Foundation Model Startup Funding
In our updated market reports, you will find everything you need
SUMMARY
This report analyzes publicly disclosed equity rounds raised by pure-play foundation model companies between August 2025 and July 2026, a 12-month window covering every geography. We kept disclosed equity rounds of $300K or more, and excluded debt-only, secondary-only, public-company financings, acquisitions, and companies not primarily built around reusable foundation models.
Over this period, the foundation model market raised $51.06B across 21 disclosed deals and 21 unique companies. This is a large market in dollars, but not a broad one in deal count.
Capital in the foundation model market is extremely concentrated. The top 1 deal represents 39.17% of disclosed capital, the top 3 deals represent 80.88%, and the top 10 deals represent 96.30%.
The median round size is $320M, while the average round size is $2.43B. That wide gap shows how much the market is shaped by outliers such as xAI, Anthropic, and OpenAI.
Deal flow averaged 1.75 rounds per month, but capital was highly uneven. August 2025, September 2025, and January 2026 together contributed $44.22B, or 86.60% of all disclosed capital.
Language Models dominate the foundation model market by capital raised. They account for $45.79B, or 89.68% of disclosed dollars, from only 6 of 21 deals.
Model Hosting Platforms are the second-largest category by dollars, with $1.85B raised. This shows that investors are also funding the deployment layer around foundation models, not only the labs building them.
The foundation model market is late-stage in capital terms. Series C and later rounds plus Growth Equity account for $46.46B, or 90.98% of total disclosed funding.
North America dominates the market with $46.86B raised, or 91.77% of total capital. Europe produced a meaningful 5 deals, but captured only 8.11% of dollars.
The investor base shows strong strategic repetition. NVIDIA appears across model labs, scientific AI, vision models, code models, and world models, making it the clearest recurring ecosystem investor in the dataset.
What are all the funding deals in the foundation model market from August 2025 to July 2026?
The table below lists every disclosed equity round raised by pure-play foundation model companies between August 2025 and July 2026. We count as “pure-play” foundation model companies those building large reusable AI models, scientific foundation models, multimodal systems, audio models, code models, vision models, or model platforms built specifically for deploying and adapting those models.
| Company | What they do | Category | Date | Stage | Deal size | Region | Main investors | Source |
|---|---|---|---|---|---|---|---|---|
| OpenAI | Builds large reusable frontier AI models and products such as ChatGPT, APIs, and multimodal foundation model systems | Language Models | Aug 2025 | Growth Equity | $8,300M | North America | Not disclosed in dataset | TechCrunch |
| Chai Discovery | Builds AI foundation models for molecular design, protein interaction prediction, and de novo antibody design | Scientific Foundation Models | Aug 2025 | Series A | $70M | North America | Menlo Ventures; Google-linked investors or executives | Business Wire |
| Cohere | Builds enterprise-focused language, multilingual, multimodal, speech, coding, and retrieval models for business and sovereign AI use cases | Language Models | Aug 2025 | Growth Equity | $500M | North America | NVIDIA; Salesforce Ventures | Cohere |
| Anthropic | Builds Claude frontier AI models for enterprise, developer, consumer, coding, and safety-focused AI applications | Language Models | Sep 2025 | Series D+ | $13,000M | North America | Lightspeed Venture Partners | Anthropic |
| Mistral AI | Builds open and enterprise foundation models, assistants, agents, and customization tools for multilingual and enterprise AI | Language Models | Sep 2025 | Series C | $1,990M | Europe | NVIDIA; a16z; General Catalyst; Lightspeed Venture Partners; Index Ventures | Mistral AI |
| PixVerse | Builds AI video generation models and a consumer-facing platform for text-to-video and generative video creation | Vision Models | Sep 2025 | Series B | $60M | Asia-Pacific | Not disclosed in dataset | PR Newswire |
| Periodic Labs | Builds AI systems for autonomous scientific discovery, including hypothesis generation and experimental science workflows | Scientific Foundation Models | Sep 2025 | Seed | $300M | North America | NVIDIA; a16z; Bezos Expeditions; Google-linked investors or executives | TechCrunch |
| Reflection AI | Builds open frontier AI models and infrastructure intended to compete with closed model labs and open-source frontier model providers | Language Models | Oct 2025 | Series B | $2,000M | North America | Google-linked investors or executives | TechCrunch |
