What’s getting funded in AI right now?
In our AI infrastructure market deck, you will find everything you need to understand the market
SUMMARY
What’s getting funded in AI right now? Frontier labs, physical AI, infrastructure, and applications that can take over expensive pieces of real work are pulling in the strongest capital.
The boom is enormous, but it is not broad. In the latest full quarter, 89% of AI funding went into just 142 rounds of at least $100 million, so the headline market and the experience of a normal startup are now two very different things.
Frontier labs have effectively become their own venture-capital market. A tiny group of research teams can raise hundreds of millions or billions before building a normal commercial business, with founder pedigree and perceived technical scarcity doing work that revenue would normally have to do.
Infrastructure is still hot, but the center of gravity is widening beyond training compute. Investors are funding the deployment layer too: interconnects, optical networking, model routing, search infrastructure, data systems and the software that keeps production AI running.
Physical AI looks like the clearest new wave because both dollars and deal activity are moving up together. That is different from a category whose growth comes mostly from one or two giant rounds, even though autonomous-vehicle financings still inflate the totals.
Among applications, coding and customer service stand out for a simple reason: investors can already see usage, workflow change and measurable economic value. Those markets do not need as much imagination as a generic assistant does.
Legal and healthcare AI are becoming durable vertical categories rather than temporary extensions of the chatbot cycle. Domain knowledge, workflow integration and distribution are giving specialist companies room even as the underlying general-purpose models improve.
AI security is unusually interesting at seed because agent adoption creates new problems at the same time it creates new products. Identity, permissions, governance and monitoring become more valuable once software can read data, call APIs and act without a human approving every click.
Consumer AI can still attract huge rounds, but the bar is much higher. Investors are rewarding products that create exceptional engagement or a genuinely new behavior; another undifferentiated chatbot, image tool or assistant has a much harder story to tell.
The deeper shift may be in what AI companies are allowed to sell against. The strongest application companies increasingly compete for labor budgets, not just software budgets, which can make the addressable market far larger when the product actually completes work instead of merely helping someone do it.
Geography remains lopsided. The U.S. dominates funding, China has enough scale to sustain a serious second frontier ecosystem, and Europe is producing strong clusters in coding, legal AI, robotics and frontier research rather than one unified AI capital center.
This market map, featured in our AI infrastructure market deck, highlights top companies and startups in the AI infrastructure market
Is AI funding really booming right now?
Yes, AI startup funding is at extraordinary levels today, but the boom is much narrower than the headline totals suggest.
CB Insights counted $149.5 billion of global AI equity funding in the latest full quarter, the second-highest quarterly total it has ever recorded. The striking number sits underneath it: 89% of those dollars went into only 142 rounds worth at least $100 million. Those mega-rounds represented roughly 6% of AI deals.
The broader venture market shows the same concentration. Crunchbase calculated that startups globally raised about $510 billion in the first half of 2026, already more than the $440 billion invested during all of 2025. OpenAI and Anthropic alone accounted for $217 billion, or 43% of every venture dollar invested globally during that period.
The trend has continued lately. Crunchbase counted $65 billion of global venture funding in July, up 100% year over year, with a record 14 billion-dollar rounds. AI companies took about $35 billion, or 53% of the month’s total.
So AI really is absorbing unprecedented amounts of money. For the average startup, though, fundraising has not suddenly become easy. CB Insights says overall venture deal count recently fell to its lowest level in more than a decade. The current market rewards companies investors already believe can become very large, and it rewards them with checks that would have looked almost impossible a few years ago.
| What we measure | Latest evidence | What it tells us |
|---|---|---|
| AI funding in the latest full quarter | $149.5B | AI capital remains near record levels |
| Share going to $100M+ rounds | 89% | Most dollars are concentrated at the top |
| Mega-round count | 142 | Large rounds are common, but still a small share of deals |
| New AI unicorns | 37 | Investors are still creating new category leaders |
| AI share of July global funding | 53% | AI dominance has continued into the current quarter |
Are frontier AI labs still swallowing most of the money?
Frontier AI labs still swallow the largest checks in venture capital today, and investors are even willing to finance brand-new labs before they have meaningful revenue.
