DEV Community

Md Kawsar
Md Kawsar

Posted on

The 10 most funded AI companies right now and what they're actually building

In H1 2026, two companies raised more than the entire US VC market typically deploys in a year. Funding headlines reveal very little about the bets behind them, though. Here's what each company is doing with its capital, and what the numbers leave unexplained.

The H1 2026 figures are hard to wrap your head around. According to Crunchbase, global startup investment reached $510 billion in the first six months of the year. OpenAI and Anthropic alone took in $217 billion of that, or 43 percent of all startup funding worldwide. And Q1's $297 billion, raised in a single quarter, was more than ten times the quarterly peak of the dot-com boom.

Coverage usually stops at the valuation and the latest record-setting round. Far less attention goes to what each company is doing with the money, which of its bets are working, and which ones stay open questions no matter how many zeros the check carries.

This post walks through the ten most funded AI companies by disclosed capital. For each one, it explains the products and strategy behind the money and names the honest question that even the biggest round can't answer.

The foundation model labs

1. OpenAI : approximately $180 billion raised, $852 billion valuation

OpenAI describes its goal as artificial general intelligence, but in 2026 what it actually operates is closer to a diversified software company than a research lab. ChatGPT passed 200 million monthly active users, the fastest any consumer application has ever reached that mark. The API processes more than 15 billion tokens a minute. Enterprise accounts make up 40 percent of revenue and are closing in on parity with consumer.

The latest capital is paying for a wide slate: GPT-5 and its successors, the Codex CLI and Code Interpreter products aimed at developer workflows, Sora for video generation (up against Runway and Google's Veo), the Operator line for autonomous agent workflows, and OpenAI for Government, a dedicated enterprise track for federal agencies. The company has also completed seventeen acquisitions in three years, six of them in 2026 alone. That adds up to the product breadth you'd expect from a publicly listed technology company.

An IPO is targeted for September 2026 at roughly $1 trillion, which would make it the largest in history if the timeline holds. Revenue is about $2 billion a month and rising, yet the company doesn't expect profitability until 2029-2030. The open question: what multiple of current or projected revenue does a $1 trillion public valuation price in, and what growth rate would it take to justify that to public-market investors?

2. Anthropic : approximately $125 billion raised, $965 billion valuation

After its $65 billion Series H closed in late May 2026, Anthropic became the most valuable standalone AI startup in the world, overtaking OpenAI's latest private-round valuation. Run-rate revenue passed $47 billion in May 2026, and the company projected its first operating profit, around $559 million, for Q2 2026. For a company that spent its first three years pouring capital into research infrastructure, that revenue trajectory is the most important signal in its recent history.

Anthropic's work breaks into three tracks. The first is the Claude model family (Claude 4.6 Sonnet and Opus as of mid-2026), which most users encounter directly. Claude Code has become a breakout product by itself, passing $2.5 billion in run-rate with more than 500 enterprise customers each spending over $100,000 a year. The second is the Model Context Protocol (MCP), which Anthropic open-sourced and donated to the Linux Foundation as an attempt to set the connectivity standard for how AI models reach external tools and data. If MCP becomes the infrastructure layer for AI integrations, Anthropic gains considerable indirect influence over how the whole category evolves. The third is Constitutional AI and interpretability research, a safety methodology meant to make model behavior explainable and adjustable. It functions both as a research program and as a selling point for regulated enterprise customers.

Confidential IPO paperwork was filed on June 1, 2026, with a listing possible as early as October 2026. The open question: can revenue growth sustain the valuation multiple that a debut priced above $1 trillion would demand, and does the safety positioning earn a lasting price premium over OpenAI, or does it eventually shrink to commodity model pricing?

3. xAI : approximately $42.7 billion raised, valuing the combined xAI/SpaceX entity at $1.75 trillion at IPO

xAI is the hardest company here to assess, because its funding, products, and corporate structure are all shifting at the same moment. In early 2026, xAI effectively folded its strategic interests into SpaceX. The SpaceX IPO, priced on June 11 under SPCX on Nasdaq at a targeted $1.75 trillion valuation, therefore gives public investors a single vehicle with exposure to both the aerospace business and the AI lab.

