Mecka AI is closing in on a $500 million valuation in a new funding round led by Sequoia Capital, a remarkably fast climb for a startup that announced its Series A just months ago. The company sells training data for robotics, and the speed of this deal tells you everything you need to know about how valuable that’s become.

The AI industry spent the last few years obsessed with text and images. Now the focus is shifting to the physical world, and companies building robots need massive amounts of high-quality data to train their models. Mecka is betting it can be the pick-and-shovel play for that gold rush.

Details on the round are still coming together, but the valuation alone is notable. Two-year-old companies don’t typically command near-unicorn prices unless investors see a clear path to dominance in a market that’s about to explode. Sequoia’s involvement suggests this isn’t just hype. The firm has a track record of spotting infrastructure plays early, and robot training data is looking more like critical infrastructure every quarter.

The timing makes sense. Major AI labs are racing to build models that can control robots in manufacturing, warehousing, and eventually homes. They all need the same thing: diverse, labeled datasets showing robots performing tasks in real-world conditions. Mecka is aggregating that data and selling access, which means it benefits no matter which robotics company or AI lab wins.

Two other deals worth watching

While Mecka’s fundraise is the standout, two other companies are showing different versions of AI industry momentum.

Moonshot AI, the Chinese company behind the Kimi chatbot, is targeting $2 billion in annual revenue. That’s an aggressive number, but there’s some data to back it up. OpenRouter shows Moonshot’s K3 models generating as many as 300 billion tokens per day on its system. Usage has dipped slightly in recent months, but the scale is still massive. If Moonshot can monetize even a fraction of that activity, $2 billion starts to look plausible.

The revenue target also reflects how quickly Chinese AI companies are moving from research projects to real businesses. While U.S. labs have focused on raising capital and building frontier models, Chinese competitors have been finding customers and charging them. Moonshot’s push suggests it sees an opening to lock in enterprise contracts before Western companies catch up in China.

Nscale, meanwhile, is adding former OpenAI executive Fidji Simo to its board ahead of a potential IPO. Simo was the No. 2 exec at OpenAI and led Instacart through its public debut in 2023, so the hire signals Nscale is serious about going public. The company hasn’t announced timing, but board additions like this typically happen six to twelve months before a filing.

Simo’s dual experience at a frontier AI lab and a consumer tech IPO makes her a particularly useful advisor. Nscale will need to convince public market investors that AI infrastructure is a durable business, not just a temporary buildout cycle. That’s a tougher pitch today than it would have been a year ago, when anything AI-adjacent could command a premium. Simo’s credibility could help.

The pattern here

All three of these stories point to the same shift: AI companies are moving from “build cool technology” mode to “prove you can make money” mode. Mecka is raising at a huge valuation because investors believe robot training data will be a real market with real margins. Moonshot is setting revenue targets because Chinese investors want to see profitability, not just user growth. Nscale is preparing for an IPO because private capital is getting more expensive and public markets demand clearer business models.

That’s a healthier dynamic than the last few years, when companies could raise massive rounds based on vibes and benchmarks. The money is still flowing, but it’s flowing to companies that can explain how they’ll capture value, not just create it.