| Lila Sciences | Builds scientific superintelligence systems and AI Science Factories for autonomous research across life sciences and physical sciences | Scientific Foundation Models | Oct 2025 | Series A | $350M | North America | NVIDIA; General Catalyst | Lila Sciences |
| Fireworks AI | Provides an AI inference cloud for deploying, fine-tuning, evaluating, and serving open and custom generative AI models | Model Hosting Platforms | Oct 2025 | Series C | $250M | North America | Lightspeed Venture Partners; Index Ventures; Sequoia Capital; Salesforce Ventures | Business Wire |
| Inception | Builds diffusion large language models, including fast code and text models designed for real-time AI applications | Code Models | Nov 2025 | Seed | $50M | North America | NVIDIA; Menlo Ventures | Business Wire |
| Wispr | Builds voice-first foundation models for speech-driven computing and a voice operating system layer | Audio Models | Nov 2025 | Series A | $25M | North America | NEA | PR Newswire |
| Black Forest Labs | Builds frontier image generation models, including the FLUX family of visual foundation models | Vision Models | Dec 2025 | Series B | $300M | Europe | NVIDIA; a16z; Google-linked investors or executives | Black Forest Labs |
| xAI | Builds Grok and advanced AI models, consumer and enterprise AI products, and supporting infrastructure | Language Models | Jan 2026 | Series D+ | $20,000M | North America | Not disclosed in dataset | xAI |
| ElevenLabs | Builds voice and audio AI models for speech synthesis, dubbing, voice agents, and multilingual audio generation | Audio Models | Feb 2026 | Series D+ | $500M | Europe | Sequoia Capital | TechCrunch |
| Runway | Builds AI video generation models and world models for creative media, simulation, and generative video workflows | Vision Models | Feb 2026 | Series D+ | $315M | North America | NVIDIA | TechCrunch |
| Advanced Machine Intelligence / AMI Labs | Builds world models intended to learn from reality rather than only from language, with API and open-source ambitions | Multimodal Models | Mar 2026 | Seed | $1,030M | Europe | NVIDIA; Bezos Expeditions | TechCrunch |
| Baseten | Provides inference infrastructure for running, optimizing, scaling, and deploying AI models in production | Model Hosting Platforms | Jun 2026 | Series D+ | $1,500M | North America | Not disclosed in dataset | Baseten |
| Runpod | Provides an AI developer cloud for training, fine-tuning, inference, and model deployment workflows | Model Hosting Platforms | Jun 2026 | Growth Equity | $100M | North America | Summit Partners | PR Newswire |
| General Intuition | Builds spatial and world-model AI systems trained on gameplay data to perceive, predict, and act across environments | Multimodal Models | Jun 2026 | Series A | $320M | Europe | General Catalyst; Bezos Expeditions | Axios |
| TwelveLabs | Builds video understanding and video intelligence foundation models for perception, knowledge, and reasoning over video | Vision Models | Jul 2026 | Series B | $100M | North America | Index Ventures; NEA | TwelveLabs |
OUR METHODOLOGY TO BUILD THIS TRACKER
We built this foundation model funding tracker by reviewing every publicly disclosed equity round raised by pure-play foundation model companies between August 2025 and July 2026. A company counts as pure-play when more than 80% of its activity is dedicated to large reusable AI models, multimodal models, vision models, audio models, code models, scientific foundation models, or foundation-model-specific deployment infrastructure.
We applied four filters to build the dataset. First, we only included equity rounds, so debt-only rounds, secondary-only transactions, public-company financings, grants, acquisitions, and revenue financing are excluded. Second, we only counted rounds of $300K or more. Third, we only kept pure-play foundation model companies. And fourth, every entry had to be confirmed by a direct company announcement, a press release, or a tier-1 media report, with the source URL preserved for every row.
The final dataset contains 21 disclosed deals across 21 unique companies, and every average, median, share, and concentration ratio is computed on that disclosed sample. Privately raised rounds that were never publicly announced are necessarily missing, which is a known limitation of any public-only foundation model funding tracker.
How active has fundraising been in the foundation model market?
As of July 2026, fundraising in the foundation model market has been very large in dollars but selective in deal count. Over the past 12 months, the dataset includes 21 disclosed equity rounds and $51.06B raised across 21 unique companies.
The deal count translates to an average of 1.75 disclosed rounds per month and a median of 1.5 deals per month. That means the foundation model market produced regular activity, but not a deep long tail of financings.
