July.
The increase is not a normal broad-market recovery. In year-to-date 2026, the top 3 rounds captured 97.3% of all capital, which means the market’s funding total is dominated by OpenAI, Anthropic, and xAI rather than by the median company.
Deal activity is also rising in the freshest period. Year-to-date 2026 produced 20 deals, compared with 13 deals over the comparable period in 2025, so more companies are getting funded even though the dollars are still extremely concentrated.
Round sizes are moving up sharply. The median foundation model round rose from $106M in 2024 to $302.5M in 2025, then settled at $277.5M in year-to-date 2026. The average round is far higher because a few giant late-stage financings distort the market.
Language Models dominate the foundation model market by capital. They captured 92.3% of full-year 2025 funding and 97.4% of year-to-date 2026 funding, even though they represented only 29.2% of 2025 deals and 20.0% of 2026 deals so far.
Multimodal Models are gaining momentum by deal count. The category produced no qualifying deals over the comparable 2025 period, but it generated 7 deals and $3.7B in year-to-date 2026, making it the clearest new formation area outside general-purpose language models.
Scientific Foundation Models remain active but proof-gated. The category produced 5 deals in 2025 and 4 deals in year-to-date 2026, but its dollar share stayed small, suggesting investors are funding specific scientific primitives rather than treating the category like a frontier LLM arms race.
North America is becoming even more dominant. It captured 87.1% of capital in 2024, 96.8% in 2025, and 99.8% in year-to-date 2026, while Europe and Asia-Pacific have lost relative share in the disclosed public funding record.
The foundation model market is now bifurcated. At the top, a few frontier labs raise infrastructure-scale rounds from strategic, sovereign, institutional, and venture investors. Beneath them, a broader set of startups is still forming around world models, physical AI, scientific models, inference routing, and adaptation tools.
Is more or less capital going into the foundation model market?
More capital is going into the foundation model market, and the increase is dramatic. Year-to-date 2026 funding reached about $176.7B, compared with about $51.7B over the same period in 2025, which means capital has grown by roughly 3.4x in the freshest comparison.
The full-year comparison tells the same story. Full-year 2025 funding was about $71.5B, up from about $28.7B in 2024, so the market had already grown by roughly 2.5x before the extraordinary 2026 financing wave.
The important caveat is that the increase is not broad in the way a normal venture-market expansion would be broad. In year-to-date 2026, the top 3 rounds captured 97.3% of all capital. Over the comparable 2025 period, the top 3 rounds captured 93.7%. So the foundation model market is receiving far more capital, but that capital is overwhelmingly concentrated in a few frontier labs.
Excluding the largest round makes the point even clearer. Year-to-date 2026 funding falls from $176.7B to $54.7B if the largest deal is removed. That remaining number is still larger than the comparable 2025 market excluding its largest deal, but the difference between the headline figure and the adjusted figure shows how much one company can reshape the entire market narrative.
The practical interpretation is straightforward. The foundation model market is expanding, but it is expanding as a strategic infrastructure category, not as a conventional startup category. More money is going in, but the marginal dollar is mostly being allocated to companies that investors believe can operate at frontier-model scale.
Is foundation model funding activity driven by more deals or larger rounds?
Foundation model funding activity is driven much more by larger rounds than by more deals. Year-to-date 2026 deal count rose from 13 deals over the comparable 2025 period to 20 deals, but capital rose from about $51.7B to about $176.7B, which means funding grew much faster than deal activity.
The full-year comparison confirms the same pattern even more cleanly. In 2025, the foundation model market had fewer deals than in 2024, falling from 35 to 24, yet total capital rose from $28.7B to $71.5B. That is the clearest possible sign that larger rounds, not more rounds, drove the market higher.
The typical round also moved up. The median round rose from $106M in 2024 to $302.5M in 2025, before landing at $277.5M in year-to-date 2026. That means the middle of the market became larger, not just the top.
Still, the average is not representative of the normal company. In year-to-date 2026, the average round was about $8.8B, while the median was $277.5M. The largest round was 439.6x the median round. That gap makes average round size a poor indicator of what most funded foundation model companies are actually raising.