Crunchbase calculated that foundational AI companies raised $178 billion across only 24 deals in the first quarter of 2026. That was already twice the amount the category raised during all of 2025.
The size of individual bets keeps stretching upward. Safe Superintelligence, founded by former OpenAI chief scientist Ilya Sutskever, recently secured a reported $5 billion investment from Nvidia. DeepSeek raised roughly $7.4 billion at a reported $50 billion valuation. Moonshot AI followed with a $3.5 billion round. Earlier this year, Ineffable Intelligence raised $1.1 billion at seed, while River AI has already raised $1.1 billion across its seed and Series A rounds despite being founded only this year.
Founder pedigree clearly matters here. River AI founder Igor Babuschkin previously worked at DeepMind, OpenAI and xAI. Ineffable Intelligence was founded by former DeepMind researcher David Silver. Safe Superintelligence is led by one of the central researchers behind modern large language models.
There is also more variety in what these labs are trying to build. Some are pursuing conventional frontier models, while others are betting on reinforcement learning, continuous learning, open-source architectures, robotics or radically different training methods. Crunchbase counted 10 frontier labs joining its Unicorn Board in June alone.
This is one of the strangest corners of venture capital right now: an exceptional research team can raise hundreds of millions, sometimes billions, before building a normal commercial business. Very few founders can play that game. Investors are still aggressively financing the ones who can.
If you want more recent data on this point, please see our latest AI infrastructure market report.
As this chart shows, and as featured in our AI infrastructure market deck, search interest in AI infrastructure has risen sharply
Is AI infrastructure still hot now that AI apps are making real money?
AI infrastructure remains one of the hottest places to raise money, although the infrastructure investors care about these days increasingly sits around deployment, data, networking and production workloads.
Databricks offers the clearest recent example. The company just raised another $5 billion, only eight months after its previous $5 billion financing. According to Crunchbase, Databricks has now passed a $7 billion revenue run rate, up more than 80% year over year, and its valuation climbed from $134 billion late last year to roughly $190 billion.
Further down the stack, Point2 Technology recently raised $136 million for high-speed interconnect technology used in AI data centers. Semiconductor startups are also attracting unusually large rounds: MatX raised $500 million for chips designed around large-model workloads, Ayar Labs raised $500 million for optical connectivity, and Etched reportedly raised $500 million for specialized AI chips.
Investors are also financing the software sitting between raw compute and AI applications. OpenRouter raised $113 million at a $1.3 billion valuation for technology that lets developers route requests among hundreds of models. Exa raised $250 million at a $2.2 billion valuation for search infrastructure increasingly used by coding, sales and other AI agents.
The infrastructure trade has widened. Training frontier models still consumes enormous capital, but a growing number of startups are being funded around what happens after the model exists: moving data, connecting chips, choosing models, serving requests and keeping production systems reliable.
As more AI products reach real users, the plumbing underneath them is becoming more valuable.
Is physical AI actually the next big funding wave?
Physical AI has become the clearest new funding wave in AI right now, with money moving into robotics, autonomous vehicles, drones, industrial automation and the models that help machines understand the physical world.
Crunchbase’s latest analysis counted $47.4 billion invested across 521 physical-AI deals during the first half of 2026. That was almost four times the $12 billion invested during the second half of 2025 and nearly 80% above the first half of 2025.
The comparison with earlier years is even stronger. Investors put $41.9 billion into physical AI during 2022, 2023 and 2024 combined. The category raised more than that in six months this year.
Large autonomous-vehicle rounds contributed heavily, including Waymo’s $16 billion financing, but the deal activity goes much deeper. CB Insights found industrial humanoid robots were the most active individual AI market in its latest full-quarter data with 20 deals, while robot foundation models recorded another 15.
Some of the new companies are being financed on a frontier-lab scale. Advanced Machine Intelligence raised $1.03 billion at seed. Generalist AI raised $400 million to develop intelligence that can control robots across complex tasks. Project Prometheus raised $10 billion in April around AI applied to manufacturing and the physical world.
