Raise a big round. Announce a partnership. Get mentioned in TechCrunch. Half the companies on “fastest growing” lists are growing headcount, not revenue. Or growing revenue from one enterprise pilot that might not renew.
I’m not doing that kind of list.
The companies below have real customers. Real revenue. Products people keep using after the first month. That’s a much shorter list than most people want to admit.
See more technology coverage at Futurelume
Before the List — What I Mean by “Growing”
Revenue doubling year over year with decent retention. Customer counts expanding without acquisition costs going insane. Products that get stickier over time, not just bigger.
Not: valuation bumps. Not: headcount. Not: press release partnerships. Not: ARR that includes pilots nobody’s actually paid for.
The companies below pass the first test. Most on other lists pass only the second.
Mistral AI
A year ago Mistral was interesting. Now it’s a real business.
Why It’s Growing
The open-weight model strategy worked better than most expected. Releasing capable models openly built a real developer base fast. Enterprise customers who can’t route data through US infrastructure now have a credible alternative. The API business is growing. The enterprise contracts are landing.
What Makes It Hard to Copy
European data sovereignty rules aren’t going away. Mistral is the default answer when a company says “we need a capable model that doesn’t touch American servers.” That’s a structural tailwind, not a marketing position.
What I’m Watching
The capability gap with OpenAI and Anthropic at the frontier is real. Mistral wins on compliance. It needs to keep closing the performance gap to win on everything else.
Perplexity
Perplexity did something most AI products haven’t managed: built a daily habit.
Why It’s Growing
The answer engine format — cite your sources, give a direct answer, let people dig deeper — turns out to be genuinely useful for how people actually search. Daily active users are growing. The enterprise tier is landing. Word of mouth is real because the product visibly improves.
What Makes It Hard to Copy
People who switch their default search behavior to Perplexity tend to stay. Habit formation is the retention metric that matters most for a daily-use product. That’s hard to buy your way into.
What I’m Watching
Google is not sitting still. When the incumbent owns the distribution channel and has unlimited resources, the quality gap needs to stay wide. That’s a lot of pressure to maintain.
Harvey
Harvey did the hard thing most vertical AI companies avoid: they went deep into one domain instead of wide across many.
Why It’s Growing
Law firms pay enterprise prices for AI that actually understands legal language and legal workflow. Harvey didn’t try to also serve healthcare and finance and construction. They focused. The revenue per customer is high. The retention is high. Referrals inside the legal industry — where reputation travels fast — are strong.
What Makes It Hard to Copy
Domain-specific training data and workflow integration that a general-purpose tool can’t replicate without the same years of investment. The moat gets wider over time as more legal data gets processed.
What I’m Watching
Every major law firm software vendor is now building AI features. Embedded competition from existing workflow tools is the thing to watch closely.
Figure
Figure is on this list not because humanoid robots are everywhere — they’re not — but because the revenue from enterprise pilots is real in a way it wasn’t eighteen months ago.
Why It’s Growing
BMW and several logistics operators are running real deployments. Not demos. Real production environments. The units are working. The cost per unit is coming down. The feedback loop between field deployment and development is tightening faster than expected.
What Makes It Hard to Copy
Hardware plus software is a much harder moat than software alone. The physical world data — how robots actually interact with real environments — compounds in ways that pure software training data doesn’t.
What I’m Watching
The jump from successful pilots to scaled deployment is where physical AI companies have historically stalled. Manufacturing capacity, reliability at scale, and maintenance economics are all still unproven at the volumes the numbers need.
ElevenLabs
ElevenLabs started as a voice cloning tool. It’s quietly becoming infrastructure.
Why It’s Growing
The API is embedded in products across media, gaming, education, and enterprise. When a company needs to add voice to their product, ElevenLabs is often where they land first. Infrastructure-level positioning is hard to dislodge once it’s in the stack.
What Makes It Hard to Copy
The voice quality gap between ElevenLabs and open-source alternatives is still meaningful. Developer experience is strong. Being the default in a category creates compounding network effects that are hard to break.
What I’m Watching
The major model providers are coming for this space. Quality leadership in AI doesn’t last forever. The window to entrench the infrastructure position is narrowing.
The Pattern Across All Five
| Company | Domain | Why It’s Hard to Copy |
| Mistral | Foundation models | Geography + compliance tailwind |
| Perplexity | Search | Daily habit + strong retention |
| Harvey | Legal AI | Domain depth + specialized data |
| Figure | Physical AI | Hardware + real-world embodiment data |
| ElevenLabs | Voice AI | Quality lead + API infrastructure |
None of them are trying to be everything. Each picked a specific problem, went deep, and built something genuinely hard to replicate quickly. That’s the pattern.
Who’s Not on This List and Why
A few categories conspicuously absent.
General-purpose AI assistants competing directly with ChatGPT — the market is dominated and the differentiation is basically nonexistent for most players. Horizontal AI platforms promising to automate everything — the go-to-market is too diffuse to build real retention. AI content tools — margins have collapsed and commoditization is nearly complete.
Fast growth in these categories is real. So is the risk. Switching costs are low, incumbents are better resourced, and unit economics are brutal.
The fastest growing AI companies worth watching in 2026 aren’t the ones with the biggest funding announcements.
They’re the ones with a clear answer to “why can’t someone else just do this tomorrow?”
Domain depth. Structural tailwinds. Products that get harder to leave over time.
That’s what real growth looks like. Everything else is just noise with a good PR team behind it.