So the right reading is that more deals helped, especially in 2026, but bigger checks are the main driver. The foundation model market is now defined by the ability of a small number of companies to absorb enormous amounts of capital.
Is foundation model capital moving toward later-stage or earlier-stage companies?
Foundation model capital is moving decisively toward later-stage companies, even though early-stage company formation is still active by deal count. In year-to-date 2026, Series D+ rounds captured 98.3% of all capital, while Seed and Series A together captured only 1.4%.
The deal-count picture looks more balanced, but it should not be mistaken for capital balance. Series A was the most common stage in year-to-date 2026, with 10 of 20 deals. But those Series A deals represented only about $2.4B of the $176.7B raised.
By contrast, 5 Series D+ rounds represented about $173.7B. That means the funding market is early-stage by formation activity and late-stage by dollar allocation.
The 2025 full-year comparison points in the same direction. Seed and Series A represented 29.2% of 2025 deals but only 4.2% of capital, while late-stage rounds, including Series B and above, Growth Equity, and Unknown stage rounds, represented 95.8% of capital.
The foundation model market is therefore not abandoning new companies. Investors are still backing new model theses. But the serious capital is moving toward later-stage labs and platforms that already have technical credibility, distribution, compute relationships, or strategic importance.
Is the foundation model market maturing or still experimental?
The foundation model market is maturing at the center and still experimental at the edges. The maturity signal is obvious in the capital structure: year-to-date 2026 late-stage rounds represented 98.6% of funding, and follow-on financings captured 98.9% of all capital.
That means the largest checks are going to companies investors already consider validated. OpenAI, Anthropic, xAI, Runway, Skild AI, DeepInfra, OpenRouter, and similar companies are not being funded as speculative formation bets. They are being financed as strategic AI infrastructure or scaled model platforms.
But the experimentation signal is also real. First financings represented 40.0% of year-to-date 2026 deals, up from 15.4% over the comparable 2025 period and 16.7% for full-year 2025. That is a strong sign that new foundation model company formation has not stopped.
The experimental activity is concentrated outside the most crowded general-purpose language-model race. Newer companies are forming around multimodal models, robotics foundation models, world models, tabular and time-series models, plant biology models, physical AI, and fine-tuning/adaptation systems.
The best interpretation is that the foundation model market has split into two markets. Frontier language-model labs are mature, capital-intensive, and institutionally financed. Specialized model categories remain exploratory, but they are being explored with unusually large early-stage checks.
Are new startups still entering the foundation model market?
Yes, new startups are still entering the foundation model market, and the year-to-date 2026 signal is stronger than the 2025 signal. First financings represented 40.0% of year-to-date 2026 deals, compared with 15.4% over the comparable 2025 period and 16.7% for full-year 2025.
That is a meaningful increase in new-company formation. The foundation model market is not closed to new entrants, even though the largest capital pools are controlled by proven companies.
The capital share tells the other side of the story. First financings captured only 1.1% of year-to-date 2026 capital. In full-year 2025, first financings captured 3.6% of capital. In full-year 2024, first financings captured 5.3% of capital. New startups are entering, but they are not receiving a large share of total dollars.
The category mix matters. In year-to-date 2026, Multimodal Models had 4 first financings out of 7 deals, Scientific Foundation Models had 2 first financings out of 4 deals, and Fine Tuning Services had 1 first financing. That suggests new entrants are not mainly trying to be the next OpenAI; they are trying to own a new model primitive or deployment bottleneck.
The practical takeaway is that the foundation model market remains open to differentiated new companies, but not to undifferentiated LLM imitation. New startups are most visible where the market still has unresolved technical surfaces: physical AI, robotics, world modeling, scientific prediction, and adaptive model systems.
Are more investors entering the foundation model market?
More investors appear to be entering the foundation model market in 2026, but the increase needs to be interpreted carefully because many investors are clustered around the same giant rounds. Year-to-date 2026 had about 113 explicitly named investors, compared with about 50 over the comparable 2025 period.
The number of unique tier-1 investors also rose sharply, from about 25 over the comparable 2025 period to about 53 in year-to-date 2026. On the surface, that looks like a major broadening of the capital base.