This is one of the few areas where both deal count and funding dollars are rising sharply at the same time. Investors are betting that AI’s next large market will involve machines doing work in warehouses, factories, roads, laboratories and other physical environments.
| Physical-AI measure | Current evidence | Change |
|---|---|---|
| First-half funding | $47.4B | Nearly 4x H2 2025 |
| First-half deals | 521 | Up from 470 in H2 2025 |
| Industrial humanoid deals in latest quarter | 20 | No. 1 AI market by deal count |
| Robot foundation-model deals | 15 | Among the most active AI markets |
| 2022–2024 combined funding | $41.9B | Less than the latest six-month total |
If you want more recent data on this point, please see our latest AI infrastructure market report.
This chart, included in our AI infrastructure market deck, shows annual VC investment in AI infrastructure startups
Are AI coding startups still getting huge checks?
AI coding is still one of the easiest application categories to finance because several independent companies are already showing the kind of growth investors usually see much later.
Lovable recently raised $400 million at a $13.3 billion valuation, roughly doubling the $6.6 billion valuation from its previous round. The company says users have created around 60 million projects on its platform and generate roughly 900 million visits a month.
CodeRabbit raised $143 million in one of the largest rounds announced in the latest weekly funding data. The company says its AI code-review product is already used by 17,000 customers and 150,000 open-source projects. Blitzy raised $200 million at a $1.4 billion valuation for software that automates larger parts of enterprise development rather than simply suggesting code.
Replit raised $400 million earlier this year at a $9 billion valuation. Cognition, the company behind Devin, has also raised repeated large rounds as investors bet on increasingly autonomous software development.
What makes coding different from many AI categories is the number of companies showing genuine usage at the same time. Investors can see developers already changing how they write, review and ship software. They can also see several business models forming around different parts of the workflow: code generation, app creation, code review, testing, legacy-code migration and autonomous development.
Competition from OpenAI, Anthropic, Google and Microsoft has done little to stop the financing. If anything, the latest rounds suggest investors currently believe coding will be large enough to support several major independent companies.
Are AI agents still hot with investors, or has the hype cooled?
AI agents are still getting funded heavily, although investors now care much more about what the agent actually does than whether a startup uses the word “agent.”
Enterprise workflows are where the money is clustering. Freehand recently raised $75 million to automate supply-chain spending and back-office work. Exa’s $250 million round partly reflects its role as search infrastructure for agents. Sycamore raised $65 million at seed around infrastructure for deploying and controlling enterprise agents.
The market is also becoming more specialized. CB Insights counted 13 deals for coding agents and another 13 for legal agents in its latest full-quarter ranking, putting both among the most active AI categories. Financial agents, sales agents, customer-service agents and security agents are producing their own clusters of companies.
Investors appear much less interested these days in a general assistant that promises to handle almost anything. The stronger pitch is painfully specific: resolve this support case, review this pull request, research this company, process this invoice, prepare this legal document or manage this purchasing workflow.
That specificity makes the business easier to judge. A customer can measure how much work the agent completes, how often humans need to intervene and how much the company saves.
The agent market looks healthier than it did when every startup seemed to be building a generic autonomous assistant. Funding is moving toward agents with a clear job.
If you want more recent data on this point, please see our latest AI infrastructure market report.
This chart, included in our AI infrastructure market deck, shows why CoreWeave is winning in AI infrastructure
Why is customer-service AI attracting so much money?
Customer-service AI is currently attracting some of the largest application-layer checks because companies can compare the cost of an AI agent directly with the cost of handling millions of human support interactions.
Sierra raised $950 million at a $15 billion valuation, one of the largest enterprise-AI rounds of the year. The company builds AI agents that handle customer conversations and complete tasks such as returns, insurance interactions and financial-service requests.
The category has several other well-funded contenders. Parloa has reached a $3 billion valuation while focusing heavily on AI voice agents for customer interactions. Decagon reached a multibillion-dollar valuation with agents that handle support workflows for large companies. Wonderful raised $150 million at a $2 billion valuation less than two years after its founding.
Sales, marketing and customer-experience startups have attracted around $3.7 billion of funding so far this year according to Crunchbase, with AI companies taking a large share of the money.
Customer service works especially well for AI because companies already know exactly what the existing process costs. They know how many conversations arrive, how many employees handle them, how long each case takes and how many cases get resolved.
A startup that can reliably automate those conversations has a very different sales pitch from a generic productivity tool. The buyer can quickly attach a dollar value to the work.