The product side has several pieces. The Grok model family improved quickly through 2025 and now competes with GPT-5 and Claude on several reasoning benchmarks. The Memphis supercluster, possibly the largest single-site AI compute cluster in the world, frees xAI from dependence on cloud providers. Close ties to X (formerly Twitter) give Grok access to real-time public conversation that other labs can't match, a real training-data edge for recency-sensitive tasks. And Aurora OS, xAI's operating environment, is the early product layer of an AI-native computing experience that goes beyond chat.

The open question: xAI hasn't disclosed its run-rate revenue, so its roughly $200 billion valuation rests on compute infrastructure and model quality, not on documented business metrics. Its enterprise sales motion is also less developed than OpenAI's or Anthropic's, and Grok usage outside the X ecosystem has been limited. Revenue transparency will be the first test that public markets apply to the xAI thesis.

Infrastructure and data platforms

4. CoreWeave : approximately $12 billion raised pre-IPO, now public (CRWV)

CoreWeave went public in March 2025, which makes it one of the few AI companies whose equity ordinary investors can actually buy. Its business is GPU cloud infrastructure: it rents NVIDIA GPU capacity to companies that build or run AI models, at a scale and level of specialization that general-purpose providers like AWS, Azure, and Google Cloud don't match for pure AI workloads.

A $20 billion-plus compute deal with OpenAI, running through 2028, is the anchor contract that gives CoreWeave near-term revenue visibility. NVIDIA's early strategic investment, and its role as primary hardware supplier, cuts both ways: it brings priority chip allocation during GPU shortages, but it also concentrates risk in a single vendor. The underlying thesis is that demand for AI inference capacity will outgrow what the major clouds can profitably serve, leaving a durable niche for focused infrastructure. The Cerebras IPO (CBRS, May 2026, $95 billion market cap) shows that public markets will pay significant multiples for AI infrastructure, which supports CoreWeave's position.

The open question: if the major cloud providers decide to subsidize GPU pricing to win AI workloads at scale, CoreWeave's pricing power gets shaky. The OpenAI contract offers short-term protection, but what will the customer base look like when it comes up for renewal?

5. Databricks : approximately $20 billion raised, $134 billion valuation, H2 2026 IPO target

Databricks is the least glamorous name on this list, and arguably the most important to how enterprise AI actually gets deployed at scale. It pioneered the data lakehouse architecture, which gives organizations one layer for storing, processing, and analyzing the data that feeds AI models. Having two product lines that each clear $1 billion in ARR is unusual at this stage.

In 2026, the company's ambitions extend past the data platform. DBRX, its open-source large language model, is pitched as the enterprise-safe option for organizations that want AI on their own infrastructure without sending data to OpenAI or Anthropic. Governance and compliance tooling built into the platform answers the top concern of every regulated industry weighing AI adoption. And Unity Catalog, a metadata and governance layer, tracks what data models were trained on and what decisions they made, the foundation that compliance-ready AI in healthcare, financial services, and government requires.

An H2 2026 IPO at a $134 billion target would test whether public markets give enterprise data infrastructure AI-level multiples or treat it like traditional SaaS. The open question: a public Databricks at $134 billion would face real pressure next to Snowflake, which carries less than half that valuation on comparable revenue. How those two companies get priced against each other will shape how enterprise data infrastructure is valued in public markets.

6. Scale AI : $13.8 billion valuation, including Meta's $14.3 billion for a 49 percent stake

Scale AI is in an unusual position in 2026. Meta paid $14.3 billion for a 49 percent non-voting stake, a deal often mislabeled as an acquisition. Scale remains independent. Its 28-year-old founder, Alexandr Wang, left to run Meta's Superintelligence Labs, so the company is going through a leadership transition at the very moment it's worth more, and is more complex, than ever before.

