The guy who spent years selling OpenAI's vision to enterprises just said the quiet part out loud: AI isn't winner-take-all, it's winner-take-most.

The Summary

  • Zack Kass, former OpenAI Go-To-Market head, told Bloomberg that AI will create "a lot of winners" despite still being early days for investment
  • The timing matters: this is someone who knows how enterprise buyers actually adopt AI, not just how VCs pitch it
  • His broader message to businesses: the competitive advantage window is open wider than the breathless headlines suggest

The Signal

Kass spent his OpenAI years in the trenches with actual companies trying to figure out what to do with GPT-4. Not researchers. Not founders. The people writing checks and managing risk. His read on the market carries weight because he's seen which use cases stick and which ones evaporate after the demo.

The "lot of winners" framing cuts against the dominant narrative that AI is a scale game only the hyperscalers can win. If you've been watching the agent economy develop, you've seen this play out: specialized AI companies building vertical solutions are finding product-market fit faster than the foundation model labs expected. Kass is validating what the data already shows.

What's interesting is the venue. He's not saying this at a startup conference or on a podcast. He's saying it at the Citi Macro Conference in Hong Kong, speaking to institutional capital allocators trying to model out where value accrues in the AI stack. That audience doesn't care about vibes. They care about durability and margin structure.

The Implication

If you're building in AI and feeling squeezed by foundation model gravity, this is signal to keep going. The distribution advantages and domain expertise you're building matter more than the model someone else trained. For investors, the message is clear: stop trying to pick the one winner and start mapping the actual adoption curve. The companies making "a lot of winners" possible are the ones solving workflow integration, not just API access.

Source: Bloomberg Tech