Key Takeaways

  • Hyperscalers like Meta are spending hundreds of billions on AI compute annually, creating a modern “AWS problem” of idle capacity that must be monetized.
  • This economic pressure forces tech giants to enter the enterprise AI market, selling their surplus compute and agent-building capabilities to businesses.
  • Elon Musk's "EWS" (Elon Web Services) with Anthropic demonstrates a prime example of turning excess GPUs into revenue by efficiently converting "electrons into tokens."
  • Non-software-native companies, like law firm Kirkland & Ellis, attempting to build half-billion-dollar custom AI software are highly unlikely to succeed against dedicated AI players.
  • For enterprise AI adoption, the smart money is on acquisitions by AI-native firms (like Thrive Holdings) that can rapidly integrate and turbocharge existing businesses with specialized AI expertise.

The AWS Problem, Reborn for AI

Imagine spending a hundred billion dollars on something every year. That's the kind of capital expenditure hyperscalers are laying out for AI compute. Brad Gerstner pulls back the curtain, explaining that this massive spending leads directly to what he calls the “AWS problem.” It's simple: you build capacity for peak demand, but the rest of the time, half of that expensive hardware sits idle.

Gerstner points out that Amazon built AWS because Jeff Bezos had to plan for Christmas week, knowing the rest of the year a huge chunk of his servers would just be waiting. “The second you start spending a hundred billion dollars on capex annually, okay, you run into the AWS problem,” Gerstner says. “Now, I have all this compute, but I don't use it every day equally.” This isn't just an inefficiency; it’s an existential pressure. Hyperscalers must find ways to rent out that idle AI capacity, pushing them squarely into the enterprise AI market.

Elon's 'EWS': A Blueprint for Monetization

Want to see this pressure in action? Look at Elon Musk. Gerstner highlights Musk's move to create "EWS" (Elon Web Services) with Anthropic as a textbook case. Musk's core strength isn't just building rockets or cars, it’s managing vast, complex infrastructure. According to Gerstner, “nobody on earth is better at turning electrons into tokens than Elon.” Musk built significant compute capacity, then found a strategic partner in Anthropic to consume and monetize that surplus.

This isn't just about selling raw compute power; it's about enabling enterprise-grade AI applications. For companies sitting on massive GPU farms, selling access to models and specialized agents built on that hardware becomes a core business strategy. It’s an imperative born from balance sheets and capital allocation, not just a desire for new revenue streams. Companies like Meta, traditionally consumer-focused, face this exact dilemma: transform their compute into a profit center for enterprise, or risk crippling write-offs.

Why Your Custom AI Is Probably a Bad Bet

For enterprise founders, this shift changes everything about how you approach AI. Gerstner expresses serious skepticism about non-software-native companies trying to build their own AI software from scratch. He questions whether a law firm like Kirkland & Ellis, which specializes in legal briefs, can suddenly develop “killer legal software to compete with OpenAI and Anthropic.” His answer is clear: “I think that's unlikely.”

Instead, Gerstner points to acquisitions by AI-native firms, like Thrive Holdings, as the far more probable and effective path for enterprise AI adoption. These firms aren't just buying technology; they're buying specialized expertise and a track record in a domain where the learning curve is steep. This suggests that for most businesses, building a half-billion-dollar in-house AI solution is a fool's errand. The winners will be those who either buy existing AI capabilities or build deep integrations on top of hyperscaler offerings.

What to Do With This

Stop trying to build your own foundational AI infrastructure. Instead, identify which hyperscaler's AI service (e.g., Meta's, AWS's, or even a nascent 'EWS' equivalent) aligns best with your specific enterprise needs. Then, focus your team on building hyper-specialized agents and deep integrations on top of those platforms. For existing non-tech enterprises, abandon costly internal AI R&D and proactively scout AI-native startups for acquisition targets that can immediately turbocharge your core business.