The business itself is data labeling, human feedback collection, and model evaluation, the infrastructure AI models rely on for quality. Every major AI company's RLHF (reinforcement learning from human feedback) pipeline either runs on Scale's services or competes with them. Government contracts for AI in national security have grown into a significant revenue line, with Scale's FedScale division working on defense and intelligence applications. Scale is also building AI evaluation frameworks (Inspect, SEAL, and related tools) that are becoming industry standards for measuring model performance.

The open question: synthetic data generation is improving fast and threatens the labeling business at the commodity end of the market. Whether Scale can climb quickly enough toward evaluation and governance, where synthetic data can't easily substitute, will decide whether the Meta deal looks prescient or overpriced in three years.

Physical AI crossing into commercial scale

7. Waymo : approximately $35 billion total raised (including Alphabet backing), $126 billion valuation

Waymo completed 15 million paid autonomous trips in 2025, triple the year before, and now operates commercially in six US cities. Its February 2026 round of $16 billion at a $126 billion valuation was the largest autonomous vehicle financing ever, backed by Lightspeed, SoftBank, Amazon, Jeff Bezos, and Coatue. The money is explicitly tied to global expansion: more than 20 new US cities and a first international launch in London are the 2026 targets.

It's more accurate to see Waymo as a safety data business than a taxi business. Its vehicles have been documented at 82 to 90 percent fewer serious-injury crashes than human drivers, and that safety record is an asset that takes decades to build and can't be bought or synthesized. Each trip adds to a behavioral dataset that makes the next generation of models safer and more capable, a compounding advantage fundamentally different from software network effects. And the unit economics are coming into view: at 15 million trips a year across six cities, the revenue and cost structure is no longer theoretical.

Goldman Sachs projects the US robotaxi market at $48 billion by 2035, and Waymo's expansion pace and safety record put it in position to capture a dominant share. The open question: Tesla's Cybercab launch brings a credible competitor with lower per-unit hardware costs. Does Waymo's safety lead outweigh Tesla's manufacturing scale? That's the central question for the category.

8. Figure AI : approximately $2.6 billion raised, $48 billion valuation

Figure AI has production humanoid robots working on BMW manufacturing lines, which sets it apart from every other humanoid robotics company still stuck in demo mode. Its target market is the $1.5 trillion a year that global physical labor shortages cost, a number big enough to justify large capital rounds, provided the robots perform reliably at industrial scale over time.

One critical difference separates Figure from the robotic automation of the past, which relied on specialized machines built for specific tasks. Figure's robots are trained on general-purpose learning systems that let them pick up new tasks through demonstration rather than reprogramming. The BMW deployment is the first serious production test of that approach. OpenAI's software partnership with Figure also gives the robots access to frontier-model reasoning, linking the physical AI layer to the foundation model layer in a combination neither company could build alone.

The open question: Tesla's Optimus program has manufacturing-scale advantages that Figure, as a purpose-built robotics company, can't currently match. If Tesla vertically integrates robot production at the scale it has mastered in cars, Figure's per-unit economics come under structural pressure that capital alone can't relieve.

The long bets

9. Safe Superintelligence (SSI) : $3 billion raised

Ilya Sutskever, who co-founded OpenAI and served as its chief scientist through the most consequential stretch of its development, co-founded SSI. Its stated mission is to build AI that stays reliably safe at the scale of artificial general intelligence, free of the commercial pressure of a product cycle. As of mid-2026, SSI has no products, no announced timeline, and no disclosed research output. The full $3 billion was raised on the strength of one thesis: that this team can do what it says it's trying to do.

That makes SSI the toughest company on the list to judge with any conventional investment or business framework. The case for it is that safety alignment is the most important unsolved problem in AI, and that a team with Sutskever's research background, insulated from product pressure, is better placed to solve it than safety groups inside product-driven companies. The case against is that three billion dollars with no output and no timeline is an enormous bet on a thesis with no near-term falsifiable predictions attached.

The signal to watch in H2 2026 is whether SSI publishes research. A company that ships no products and publishes no technical work is functionally invisible to anyone trying to judge whether the investment thesis is on track. Output is the only data point that matters here.