Dollar flow is much more uneven than deal flow. Average capital raised per month was $4.26B, while the median month was $922.5M, showing that a few months carried most of the funding signal.
August 2025, September 2025, and January 2026 were the decisive months. Together they contributed $44.22B, or 86.60% of total disclosed capital, while April and May 2026 had no qualifying deals.
How concentrated has fundraising been in the foundation model market?
As of July 2026, fundraising in the foundation model market is extremely concentrated at the top. Over the past 12 months, the largest deal accounts for 39.17% of total disclosed capital, the top 3 deals reach 80.88%, and the top 5 reach 88.70%.
The top 10 deals account for 96.30% of all dollars. That means the lower half of the company list is strategically interesting, but financially marginal in aggregate capital analysis.
This concentration is not just a company-level pattern. Language Models account for only 28.57% of deals, but they capture 89.68% of disclosed capital.
The practical reading rule is simple. Any statement about the foundation model market’s total funding should separate xAI, Anthropic, OpenAI, Reflection AI, and Mistral AI from the rest of the dataset.
How much of the foundation model funding signal is driven by outliers?
As of July 2026, most of the foundation model funding signal is driven by outliers. Over the past 12 months, the market raised $51.06B, but the median round was $320M and the average round was $2.43B.
That average-to-median gap is the clearest sign of an outlier-led market. A few multi-billion-dollar rounds pull the mean far above what a typical disclosed financing looks like.
Rounds above $50M are almost the default public financing unit in the foundation model market. There were 19 deals above $50M, equal to 90.48% of all disclosed rounds.
The contrast below $50M is striking. Total capital raised excluding rounds above $50M was only $75M, which means smaller visible rounds barely register in the market’s dollar totals.
Is the foundation model market broad with many targets, or narrow with few fundable companies?
As of July 2026, the foundation model market is narrow rather than broad. Over the past 12 months, only 21 unique companies raised qualifying disclosed equity rounds, despite the market’s very large headline funding total.
This narrowness matters because foundation models are capital intensive, talent intensive, and compute intensive. The dataset suggests that visible public financing mostly goes to companies with strong research, infrastructure, or distribution credibility.
The absence of sub-$20M qualifying deals reinforces the point. Pure-play foundation model startups that reach public financing announcements are already institutionally validated, infrastructure-heavy, or both.
The market is also narrow by category. Language Models, Vision Models, Scientific Foundation Models, and Model Hosting Platforms account for 16 of the 21 disclosed deals.
Is the foundation model market mostly an early-stage formation market or a late-stage scaling market?
As of July 2026, the foundation model market is late-stage in dollars but mixed by deal count. Over the past 12 months, late-stage rounds, Series C and later plus Growth Equity, captured $46.46B, or 90.98% of total disclosed capital.
Series D+ alone raised $35.32B across 5 deals. That stage represents 69.17% of total capital, showing that the largest checks went to scaled companies rather than new entrants.
Early-stage rounds still appear in the dataset, but they are financially dwarfed. Seed through Series B rounds raised $4.61B, or only 9.02% of total disclosed funding.
Stage labels can be misleading in this market. Seed rounds include Periodic Labs at $300M and AMI Labs at $1.03B, which are closer to infrastructure-scale company formation than ordinary software seed funding.
Which categories attract the most investor attention in foundation models?
As of July 2026, Language Models attract the most investor attention in the foundation model market by capital raised. Over the past 12 months, the category raised $45.79B, or 89.68% of all disclosed dollars, across 6 deals.
Deal count tells a more balanced story. Language Models account for 28.57% of deals, Vision Models for 19.05%, Scientific Foundation Models for 14.29%, and Model Hosting Platforms for 14.29%.
This means the market is not only about language model labs. Investors are also backing vision, scientific discovery, multimodal systems, audio models, code models, and the infrastructure layer used to deploy models.
Still, the center of gravity is clear. Language Models are the only category where investors are consistently underwriting platform-scale balance sheets measured in billions of dollars.
Which categories attract disproportionately large checks in the foundation model market?
As of July 2026, Language Models attract disproportionately large checks in the foundation model market. Over the past 12 months, the category had a capital share to deal share ratio of 3.14, far above every other category.