Is healthcare AI still one of the strongest vertical bets?
Healthcare AI remains one of the strongest vertical funding markets because startups can attack both expensive administrative work and high-value clinical or scientific problems.
The market has already shown unusual staying power. Crunchbase estimated that AI-powered health companies raised roughly $14 billion in 2025, up about 63% from the previous year, and large financings have continued.
The money is spread across very different businesses. OpenEvidence has raised large rounds for an AI medical search product aimed at doctors. Abridge has repeatedly attracted major funding for automating clinical documentation. AI drug-discovery companies such as Isomorphic Labs operate at the other end of the risk spectrum, where investors are funding scientific platforms that may need years before producing meaningful drug revenue.
Aureka Biotechnologies provides a fresher example. The company recently raised $100 million for a “lab-in-the-loop” system that combines AI with physical experiments to improve drug discovery. CB Insights also highlighted AI drug discovery as one of the areas producing some of the largest recent digital-health rounds.
Healthcare gives investors several ways to play AI. Documentation and administrative tools can prove revenue quickly. Medical search can build habitual clinician usage. Drug discovery offers a much bigger scientific upside with much longer timelines.
Healthcare looks more durable than many fashionable verticals because capital keeps returning across different business models and funding stages.
This chart, included in our AI infrastructure market deck, shows annual funding in AI infrastructure startups
Has legal AI become a real venture category?
Legal AI has clearly become a real venture category, with enough large companies and repeated funding rounds that the market can no longer be explained by one breakout startup.
Crunchbase recorded $4.08 billion of legal-tech funding in 2025, up 77% from roughly $2.3 billion the previous year. More than $1.3 billion had already been invested again by late April this year.
Legora shows how quickly valuations can move when investors see strong adoption. The Swedish company raised $550 million at a $5.55 billion valuation and then added another $50 million from Nvidia’s venture arm. Only a few months earlier, Legora had been valued at $1.8 billion.
Harvey has raised more than $1 billion since launch, including four separate rounds in 2025. Other companies are building around tax research, corporate legal departments, case preparation and plaintiff-side litigation.
The interesting part is how specific many of the successful products have become. Lawyers spend huge amounts of time searching documents, reviewing evidence, drafting repetitive material and organizing cases. Those workflows give AI startups concrete jobs to automate while leaving the lawyer in control of the final judgment.
The recent deal count supports the same conclusion: legal agents were tied with coding agents among CB Insights’ most active AI markets.
Legal AI is one of the better examples of a vertical where domain knowledge, workflow integration and distribution can still create room around increasingly powerful general-purpose models.
Is AI security becoming one of the hottest seed markets?
AI security is becoming one of the hottest AI seed categories right now, and the unusually large number of rounds makes the trend harder to dismiss as a few headline deals.
Crunchbase counted $855 million across more than 150 reported AI-security seed rounds so far this year. At the current pace, the category is heading toward an all-time high for seed investment.
The breadth is striking. Oak raised $60 million at seed for identity intelligence built for the AI era. Cylake raised $45 million for an AI-native security platform. JetStream Security launched with $34 million for AI governance and enterprise protection.
Older cybersecurity problems are also being rebuilt around AI. Tenex.AI raised $250 million for an AI-native security service. XBow raised $120 million for automated offensive security. Upwind Security raised $250 million for cloud security.
Agent adoption is creating another layer of demand. Companies now have software that can read data, call APIs and take actions without a person clicking every button. Someone has to decide which agents can access which systems, verify their identities, monitor what they do and stop them when something goes wrong.
Seed investors appear to be betting that those problems will grow alongside agent adoption. Security has the advantage of getting more important as AI use expands.
This chart, included in our AI infrastructure market deck, compares the main business model options for AI cloud infrastructure providers
Can consumer AI startups still raise serious money?
Consumer AI startups can still raise very serious money, but investors currently reserve the biggest checks for products showing unusual engagement or the possibility of creating a new consumer behavior.
AI music is a good example. Suno raised $400 million at a $5.4 billion valuation while facing major copyright lawsuits from music companies. Investors were willing to accept that legal risk because users were already creating and paying for AI-generated music at meaningful scale.
