The most promising private AI companies shaping what comes after the hype.
Curated by Forbes · Annotated by Mike Litman
The Forbes AI 50 selects for momentum, not just funding. Private companies only. Revenue traction, technical credibility, and talent density all factor in. By the time you see a company on this list, the bet is already made – the question is whether you understood why.
This talk maps the 50, groups them by what they actually do, and pulls out the signals hiding in plain sight.
The biggest AI operations in the world – Microsoft, Google, Meta, Amazon, xAI – are all invisible here. Forbes only covers private companies. So what you’re actually looking at is the AI landscape that exists outside Big Tech’s gravitational pull. Fifty companies building without a $3 trillion market cap to fall back on.
That’s a very different kind of bet. And it’s the reason this list is more interesting than it first appears.
Nine companies on this list exist purely to make AI possible or faster. Anthropic and OpenAI are the model factories. Mistral AI and Cohere serve the enterprise and open-source flanks. The inference layer – SambaNova, Fireworks AI, Baseten, Together AI, Crusoe – exists because the frontier models are expensive and slow, and there’s a multi-billion pound arbitrage in making them not be.
Image, video, voice, music – the full creative stack is present on this list. Midjourney, Black Forest Labs, and Krea.ai own still imagery. Runway, HeyGen, and Synthesia own motion. ElevenLabs owns voice. Suno owns music. Ten companies. One conclusion: there is no media format left that doesn’t have a credible AI-native alternative.
Cursor, Replit, Lovable, and Cognition are each betting that the role of the software engineer transforms in the next three years. Meanwhile, Glean, Databricks, Clay, Perplexity, and Notion are fighting for the knowledge worker’s daily workflow. These aren’t AI features bolted onto existing tools – they’re rewrites of the tools themselves.
Law has Harvey and Legora. Medicine has Abridge, OpenEvidence, and Chai Discovery. Finance has Rogo. Hiring has Mercor. Real estate has EliseAI. Customer service has Decagon and Sierra. Language learning has Speak. Security has Cyera. Twelve companies, each making the same bet: that every regulated, expertise-dense profession is worth rebuilding from scratch around AI.
This is the category that wasn’t here two years ago. Physical Intelligence and Skild AI are building foundation models for robots – the same paradigm shift that GPT-3 was for text, applied to arms and fingers. Applied Intuition is hardening autonomous vehicles. World Labs, Fei-Fei Li’s company, is building spatial intelligence. And then there’s Safe Superintelligence Inc. – Ilya Sutskever’s post-OpenAI bet that the only thing worth building is the safe version of what comes next.
The overwhelming majority of this list is based in the United States – concentrated in San Francisco, New York, and a handful of US tech hubs. Mistral AI is the most prominent European entrant, built in Paris as a deliberate argument that frontier AI doesn’t have to come from California. Synthesia has UK roots. But the capital flows, the talent pipelines, and the model infrastructure are all pointing in the same direction.
If you’re building or buying AI outside the United States, this list is mostly a map of what’s coming to you – not what you built.
When the application layer commoditises, what’s the moat?
The companies on this list are building fast. But the models they rely on are getting cheaper every quarter, and the tools they’re building are getting easier to clone. The ones that survive will have data moats, distribution moats, or regulatory moats – not just better prompts. The 2028 list will tell us who figured that out.
mikelitman.me · hello@mikelitman.me