Analysis
Fortune's argument this week is the right one, and I don't think Anthropic can actually meet the bar it implies: Fortune reports that Anthropic needs to eventually deliver Amazon-style earnings power to justify a $2 trillion valuation -- and right now, the company has barely turned a profit. I've been skeptical of the Amazon comparison since it first started circulating, and this week's numbers only sharpen that skepticism.
Here's the actual math nobody wants to say out loud in the investor meetings Krishna Rao is currently running: $47 billion in run-rate revenue against a $2 trillion target valuation is a 43x multiple. Amazon, even during its most aggressive growth-over-profit years in the late 1990s and early 2000s, never traded at anything close to that on revenue -- and Amazon's story eventually worked because Jeff Bezos ruthlessly reinvested cash flow into logistics infrastructure that became a genuine moat, then flipped the profit switch on AWS. Pulse has previously covered Amazon's own AI infrastructure spending and how it stacks up against the labs it backs, including Anthropic. Anthropic hasn't shown the public market any evidence of that second act yet. Run-rate revenue tells you growth is real; it tells you nothing about whether the unit economics of serving that revenue are improving.
“Pulse has previously covered Amazon's own AI infrastructure spending and how it stacks up against the labs it backs, including Anthropic.”
The bull case is straightforward and worth stating honestly: if Claude's enterprise and coding-agent revenue keeps compounding at anything close to its recent pace -- roughly 5x from $10 billion to $47 billion in under six months -- then the multiple compresses fast on its own, and Anthropic doesn't need to prove profitability today, just trajectory. Compute costs are also falling per-token industry-wide as inference gets more efficient, which could improve margins without Anthropic changing anything about its business model.
Room for disagreement: the strongest counter to my skepticism is that public markets have repeatedly paid growth-stage multiples for category-defining infrastructure companies precisely because the alternative -- waiting for profitability proof before investing -- means missing the position entirely. Snowflake, CrowdStrike and Palantir all traded at valuations that looked irrational on trailing revenue and were broadly right in hindsight. If Anthropic is genuinely building the enterprise AI equivalent of AWS, a premium multiple during the land-grab phase is defensible, and I could simply be wrong about the timeline for margin proof.
What I'd actually watch: whether Anthropic's eventual prospectus discloses gross margin by product line -- API revenue versus Claude Code versus enterprise contracts -- because that breakdown, not the top-line run-rate number, is what will tell you whether this is Amazon in 1999 or a much more fragile story wearing Amazon's playbook as a costume.