The full-year comparison is more nuanced. Full-year 2025 had about 75 disclosed investors, down from 121 in 2024, even though capital rose sharply. So 2025 was a year when fewer investors funded much larger rounds. The 2026 year-to-date figures suggest investor breadth returned, but mainly through very large financings that pulled in strategic, sovereign, institutional, and corporate capital.
This matters because investor count does not mean risk appetite is spreading evenly. Many investors are entering through OpenAI, Anthropic, xAI, or other highly credentialed companies rather than through small unproven startups.
The foundation model market is therefore attracting more types of capital, not just more venture funds. The capital stack now includes hyperscalers, GPU suppliers, sovereign wealth funds, asset managers, corporate strategics, growth investors, and traditional AI venture firms. The market has become too large for venture capital alone.
Are top investors getting more or less active in the foundation model market?
Top investors are getting more active in the foundation model market where it matters most: the large, strategically important rounds. NVIDIA or NVentures appeared in 8 disclosed year-to-date 2026 deals, matching its full-year 2025 activity and reinforcing its role as one of the most important strategic investors in the market.
The investor mix has shifted from classic venture syndicates toward strategic capital stacks. In year-to-date 2026, repeated investors included NVIDIA, Fidelity, MGX, Khosla Ventures, General Catalyst, Andreessen Horowitz, Sequoia, Lightspeed, Salesforce Ventures, Coatue, Temasek, TPG, QIA, BlackRock-affiliated funds, Blackstone, and Insight Partners.
The full-year trend shows the same broad direction. In 2024, Andreessen Horowitz appeared in 10 deals and NVIDIA in 8. In 2025, Andreessen Horowitz and NVIDIA each appeared in about 8 deals. In 2026, NVIDIA remained highly active, while the largest rounds also drew in more sovereign, institutional, hyperscaler, and strategic participation.
The qualitative change is more important than the raw deal count. Top investors are no longer just backing promising AI startups. They are helping finance infrastructure-scale companies that need compute, data-center access, distribution, and geopolitical capital relationships.
The strongest reading is that top investors are becoming more active at the strategic center of the foundation model market. The most important investors are not merely writing checks; they are helping companies secure compute, credibility, and platform positioning.
Which foundation model subcategories are gaining momentum?
The foundation model subcategories gaining momentum are Language Models by capital, Multimodal Models by deal count, and Scientific Foundation Models by new-company formation. Each category is gaining momentum in a different way, so one metric is not enough.
Language Models have the strongest capital momentum. They raised about $172.1B in year-to-date 2026, up from about $50.5B over the comparable 2025 period. They captured 97.4% of year-to-date 2026 capital, which makes the category the overwhelming funding center of the market.
Multimodal Models have the strongest deal momentum. The category had no qualifying deals over the comparable 2025 period, but it produced 7 deals and $3.7B in year-to-date 2026. It also exceeded full-year 2025 deal count and capital by early July.
Scientific Foundation Models are gaining formation momentum rather than capital dominance. They produced 4 year-to-date 2026 deals, compared with 3 over the comparable 2025 period, and funding was roughly stable at about $298M versus $291M.
The important interpretation is that capital and experimentation are moving to different places. Language Models are where the biggest checks go. Multimodal and scientific models are where investors are testing the next set of model primitives.
Which foundation model subcategories are losing momentum?
The foundation model subcategories losing momentum are standalone Audio Models, standalone Code Models, and, in relative capital terms, Model Hosting Platforms. The underlying use cases are not necessarily weakening, but their financing is becoming less visible as standalone pure-play foundation model categories.
Audio Models show the clearest decline in deal activity. Full-year 2024 had 2 Audio Model deals worth $205M. Full-year 2025 had 1 Audio Model deal worth $180M. Year-to-date 2026 had no qualifying Audio Model deals.
Code Models have also disappeared as a standalone financing category. Full-year 2024 had 2 Code Model deals worth $820M. Full-year 2025 had no qualifying Code Model deals, and year-to-date 2026 also had none. The likely explanation is absorption: code capability is increasingly financed inside broader language-model platforms or developer-tool companies, not as a pure-play foundation model category.
