The $47 Billion Question: Why AI Companies Are Still Betting on Growth Nobody Can Prove Yet

I was sitting in a client meeting last week when someone asked me the question that keeps me up at night: "So we've spent six figures on AI tooling this year. What exactly did it save us?" The answer I gave was honest and uncomfortable: "I'm not entirely sure."

That conversation is happening in a thousand conference rooms right now, and it's exactly why Anthropic's push toward an IPO at a nearly trillion-dollar valuation feels both aggressive and terrifyingly fragile. The company is about to go public on a bet that corporate AI spending will eventually pay off—but right now, that's largely a belief, not a proven track record.

The Capital Treadmill Nobody Talks About

Reading Daniela Amodei's justification for the IPO hit me differently than the usual startup fundraising noise. She's not talking about disruption or market dominance. She's talking about physics: training frontier AI models costs obscene amounts of money. Serving those models at scale costs even more.

That's actually honest. But it's also circular logic presented as strategy.

Here's what I see happening: Anthropic burned through billions in the private markets, so they need public markets to keep the machine running. They're not going public because the unit economics proved themselves. They're going public because the burn rate demanded it. OpenAI, xAI—everyone in this space has the same problem. You can't scale inference costs without exponential capital, and private investors only have so much dry powder.

The $1.25 billion per month partnership with xAI tells you everything. That's not a partnership—that's dependency management dressed up as a strategic relationship.

The Use Case Problem That Still Isn't Solved

What bothered me most about Amodei's statements was the vagueness around actual ROI. She mentioned coding, financial services, legal, healthcare. As a developer, I'll confirm: coding is genuinely helpful. I use Claude regularly, and it accelerates specific tasks. But nobody—and I mean nobody—is seeing a 4x productivity multiplier across their entire codebase.

The efficiency gains we're seeing are real but narrow. We're good at assistance for boilerplate, documentation, and refactoring. We're not good at architectural decisions or complex debugging. And we're definitely not good at replacing junior engineers, despite what the hype suggests.

Amodei's hope that AI will eventually be "more incorporated into the day-to-day of how humans do our work" sounds reasonable. It's also doing a lot of heavy lifting. That's not a business plan—that's a wish.

Why I'm Skeptical of Their Approach

The deliberate choice not to build their own data centers actually concerns me more than it should reassure me. Amodei frames it as prudence—avoiding overextension. But it feels like the opposite: avoiding the cost transparency that comes with vertical integration.

If you own the compute infrastructure, you can't hide capital allocation failures. Every unused chip is a visible loss. By outsourcing to xAI and others, Anthropic gets flexibility but also deniability about utilization rates.

From a pure engineering standpoint, this is backward. You want tight feedback loops between product demand and compute provision. Buying compute month-to-month from a partner gives you neither the leverage nor the visibility.

What Actually Needs to Happen

For this IPO to make sense long-term, companies like Anthropic need to move from "AI is valuable" to "AI delivers X% productivity gain and costs Y% of current spend." That's not happening yet.

As developers, what we should actually be watching is whether the market forces a reckoning. Uber's admission that not all their AI spending was productive is the real story here. That's a signal that someone's actually measuring ROI instead of just assuming it.

If I were building a product right now, I'd be ruthless: integrate AI where you can measure the impact, skip it everywhere else. The companies that survive this cycle won't be the ones with the most compute access. They'll be the ones who can actually tell you what their AI spending returns.

The Real Question

Before you celebrate or dismiss Anthropic's IPO plans, ask yourself: what's the first AI investment you'd cut if you had to prove its ROI tomorrow? And if you can't answer that, maybe the real doubt isn't about AI's returns—it's about whether anyone's actually measuring them.

Source: This post was inspired by "Ahead of its IPO, Anthropic's Daniela Amodei shrugs off doubts about AI's returns" by TechCrunch. Read the original article