Anthropic's large-scale IPO appears to be a vote of confidence in the AI trade, but Bloomberg economists offer the opposite interpretation.
On November 19, according to analysis by Bloomberg economics economist Anna Wong, since September — the cutoff date for the financial data contained in Anthropic's prospectus (September 30) — its effective price per million tokens has declined sharply.
Over the same period, token usage surged, but market share did not rise accordingly and even edged slightly lower. After conceding a considerable portion of its price premium, the company did not gain a stronger competitive position.
Simply put, technological leadership and robust AI demand have not translated into durable pricing power. For investors, this means the valuation logic underpinning the AI trade is facing a test from fundamentals.
The IPO itself will bring the economics of frontier AI into public markets. From now on, pricing, margins, and returns on capital will be scrutinized on a quarterly basis. In hindsight, this IPO may look less like the starting point of a new leg up in the AI rally and more like the last wave of fuel.
As a leading player in frontier models and an iconic company of the AI boom, Anthropic has exposed precisely the soft spot at the core of this trade: the pricing power of frontier intelligence is being eroded.
Pricing Power: What Exactly Constitutes a Moat
A durable moat ultimately comes down to the ability to extract economic rents. Anna Wong's core argument is built on this standard: whether a company can sustain prices above competitive levels over the long term determines whether its valuation holds up.
Measured by this yardstick, the picture Anthropic currently presents is not ideal. During the period before the cutoff of the prospectus financial data, its effective token price fell steeply, its price premium was substantially conceded, and its competitive position did not improve as a result. Its technological edge did not translate into bargaining power on the pricing side.
Wong summarized the entire argument with two charts: first, the effective price per million tokens has clearly trended downward recently; second, token usage share has not risen but instead slipped slightly.
This divergence between price and share reveals the market structure in which Anthropic operates — it resembles more of an oligopolist without durable pricing power. Growth must be driven by price cuts to boost usage, rather than by product differentiation or customer stickiness to support prices. Wong wrote:
This is a difficult business.
It should be noted that Anthropic's massive fundraising this time may still add another burst of fuel to the AI trade in the short term. Support on the funding side can often extend a rally at the sentiment level.
If Pricing Power Falls Short of What Valuation Implies, Repricing May Accelerate
But the other side of the IPO is transparency. After listing, pricing, margins, and returns on capital will be disclosed and examined quarter by quarter, and the pricing power issues that previously existed only in private markets will be placed under the spotlight of public markets. The assumptions embedded in the valuation will also, for the first time, be reconciled against continuous data.
The core risk Wong highlights is this: if Anthropic's future results show that the company's pricing power is weaker than its valuation suggests, then the repricing that follows could accelerate a broad giveback in AI sector valuations.
From a broader market perspective, she also recently noted that, based on the rate analysis model commonly used by the Federal Reserve (the Kim and Wright model), the risk compensation (spread) currently demanded by the market has already widened significantly.
At the same time, against the backdrop of tightening financing conditions, a large number of the more than 900 supply chain bottlenecks mentioned by listed companies in the previous earnings season have yet to ease. Together, these high-frequency signals form the basis of her ongoing tracking of market and credit risks.
Comments