The average Language Models deal was $7.63B, and the median was $5.15B. Those numbers reflect platform-scale financings for xAI, Anthropic, OpenAI, Reflection AI, Mistral AI, and Cohere.
Multimodal Models rank second by average deal size at $675M, helped by AMI Labs and General Intuition. Model Hosting Platforms follow at $616.7M, driven mainly by Baseten’s $1.5B round.
Vision Models, Audio Models, Scientific Foundation Models, and Code Models all receive meaningful funding, but their check-size profiles are much smaller. They look like specialized model segments rather than frontier balance-sheet wars.
Which geographies matter most for fundraising in the foundation model market?
As of July 2026, North America is the geography that matters most for fundraising in the foundation model market. Over the past 12 months, the region captured $46.86B, or 91.77% of disclosed capital, across 15 deals.
Europe is credible by deal count but much smaller by dollars. It produced 5 deals, equal to 23.81% of the dataset, but raised $4.14B, or only 8.11% of disclosed capital.
Asia-Pacific is almost absent under this strict pure-play definition. PixVerse is the only qualifying Asia-Pacific deal, with $60M raised in September 2025.
The geography split shows that the largest capital pools remain concentrated around North American labs. Europe has visible lab formation, but not yet the same scale of funding concentration.
Is the foundation model opportunity set broad or concentrated in one hub?
As of July 2026, the foundation model opportunity set is concentrated in one dominant hub, with a secondary European cluster. Over the past 12 months, North America accounted for 71.43% of deals and 91.77% of dollars.
The difference between deal share and capital share is important. North America does not merely produce more companies; it produces the largest financings by a very wide margin.
Europe’s 5 deals show that the region is a real foundation model market, especially around Mistral AI, AMI Labs, ElevenLabs, Black Forest Labs, and General Intuition. But its 8.11% capital share remains far below North America’s.
Latin America, the Middle East, and Africa produced no qualifying disclosed foundation model deals in the dataset. That does not mean there is no AI activity there, only that no strict pure-play financings appeared in the public sample.
Is the foundation model market a market of small experiments or scaled financings?
As of July 2026, the foundation model market is a market of scaled financings, not small experiments. Over the past 12 months, 19 of 21 disclosed deals were above $50M, and 15 were above $100M.
There were no disclosed rounds below $20M in the dataset. Only 2 deals fell between $20M and $50M, while every other round was already at megaround scale.
The median disclosed round size was $320M. In most venture markets, that would be a late-stage or growth round; in the foundation model market, it is close to the visible norm.
The reason is structural. Building foundation models often requires frontier research talent, large-scale compute access, proprietary data or environments, and distribution into model usage.
Who are the investors that appear the most in foundation model fundraising?
As of July 2026, NVIDIA is the most visible repeat investor in foundation model fundraising. Over the past 12 months, NVIDIA or NVentures appeared across Cohere, Mistral AI, Periodic Labs, Lila Sciences, Black Forest Labs, Inception, Runway, and AMI Labs.
That pattern makes NVIDIA look less like a normal financial investor and more like an ecosystem allocator. Its repeated presence maps to future demand for compute consumption across model categories.
Andreessen Horowitz appears in Mistral AI, Periodic Labs, and Black Forest Labs. General Catalyst appears in Mistral AI, Lila Sciences, and General Intuition, while Bezos-linked capital appears in Periodic Labs, AMI Labs, and General Intuition.
Several other investors appear twice, including Lightspeed Venture Partners, Index Ventures, Menlo Ventures, Sequoia Capital, Salesforce Ventures, NEA, and Google-linked investors or executives. The repeat-investor pattern confirms that conviction clusters around a small number of high-credibility labs and platforms.
One important caveat is that round announcements rarely disclose individual investor check sizes. Any investor ranking should be read as repeated participation, not exact capital committed.
INSIGHTS
The insights below come from reviewing every disclosed equity round in the foundation model market between August 2025 and July 2026. They are not row-by-row summaries. They are the reusable patterns that kept showing up across the 21-deal dataset, and they are meant to stay useful when reading any future foundation model funding announcement.
- The foundation model market is a capital-concentration market, not a broad funding market. The top 3 deals alone account for 80.88% of all disclosed dollars. Any market-size conclusion that does not isolate xAI, Anthropic, and OpenAI will overstate the funding environment for the median company.