Model Hosting Platforms remain active, but their capital intensity has weakened. The category had 3 year-to-date 2026 deals, the same as the comparable 2025 period, but capital fell from $429M to $235M. That suggests inference, routing, and deployment infrastructure are still investable, but investors are not valuing them like frontier-model ownership.
Fine Tuning Services remain too small to call a strong recovery. The category had no qualifying full-year 2025 deals and 1 year-to-date 2026 deal worth $50M. That is a sign of life, but not yet a durable trend.
Which regions are gaining momentum in foundation model funding?
North America is gaining the most momentum in foundation model funding, and the increase is visible in both capital and deal count. North America captured 99.8% of year-to-date 2026 capital, up from 99.5% over the comparable 2025 period and 96.8% for full-year 2025.
The full-year trend shows how much the market has concentrated around North America. North America captured 87.1% of capital in 2024, 96.8% in 2025, and 99.8% in year-to-date 2026. Its deal share also rose from 68.6% in 2024 to 83.3% in 2025, then remained high at 80.0% in year-to-date 2026.
The Middle East shows a smaller but notable emerging signal. Year-to-date 2026 included 2 Middle East deals and $320M of capital, after no qualifying full-year 2024 or 2025 deals in the market figures. That is not yet a broad regional ecosystem, but it is a new sign of participation around world models, physical AI, and strategic AI infrastructure.
Asia-Pacific also reappeared in year-to-date 2026 with 1 deal worth $15M after no qualifying full-year 2025 deals in the assembled figures. But the dollar amount is too small to call a major regional comeback.
The honest conclusion is that North America is gaining real momentum, while the Middle East is gaining early visibility. Other regions are present, but not yet contributing enough capital to change the center of gravity.
Which regions are losing momentum in foundation model funding?
Europe and Asia-Pacific are losing relative momentum in foundation model funding, especially compared with their earlier visibility. Europe had 3 deals and about $271M over the comparable 2025 period, but only 1 deal and about $7M in year-to-date 2026.
The full-year comparison is less severe but still shows relative weakness. Europe raised about $898M in 2024 and $2.261B in 2025, helped by companies such as Mistral AI, ElevenLabs, Bioptimus, Latent Labs, and other technical challengers. But Europe’s share of total capital stayed low because North American rounds grew far faster.
Asia-Pacific’s decline is sharper in the public funding record. Asia-Pacific had 5 full-year 2024 deals and about $2.803B of capital, driven by Chinese and Japanese foundation model companies. Full-year 2025 had no qualifying Asia-Pacific deals in the assembled figures, and year-to-date 2026 had only 1 small deal.
Some caution is necessary because Chinese domestic rounds and private strategic financings can be harder to verify through English-language public sources. Even with that caveat, the disclosed public-market signal is clear: Asia-Pacific has lost visible funding momentum since 2024.
The practical interpretation is that Europe remains technically credible but financially underpowered, while Asia-Pacific has become much less visible in the disclosed equity dataset. North America is absorbing the capital that would be needed for either region to compete at frontier-lab scale.
Is the foundation model market becoming more global or more regionally concentrated?
The foundation model market is becoming more regionally concentrated, not more global. North America captured 87.1% of capital in 2024, 96.8% in 2025, and 99.8% in year-to-date 2026.
This concentration is not only a capital artifact caused by one mega-round. North America’s deal share also rose from 68.6% in 2024 to 83.3% in 2025, then remained at 80.0% in year-to-date 2026. That means the company-formation signal and the capital signal both point toward North America.
Europe remains relevant by company quality, but not by capital scale. Europe had 16.7% of full-year 2025 deals but only 3.2% of capital, and its year-to-date 2026 share fell close to zero. The foundation model market can still produce European technical challengers, but those companies are not yet being funded at North American frontier-lab scale.
Asia-Pacific was meaningful in 2024 but nearly absent in 2025 and only minimally present in year-to-date 2026. The Middle East is emerging selectively, but more as a strategic capital and deployment region than as a broad company-formation region.
