Anthropic's $1.5 Trillion IPO Meets Investor Cold Feet

Anthropic is expected to go public at roughly $1.5 trillion, yet some investors are now reportedly pushing for a valuation closer to $1 trillion after last week's AI selloff. That gap, $500 billion wide, is the clearest signal yet that the market has begun to interrogate the economics behind frontier AI spending. The numbers behind both Anthropic and Meta deserve the same scrutiny.

How fast is Anthropic's R&D really growing?

The company's research and development expense rose from approximately $1.9 billion in the first half of 2025 to roughly $9.7 billion a year later, according to figures cited in a recent deep-dive analysis of the pending listing. That is a fivefold increase in twelve months.

The same analysis calculates that benchmark progress per R&D dollar fell by about 60% even as progress per year doubled, and it almost halved again in the third quarter. Read that carefully. Capability gains are accelerating in absolute terms while becoming dramatically more expensive in relative terms. Both statements are true at once.

This is the efficiency question investors rarely ask aloud. If each incremental unit of model improvement costs several times more than the last, then revenue must grow even faster than R&D just to hold margins flat. The disclosed figures do not yet show that happening.

What does Meta's capex tell us?

Meta plans to spend $135 to $145 billion of capital expenditure this year, which is roughly 3.5 times what it spent in 2024. This is confirmed, budgeted spending, not a roadmap aspiration, and it anchors the entire sector's demand for compute.

The bull case for Meta rests on a simple floor: its core advertising business alone is valued at about 10x 2027 EBIT, per the analysis. That floor exists even if AI bets such as MSL and Muse underperform. Anthropic has no equivalent floor. Its entire value rests on a young, usage-based API business.

Why is Anthropic's revenue less defensible than it looks?

The analysis identifies three specific risks to Anthropic's usage-based API revenue:

• Customer insourcing, as large clients build models in-house
• Rapid price compression across the frontier-model market
• Ease of switching to rival models, which limits pricing power

Each risk attacks the same weak point. Metered API revenue is loyal to the benchmark leaderboard, not to the vendor. When a rival model posts better results at a lower price, switching costs are minimal. Meta's ad business, by contrast, monetizes through entrenched advertiser relationships and distribution, not raw model performance.

There is also a disclosure asymmetry worth naming. OpenAI does not file with the SEC, and Google's AI spending is buried inside a much larger business, so reliable estimates for both are difficult to construct. Anthropic, by going public, will be the first frontier lab to expose its full cost structure to quarterly scrutiny. The market will have one clean data point where it previously had none.

What should investors actually track?

The analysis proposes three concrete metrics, and they are worth repeating because they convert a valuation debate into something measurable:

• The share of compute serving paying customers, rather than internal research
• Revenue per R&D dollar, which directly tests the efficiency problem above
• The proportion of subscription versus metered revenue, since subscriptions are stickier than usage-based billing

If Anthropic's revenue per R&D dollar keeps falling at the recent pace, no valuation premium survives contact with that arithmetic. If the subscription mix rises, the switching-risk argument weakens. These three lines in future filings will say more than any investor presentation.

Where does this leave the valuation debate?

The push from $1.5 trillion toward $1 trillion is not a verdict on AI capability. It is a verdict on unit economics that remain, for now, undisclosed in full. Benchmark progress doubled in a year; the cost of that progress in R&D terms rose far faster, and the third quarter showed the trend worsening rather than stabilizing.

Meta's position is different in kind. Its $135 to $145 billion capex is a bet the ad business can absorb, and at roughly 10x 2027 EBIT for the core alone, the floor holds even under disappointment. Anthropic's public filings, once they arrive, will reveal whether a fivefold R&D escalation bought a durable business or an expensive seat at a price war.