Multimodal creation is attracting money too. Kling AI raised $2.8 billion for AI video generation, making it one of the largest venture rounds announced in July. Meshy raised substantial capital around AI-generated 3D content.
Hardware and new interfaces remain another area investors are willing to explore. Smart glasses, voice devices and other AI-native hardware companies continue to raise money because a successful product could become the interface through which consumers interact with AI all day.
Consumer AI faces a tougher test than enterprise AI, though. A business buyer can justify spending through headcount savings or higher productivity. Consumers can leave an app in seconds when something more entertaining or cheaper appears.
Consumer AI remains very fundable when usage looks exceptional. The middle of the market is much harder: another chatbot, image generator or assistant with no distinctive distribution has little reason to command a large round these days.
Are investors starting to fund AI against labor budgets instead of software budgets?
Increasingly, yes: investors are backing AI companies that can charge for work previously done by people, which gives those startups access to budgets far larger than a normal software subscription.
Traditional SaaS usually sells seats. The customer pays for software that helps an employee work. A growing group of AI companies are trying to sell the completed task itself.
We can see the pattern in customer support, where agents resolve conversations; in coding, where systems review or generate production software; in legal AI, where tools prepare documents and analyze cases; and in back-office automation, where agents handle procurement, finance or administrative tasks.
Pricing is beginning to follow. Silicon Valley Bank’s recent survey of more than 120 venture-backed enterprise software companies found that 37% currently relied exclusively on subscriptions, while only 26% expected to remain subscription-only as usage- and outcome-based pricing spreads.
This changes what an AI company’s market can look like. A legal software startup traditionally competes for legal-software spending. A system that actually performs part of a paralegal’s work can potentially compete for part of the labor budget as well. The same logic applies to call centers, developers, accountants and many other knowledge jobs.
The companies investors seem most excited about today usually have a convincing answer to a simple question: what expensive human work does the customer stop paying for as this product gets better?
If you want more recent data on this point, please see our latest AI infrastructure market report.
This chart, featured in our AI infrastructure market deck, shows the share of revenue generated by each customer segment in the AI infrastructure market
Is AI funding becoming global, or is it still mostly American?
AI startup funding is still overwhelmingly U.S.-centric today, although China has built a serious second ecosystem and Europe is producing more large AI companies than it did a few years ago.
Crunchbase calculated that nearly 88% of global AI startup funding went to U.S.-headquartered companies during the first half of the year. The giant OpenAI and Anthropic rounds amplified that figure, but American dominance also extends through infrastructure, coding, enterprise AI and robotics.
China has become much harder to ignore. AI companies captured more than 60% of all Asian venture funding in the latest full quarter, according to Crunchbase, totaling just over $26 billion. DeepSeek raised about $7.4 billion, StepFun raised $2.5 billion and Moonshot AI later added another $3.5 billion.
Europe is smaller in dollars but increasingly produces companies that can raise at U.S.-style valuations. Sweden has produced Lovable and Legora. The U.K. has generated several frontier labs, including Ineffable Intelligence. European robotics and defense-related AI companies are also appearing in large funding rounds.
The geographic spread is wider at the company level than the funding totals suggest. Crunchbase counted 34 new unicorns in June, including 16 from the U.S., eight from China and companies from the U.K., Germany, India, the Netherlands, Belgium, Canada and Saudi Arabia.
For now, though, the funding hierarchy remains steep. The U.S. dominates the capital market, China has enough scale to build its own frontier ecosystem, and Europe has several strong clusters rather than one unified AI center.
What kind of AI startup is hardest to fund today?
The hardest AI startup to fund today is a generic application sitting on top of someone else’s model with no unusual distribution, proprietary workflow, technical bottleneck or evidence of strong customer usage.
The market itself tells us what investors prefer. Billion-dollar checks are going to frontier labs. Physical AI funding has accelerated sharply. Coding companies are raising at multibillion-dollar valuations. Security startups are producing more than 150 seed rounds. Legal and healthcare companies keep attracting repeat capital.
A generic assistant has to compete with OpenAI, Anthropic, Google, Microsoft and thousands of other startups using many of the same underlying models. When the underlying model improves, customers may get the feature directly from a platform they already use.