So the foundation model market is globally important, but the disclosed equity-financing market is not globalizing. It is concentrating around North American companies, with global capital increasingly participating in North American-led rounds.
Is foundation model capital moving toward proven winners or new opportunities?
Foundation model capital is moving overwhelmingly toward proven winners, while deal formation is moving toward new opportunities. In year-to-date 2026, follow-on financings represented 60.0% of deals but captured 98.9% of capital.
First financings tell the opposite story. They represented 40.0% of year-to-date 2026 deals but only 1.1% of capital. That means investors are willing to fund new ideas, but they reserve most balance-sheet exposure for companies that have already demonstrated credibility.
The pattern was already visible in earlier periods. Full-year 2025 first financings represented 16.7% of deals and 3.6% of capital. Full-year 2024 first financings represented 22.9% of deals and 5.3% of capital. Across all three periods, first financings are much more meaningful by count than by dollars.
The new opportunities are concentrated in categories where the market is still searching for the next defensible primitive: multimodal systems, robotics foundation models, world models, physical AI, tabular and time-series models, plant biology, manufacturing models, inference routing, and adaptation layers.
The proven winners are concentrated in frontier language models and scaled model platforms. The foundation model market is therefore exploratory at the bottom and highly selective at the top.
Is the foundation model market becoming winner-takes-most?
Yes, the foundation model market is becoming winner-takes-most by capital. In year-to-date 2026, the largest deal captured 69.0% of total capital, the top 3 captured 97.3%, the top 5 captured 98.5%, and the top 10 captured 99.6%.
The trend toward concentration has been building. In 2024, the top 3 deals captured 64.8% of capital. In 2025, the top 3 captured 81.1%. In year-to-date 2026, the top 3 captured 97.3%. The direction is unmistakable.
The median-versus-average gap shows how extreme the concentration has become. In year-to-date 2026, the median round was $277.5M, while the average round was $8.835B. The largest round was 439.6x the median.
The right phrase is winner-takes-most, not winner-takes-all. New companies are still being formed, and several subcategories remain active. But economic power is concentrating in a few companies with frontier model capability, massive distribution, strategic compute access, and the ability to attract sovereign and institutional capital.
For anyone interpreting the foundation model market, top-round concentration is now the first number to check. Without it, the funding total is almost impossible to read correctly.
Is the next wave of foundation model winners becoming visible?
The next wave of foundation model winners is becoming visible, but it is emerging around the edges of the market rather than directly displacing the largest language-model labs. The strongest emerging areas are multimodal models, world models, physical AI, robotics foundation models, scientific foundation models, and model-routing infrastructure.
Multimodal Models are the clearest formation signal. The category had 7 year-to-date 2026 deals, more than any other category, after having no qualifying deals over the comparable 2025 period. That points to a shift from text-centric AI toward models that understand video, space, robotics, physical environments, and simulation.
Scientific Foundation Models are also showing early winner formation. Companies in tabular prediction, time-series forecasting, plant biology, manufacturing, and broader scientific AI are attracting capital, but at round sizes that suggest investors still want proof before underwriting frontier-scale bets.
Model Hosting Platforms remain important as an enabling layer. DeepInfra, OpenRouter, Portkey, and similar companies suggest that routing, inference, observability, reliability, and model access are still monetizable bottlenecks.
The next wave is visible, but subordinate. The current winners are still OpenAI, Anthropic, xAI, and other scaled frontier labs. The next wave is likely to come from companies that control new high-value model loops rather than companies that simply imitate general-purpose language models.
Is the foundation model funding landscape fragmenting or consolidating?
The foundation model funding landscape is consolidating by capital and fragmenting by company formation. Both things are happening at the same time, which is why the market can look broad by deal count and extremely narrow by dollars.
Capital is consolidating. The top 3 year-to-date 2026 deals captured 97.3% of funding, up from 93.7% over the comparable 2025 period and 81.1% for full-year 2025. The bottom half of year-to-date 2026 deals captured only 0.4% of total capital.
Company formation is fragmenting. Year-to-date 2026 had 20 deals across 20 unique companies, up from 13 deals across 13 unique companies over the comparable 2025 period. First financings rose to 40.0% of deals.