10. Anduril Industries : approximately $9.8 billion raised, $28 billion valuation

Anduril is the outlier of the group. It isn't a foundation model company, an infrastructure play, or a consumer product. It builds AI-native defense systems for the United States and allied governments. The $5 billion it raised in May 2026 ranks among the largest defense technology financings ever, and reflects a structural shift in how the US government buys defense technology.

The core product is Lattice OS, a command-and-control operating system that brings autonomous surveillance, drones, interceptor weapons, and sensor networks together under one AI-driven interface. What distinguishes Anduril from traditional defense contractors is the order of operations: software first, hardware second, rather than the other way around. The Roadrunner autonomous interceptor, the Altius drone family, and the Sentry tower system all exist because the software called for them. That inversion gives Anduril faster iteration cycles than hardware-first competitors.

The open question goes beyond business metrics. All of Anduril's revenue comes from governments using AI systems to make decisions with lethal consequences. The ethical framework for this work, the standards of accountability when systems fail, and the geopolitical implications of making AI-native weapons more widely available don't appear on any cap table. They are still real, and anyone evaluating Anduril should weigh them alongside the financial projections.

What the funding map actually shows

Looking across all ten companies, a few structural patterns emerge that don't belong to any single company's story:

The capital structure has changed category. The sovereign wealth funds writing the biggest checks (GIC in Anthropic, Temasek in OpenAI, QIA in Anthropic) are infrastructure investors by instinct. They're treating frontier AI labs the way earlier sovereign funds treated energy infrastructure. That isn't a valuation argument. It's an observation about which kind of investor sees AI infrastructure as a generational bet.
Revenue backs the top two, not the top three. OpenAI ($2 billion a month) and Anthropic ($47 billion ARR run-rate) have disclosed revenue that at least partly explains their valuations. xAI's valuation rests mainly on claims about compute infrastructure and model quality, with no disclosed revenue as of July 2026.
The IPO queue is long and will be a real test. OpenAI (September 2026 target), Anthropic (October 2026 possible), Databricks (H2 2026), and Cohere (H2 2026 target) are all preparing to list in the same window. If all four go ahead, the combined float could top $3 trillion. Whether public markets absorb that at the expected valuations will be the most important data point of the AI investment cycle.
Physical AI is the most capital-efficient story. Waymo's 15 million trips and Figure's BMW deployment show real revenue and real operations on a fraction of the capital the foundation model labs have consumed. The comparison is imperfect, since the underlying technology needs different infrastructure, but the road to unit economics is shorter and easier to see.
The most useful question about any of these companies isn't how much they've raised. It's this: what specific, falsifiable claim does the funding let them test? That answer says more than the valuation figure ever will.

The question worth sitting with

Across these ten companies, total capital raised falls somewhere between $430 and $450 billion, depending on how you count strategic commitments versus equity rounds and how you treat xAI's merger with SpaceX. The figure is imprecise. The accounting is complicated, and responsible sources point out the methodological gaps in comparing these numbers.

A 2025 MIT study found that 95 percent of enterprise generative AI pilots produced no measurable profit impact. That doesn't mean AI isn't real or valuable. It does mean the path from frontier model capability to enterprise profit is longer, messier, and less linear than funding headlines suggest. The ten companies above embody the bets that the best-informed, best-resourced investors in the world have placed on how that path resolves.

The money isn't wrong about how important AI is. It may be wrong about the timeline, the specific winners, and whether today's capital structure produces sustainable businesses or just an extremely expensive discovery process. Both can be true at once.
Every one of these companies will report some version of success in H2 2026: revenue growth, partnerships, product launches, or IPO filings. The announcements aren't the real test, though. What matters is whether the economics underneath hold up across several quarters, in markets that are price-sensitive to AI value, against competitors who are just as well funded and just as determined to win the same customers.

That story is still being written. The funding is only the setup.

Enjoyed this read? Be sure to follow me for more stories that make people feel things. You can also connect with me on LinkedIn or Telegram @kawsarlog for more updates on my work.

I do freelance scraping projects, so if you need some data collected feel free to contact me on LinkedIn or email at hi@kawsarlog.com

Top comments (0)