Early-stage funding has also become more demanding overall. Crunchbase found that seed rounds have become larger since 2023 while the odds of progressing from seed to Series A have fallen sharply. The median U.S. seed round reached around $3 million last year, roughly three times its 2018 level, but more capital at seed has not produced an easier path to the next round.
Investors can still finance a young AI company with little revenue when the technical ambition is exceptional. They can also finance an application when adoption is exceptional. The uncomfortable place is between those two extremes.
These days, saying “we use AI to make this workflow faster” rarely carries a financing story by itself. Investors want evidence that the company owns something difficult to reproduce.
This chart, included in our AI infrastructure market deck, shows how GPU cloud infrastructure technology has evolved over time
So what’s actually getting funded in AI right now?
Right now, AI money is clustering around frontier intelligence, physical AI, AI infrastructure and applications that can take over expensive pieces of real work.
The biggest dollars still belong to frontier labs. OpenAI, Anthropic, DeepSeek, Safe Superintelligence, Moonshot AI and a growing group of new research labs are raising on a scale that has pushed the entire venture market into record territory.
Below them, infrastructure remains extremely strong. The newest large rounds include $5 billion for Databricks, $136 million for Point2 Technology and hundreds of millions for chip, networking, model-routing and search infrastructure companies. Investors increasingly want the systems that make AI cheaper, faster and easier to run at scale.
Physical AI currently looks like the strongest emerging wave. As seen above, its first-half funding already exceeded the amount invested across 2022 through 2024 combined. Humanoid robotics and robot foundation models also sit near the top of the latest deal-count rankings.
Applications are becoming easier to rank too. Coding is the clearest winner. Customer service follows closely because buyers can calculate the labor savings. Legal and healthcare AI have developed into real vertical markets. AI security is especially interesting at seed, where the number of funded companies is already unusually high.
Consumer AI still gets large checks when usage is exceptional, especially around creation, video and new interfaces. Generic AI apps face a much tougher market.
Across these categories, the same preference keeps showing up. Investors currently pay huge premiums for scarce intelligence, hard infrastructure, physical-world capability and software that can own a valuable piece of work. The closer an AI startup gets to one of those four things, the easier it is to understand why the money is flowing there.
If you want more recent data on this point, please see our latest AI infrastructure market report.
OUR METHODOLOGY
This analysis asks what is actually getting funded in AI right now, rather than treating a record headline total as proof that every part of the market is booming. We broke the question into the main areas competing for AI capital: overall funding concentration, frontier labs, infrastructure, physical AI, coding, agents, customer service, healthcare, legal AI, security, consumer AI, labor-replacement software and geography.
For each area, we compared the freshest evidence available in the material above, including funding volume, deal count, round size, valuation changes, repeat financings and, where it helped, evidence of product adoption or commercial traction. Individual rounds are used to show what a category looks like in practice, while broader datasets carry more weight when we make a market-level claim.
We also separate absolute scale from momentum. A category can attract billions because a few companies are enormous, while a smaller category can be more interesting if deal activity is rising across many startups. That distinction is especially important here because mega-rounds now account for most AI dollars, and broad categories such as physical AI include very large autonomous-vehicle financings alongside robotics, drones and industrial automation.
We did not rank sectors first and then search for evidence to justify the order. The conclusions come from comparing the evidence point by point and then looking for patterns that repeat across different companies and datasets. That is why some sections rely mainly on aggregate market data, while others lean more heavily on repeat financings, customer adoption or pricing behavior.
Key sources include CB Insights’ State of AI Q2 2026, Crunchbase’s H1 2026 global startup investment analysis, Crunchbase’s July 2026 venture funding review, Crunchbase’s analysis of billion-dollar-plus rounds, Crunchbase’s physical-AI funding analysis, and Crunchbase’s AI-security seed analysis.
Company-level funding and traction examples were also anchored in original announcements where available, including Lovable’s $400 million Series C, CodeRabbit’s $143 million Series C, Replit’s $400 million financing, Sierra’s $950 million financing, Legora’s $550 million Series D, OpenRouter’s $113 million Series B, and Exa’s $250 million Series C.
In our AI infrastructure market deck, we identify pain points entrepreneurs should prioritize
Related blog posts
- A complete list of business models in AI infrastructure
- What does the AI infrastructure startup landscape look like today?