The category map is also fragmenting. Year-to-date 2026 included Language Models, Multimodal Models, Scientific Foundation Models, Model Hosting Platforms, Vision Models, and Fine Tuning Services. That is a broad formation pattern, even though the dollars are heavily concentrated in language models.
The right conclusion is asymmetric. The foundation model market is consolidating at the top and fragmenting underneath. The top absorbs the capital; the bottom explores the frontier.
Where is investor attention shifting in the foundation model market?
Investor attention in the foundation model market is shifting in three directions: toward frontier language labs for capital scale, toward multimodal and world-model systems for new-category formation, and toward deployment infrastructure for monetizing model fragmentation.
The biggest capital shift is still toward frontier language models. Language Models captured 97.4% of year-to-date 2026 capital and 92.3% of full-year 2025 capital. Investors are not funding many general-purpose LLM companies; they are funding the few companies they believe can compete at global frontier scale.
The most interesting formation shift is toward multimodal and physical-world models. Multimodal Models had no qualifying deals over the comparable 2025 period, 3 full-year 2025 deals, and 7 year-to-date 2026 deals. That is a sharp acceleration in interest around world models, robotics models, spatial reasoning, video prediction, and embodied AI.
A smaller but durable shift is toward scientific foundation models. Investors are backing companies with specific data or deployment loops in biology, tabular data, time series, plant genomics, and precision manufacturing. These companies do not attract language-model-scale dollars, but they show where investors think proprietary data and scientific validation can create defensibility.
Model infrastructure remains a practical focus. Hosting, inference, routing, gateway, and observability companies are still raising, but their small capital share suggests investors see them as necessary infrastructure utilities rather than the main ownership layer of the foundation model market.
INSIGHTS
The insights below come from reviewing disclosed equity rounds in the foundation model market across 2024, 2025, and year-to-date 2026.
- The foundation model market is no longer well described as a venture category. With $176.7B raised in year-to-date 2026 and 97.3% of that capital captured by the top 3 rounds, the market now behaves more like strategic infrastructure financing than ordinary startup funding.
- Deal count and capital volume now answer different questions. Deal count shows where experimentation is happening, while capital volume shows which companies are treated as systemically important.
- The strongest capital-allocation rule is that general-purpose language-model companies are financed as platforms, while most other categories are financed as options. Language Models had only 20.0% of year-to-date 2026 deals but 97.4% of capital.
- The foundation model market is simultaneously early-stage and late-stage. Series A was the most common year-to-date 2026 stage by deal count, but Series D+ captured 98.3% of capital.
- The largest rounds are now so large that average round size has limited interpretive value. In year-to-date 2026, the average round was about $8.8B, the median was $277.5M, and the largest round was 439.6x the median.
- A single mega-round can redefine the annual market. OpenAI’s year-to-date 2026 financing alone represented 69.0% of total capital, so any market-wide funding trend must be decomposed before being interpreted.
- The foundation model market is becoming more winner-takes-most, but not less innovative. First financings rose to 40.0% of year-to-date 2026 deals, even as those first financings captured only 1.1% of capital.
- New entrants are strongest where the dominant frontier labs have not yet fully closed the product surface. Multimodal, scientific, physical-AI, and adaptation companies show more first-financing activity than general-purpose language-model companies.
- The next-wave signal is stronger in deal count than in dollars. Multimodal Models produced 7 year-to-date 2026 deals, the highest category count, but only 2.1% of capital.
- The market’s most important category shift is from text models toward world models, not from language models toward another capital center. World-model and physical-AI companies are multiplying, but they have not yet displaced LLMs as the main funding magnet.
- Scientific Foundation Models are being underwritten on specificity, not scale. Investors are funding tabular, time-series, plant-biology, biology, and manufacturing-model companies, but at sizes that imply proof-gated conviction rather than frontier-lab confidence.
- Model Hosting Platforms are investable but not treated as scarce enough to command frontier-lab capital. They represented 15.0% of year-to-date 2026 deals but only 0.13% of capital.
- Code Models disappearing as a standalone category does not mean code AI is weak. The better interpretation is that code capability has been absorbed into broad language-model platforms and developer-workflow companies that no longer look like pure-play foundation-model financings.