- Deal count and capital tell opposite stories in the foundation model market. Language Models represent only 28.57% of deals but 89.68% of capital. That means investors are funding platform-scale balance sheets, not simply backing the most active category.
- The median round is $320M, while the average round is $2.43B. This 7.6x gap is the clearest sign that the dataset is governed by extreme outliers. The average round should not be read as typical.
- The market’s fundraising health depends heavily on three months. August 2025, September 2025, and January 2026 together contribute $44.22B. Without those months, the foundation model market looks much smaller and less continuous.
- The absence of sub-$20M disclosed rounds is structurally important. Pure-play foundation model companies that reach public funding visibility are already capital-intensive, institutionally validated, or both.
- The market is late-stage in dollars but not exclusively late-stage in deal count. Series D+ and Growth Equity dominate capital, while Seed, Series A, and Series B still appear across 11 deals. Younger companies are present, but financially dwarfed.
- Seed rounds no longer behave like ordinary seed rounds in the foundation model market. Periodic Labs raised $300M, and AMI Labs raised $1.03B. These are infrastructure-scale company formations, not conventional early-stage software financings.
- Scientific Foundation Models look small by capital share but large by conviction. The category holds only 1.41% of dollars, yet its median round is $300M. Investors appear selective, but willing to fund high-pedigree science labs at scale.
- Vision Models show a different funding structure than Scientific Foundation Models. They have more deals and similar total capital, but smaller round sizes. That suggests video and image model companies are funded more like productizing labs than frontier compute balance sheets.
- Audio Models have a two-tier structure. ElevenLabs is a scaled category leader with a $500M round, while Wispr is a narrower interface bet at $25M. The category should not be interpreted as uniformly infrastructure-heavy.
- Model Hosting Platforms are the clearest alternative to owning the frontier model. Baseten, Fireworks AI, and Runpod show that investors also want exposure to model deployment, inference, and usage growth. This is a way to monetize model proliferation without taking full frontier-lab risk.
- Fine Tuning Services produced no qualifying pure-play deals in the period. This likely means fine-tuning has been absorbed into inference clouds and model platforms. It does not currently look like a standalone fundable category at scale.
- NVIDIA is the strongest investor signal in the dataset. Its repeated presence across language, science, vision, code, and multimodal companies makes it look like an ecosystem allocator. The common thread is future compute demand.
- The foundation model market has split into two investable theses. One is owning the frontier model. The other is owning the serving layer. Application-only companies are mostly absent because the dataset filters for reusable model infrastructure.
- North America dominates both deal count and capital, but the capital skew is stronger. It has 71.43% of deals and 91.77% of dollars. The largest financing capacity remains concentrated around U.S. and Canadian labs.
- Europe is credible by deal count but not comparable by capital scale. Its 23.81% deal share and 8.11% capital share show a real lab-formation ecosystem, but one still far below North American funding depth.
- Asia-Pacific is almost absent under this strict public-source definition. PixVerse is the only qualifying deal. This likely reflects a mix of corporate-internal funding, lower disclosure transparency, and companies falling outside the pure-play scope.
- The largest European rounds are not only product-market-fit stories. Mistral AI and AMI Labs are also sovereignty and architecture bets. Geopolitics is part of the investment logic in frontier AI.
- Different categories require different validation signals. Language labs are judged by valuation, frontier capability, and enterprise adoption. Model hosting platforms are judged by inference volume and usage. Scientific model labs are judged by research credibility, experimental loops, and long-term option value.
- The foundation model market is too capital-intensive for ordinary venture construction. With a $320M median round and 15 deals above $100M, category leadership increasingly requires sovereign, strategic, crossover, or late-stage capital.
TechCrunch (OpenAI), Business Wire (Chai Discovery), Cohere (Funding round), Anthropic (Series F), Mistral AI (Series C), PR Newswire (PixVerse), TechCrunch (Periodic Labs), TechCrunch (Reflection AI), Lila Sciences (Series A), Business Wire (Fireworks AI), Business Wire (Inception), PR Newswire (Wispr), Black Forest Labs (Series B), xAI (Series E), TechCrunch (ElevenLabs), TechCrunch (Runway), TechCrunch (AMI Labs), Baseten (Series F), PR Newswire (Runpod), Axios (General Intuition)
Related blog posts
- A full list of funding deals in the foundation model market
- Which companies have raised the most funding in the foundation model market?
- Which companies are the most valued in the foundation model market?
