Anthropic's $1.5 Trillion IPO Math Meets a 60% Efficiency Decline

Anthropic is expected to go public at a valuation of roughly $1.5 trillion, yet some investors are now reportedly pushing for a figure closer to $1 trillion after last week's AI selloff. That half-trillion-dollar gap, laid out in a November 11, 2026 analysis by MBI Deep Dives under the title "Anthropic Vs. Meta: Room To Be Wrong," is the sharpest public articulation yet of the central question hanging over AI equities: whether current spending levels can produce durable, defensible businesses.

The analysis, published by an author with sell-side and US long-only buy-side experience plus MBA, CFA, and FRM credentials, frames the debate through two companies with radically different risk profiles. Meta, trading under the ticker META, has an earnings floor. Anthropic, expected to list as Anthropic, PBC under the ticker ANT, does not — at least not on the same terms.

How much is Meta actually spending?

Meta plans to spend $135 to $145 billion of capital expenditure this year, a figure the analysis notes is approximately 3.5 times what the company spent in 2024. The scale matters. Meta is pouring that money into AI infrastructure while carrying an advertising business that the analysis values at about 10x 2027 EBIT (earnings before interest, taxes, depreciation, and amortization) on a standalone basis.

That ads valuation provides what the analysis calls a strong earnings floor — meaning that even if Meta's speculative AI bets, identified in the piece as MSL and Muse, underperform, the core business alone supports the equity. The downside case for Meta is cushioned. The analysis's framing is direct: Meta has room to be wrong.

Anthropic's position differs. As a pre-IPO company whose financials are only now entering public scrutiny, its valuation rests almost entirely on the trajectory of its AI products, not on a legacy cash machine underneath them.

Why are OpenAI and Google absent from the comparison?

The analysis confines its head-to-head to Anthropic and Meta for a practical reason: comparable data does not exist for the other two US AI heavyweights. OpenAI does not file with the SEC (the US Securities and Exchange Commission), so its financials remain private. Google's AI spending is buried inside a much larger business at Alphabet, making it difficult to isolate the AI investment line from search, YouTube, and cloud.

The result is an unusual market structure. Investors can scrutinize Meta's AI spending through public filings and, assuming the IPO proceeds, Anthropic's numbers through its registration documents — but the two other major AI spenders remain opaque. The analysis treats this opacity as a reason to focus on the two companies where the arithmetic can actually be checked.

What happened to Anthropic's R&D efficiency?

The most striking figure in the analysis concerns Anthropic's research and development spending. Anthropic's R&D expense rose from approximately $1.9 billion in the first half of 2025 to approximately $9.7 billion a year later — a more than fivefold increase in twelve months.

The output side of that equation moved in the opposite direction on a per-dollar basis. According to the analysis, benchmark progress per R&D dollar fell by roughly 60%, even as raw progress per year doubled. The trend then compounded: efficiency almost halved again in the third quarter.

The implication is that Anthropic is buying capability gains with dramatically escalating capital, not improving returns on each dollar deployed. For a company expected to command a valuation between $1 trillion and $1.5 trillion, that efficiency curve is the number investors will interrogate. If benchmark progress per dollar continues to halve on a quarterly cadence, the cost of maintaining frontier-model leadership compounds at a pace that only the largest balance sheets can absorb.

How defensible is Anthropic's revenue?

The analysis raises a second structural concern: the defensibility of Anthropic's usage-based API business. It identifies three specific risks:

• Customer insourcing: large customers can build their own model capabilities rather than pay Anthropic per token.
• Rapid price compression: prices for model access are falling quickly across the industry.
• Ease of switching to rival models: customers face low friction when moving to competing models, which weakens Anthropic's pricing power.

Together, these factors make Anthropic's revenue stream less defensible than a subscription business or an advertising franchise, in the analysis's assessment. A metered API (a pricing model where customers pay based on how much compute they consume) depends on customers continuously choosing to rent capability rather than own it — a choice that price compression and easy substitution make harder to bank on year over year.

The contrast with Meta is deliberate. Meta's advertising revenue comes from an entrenched auction system with deep advertiser integration. Anthropic's revenue comes from an API that a customer can, in the analysis's framing, walk away from with minimal switching cost.

What should investors track?

The analysis closes with a concrete monitoring framework rather than a buy or sell call. It names three metrics investors should follow to assess the sustainability and business quality of AI companies like Anthropic:

• The share of compute serving paying customers — a measure of whether expensive infrastructure is monetized or idle.
• Revenue per R&D dollar — the efficiency ratio whose recent decline anchors the analysis's caution.
• The proportion of subscription versus metered revenue — a gauge of revenue predictability and defensibility against switching and price compression.

Each metric maps directly onto the risks the analysis identifies. Idle compute hits margins. Falling revenue per R&D dollar signals escalating costs of frontier progress. A revenue mix tilted toward metered API usage exposes the company to the insourcing and switching dynamics the piece flags.

Why the $1 trillion versus $1.5 trillion gap matters

The reported investor push to mark Anthropic down from roughly $1.5 trillion toward $1 trillion follows last week's AI selloff, and the analysis's numbers explain the nerves. A company whose R&D expense quintupled in a year, whose benchmark progress per R&D dollar fell by roughly 60% and then almost halved again in the third quarter, and whose revenue model carries three identified structural vulnerabilities does not obviously support a premium to every company in the S&P 500 except a handful.

Meta's case runs in the opposite direction. A $135 to $145 billion capex budget — 3.5 times its 2024 level — represents enormous ambition, but the analysis pegs Meta's core ads business at roughly 10x 2027 EBIT on its own. Meta can absorb AI disappointment. Anthropic, at IPO, will be priced almost entirely on AI delivery.

The analysis's title, "Room To Be Wrong," captures the distinction. One company has an earnings floor beneath its AI bets; the other is the bet. As Anthropic's listing approaches, the market will have to decide which risk profile deserves the larger multiple — and the investor cohort reportedly arguing for $1 trillion rather than $1.5 trillion has, in the efficiency and defensibility numbers, its opening argument. How the IPO prices against that internal disagreement will set a benchmark for every private AI valuation that follows.