- Audio Models disappearing in year-to-date 2026 suggests a similar absorption dynamic. Audio remains a major AI capability, but standalone audio foundation-model financing appears less visible after earlier category leaders raised large rounds.
- The strongest diligence signal has shifted from whether a company builds a model to whether the company controls a scarce primitive. Scarce primitives include frontier talent, compute access, proprietary data, distribution, routing volume, scientific validation, or real-world feedback loops.
- NVIDIA’s repeated presence is more informative than many venture-fund logos. NVIDIA or NVentures appeared in 8 year-to-date 2026 deals, indicating that compute alignment is a recurring credibility signal across model labs, hosting platforms, and applied foundation-model categories.
- Sovereign and institutional investors are turning foundation model funding into geopolitical and infrastructure exposure. The presence of MGX, QIA, GIC, Temasek, BlackRock-affiliated funds, Blackstone, and other large institutions shows the market has outgrown ordinary VC capital supply.
- North America’s dominance is not just a byproduct of one company. North America represented 80.0% of year-to-date 2026 deals and 99.8% of capital, showing concentration in both company count and check size.
- Europe remains technically credible but financially underpowered relative to North America. Europe produced meaningful companies in 2024 and 2025, but its year-to-date 2026 share fell to one small deal and almost no capital share.
- The foundation model market is splitting into three layers: frontier platform labs, specialized model primitives, and deployment infrastructure. Each layer has different capital needs, proof standards, investor types, and likely exit paths.
- The market’s apparent breadth can be overstated if each deal is counted equally. A 20-deal year-to-date 2026 market looks broad until the bottom half of deals is shown to represent only 0.4% of capital.
- The market’s apparent concentration can also be overstated if only dollars are considered. The same year-to-date 2026 period had 20 unique companies and 40.0% first financings, which means the exploratory pipeline remains active.
- The practical forecasting rule is that the next credible challengers are more likely to emerge from new model surfaces than from direct LLM imitation. Multimodal/world models, scientific primitives, physical-AI foundation models, and infrastructure layers show the strongest formation signals outside the incumbent LLM capital race.
OUR METHODOLOGY TO BUILD THIS TRACKER
We built this foundation model funding tracker by reviewing publicly disclosed equity rounds across 2024, 2025, and year-to-date 2026. A company counts as pure-play when more than 80% of its activity is dedicated to foundation model development, model deployment infrastructure, model hosting, or model adaptation and fine-tuning services.
We define the foundation model market as large reusable AI models that can be adapted across many tasks, products, and industries. The included categories are Language Models, Multimodal Models, Vision Models, Audio Models, Code Models, Scientific Foundation Models, Model Hosting Platforms, and Fine Tuning Services.
We applied four main filters. First, we only included equity rounds, so grants, debt-only financings, acquisitions, secondary-only transactions, and non-equity strategic arrangements were excluded. Second, we only counted disclosed rounds of $300K or more. Third, we only kept pure-play companies, which means generic AI applications, AI agents, AI chips, cloud or GPU leasing, consulting, robotics companies without a foundation-model core, and broader software companies using AI were excluded. Fourth, each qualifying round had to be supported by a direct company announcement, investor announcement, press release, tier-1 media report, specialized technology source, or relevant regional publication.
Undisclosed-amount rounds were excluded because they cannot be used reliably in total funding, average round size, median round size, category share, stage share, geography share, or concentration calculations. Rounds with disclosed amounts but unclear stages were kept and marked as Unknown when the company otherwise fit the market definition.
All dates are based on the month and year of the funding announcement or the most reliable public reporting date. Dollar amounts are treated as disclosed or reported at announcement time, and non-dollar amounts are converted only when the source or accompanying reporting provides a clear dollar equivalent. The result is a public-source funding tracker, not a private PitchBook or Crunchbase export, so private undisclosed rounds and poorly reported domestic financings may be missing.
Related blog posts
- Which companies have raised the most funding in the foundation model market?
- Which companies are the most valued in the foundation model market?
- Is fundraising strong in the foundation model market right now?
