Target $1.5T, Buyers Bid $1.1T: The Valuation Haggling Before Anthropic's Roadshow

Lay out the numbers first. Anthropic plans to push its IPO before Thanksgiving, targeting a $1.5 trillion valuation. That figure did not emerge from a discounted cash flow model — it looks more like an anchor that discounts years of future compute spending into today's equity price. The problem is the other end of the anchor: the buyers. Bloomberg reported that potential buyers value the company as low as $1.1 trillion. The gap is about $400 billion, a markdown of roughly 26.7%. This is not a tentative probe within a price range; it is a direct rejection of the issuer's pricing logic by the buy side.

Straighten the timeline and the contrast sharpens. On Nov 5, the Nasdaq 100 fell 6.4% in a single session and Wall Street lost $3.2 trillion in market value in a day. On Nov 6, Asia-Pacific markets followed: South Korea's KOSPI closed down 6.02%, triggering a circuit breaker after falling as much as 8% intraday and halting trading for 20 minutes; Taiwan's weighted index fell 1,873.08 points, or 4.51%; and the Nikkei 225 dropped below 60,000 for the first time since May. It is in the middle of these two weeks that Anthropic scheduled its roadshow. The issuer is unwilling to wait for market sentiment to recover, because waiting itself discounts the valuation.

Be clear about what officials have not said. FT and CNBC both reported that $1.5 trillion is the target valuation, that investors consider it high, and that the actual listing valuation could come in below both the target and earlier expectations. But the offering size and underwriter details have not been disclosed. In other words, $1.1 trillion is the low end of buyers' bids, not a deal price; 26.7% is the gap between bids and target, not a final discount. Our calculation: pricing at the midpoint of the $1.1 trillion to $1.5 trillion range implies a valuation of about $1.3 trillion, roughly 13% below target — an arithmetic exercise, not officially confirmed.

The target valuation is written by the issuer; the deal price is signed by the buyer. That $400 billion in between is the entire subject of this roadshow.

Buyers' hesitation is not emotion; it is arithmetic. A calculation by Dwarkesh Patel is widely cited: if an AI system at the level of a human software engineer ran on H100-equivalent chips, the annual rent for that chip should exceed $250,000 — 15 times the current spot price. Read that calculation in reverse: current compute spot prices are far below the theoretical value of 'replacing human labor,' which means either model capabilities have not reached that level, or compute supply will keep expanding and hold rents down. Either way, discounting future compute spending into today's equity at a 15x premium requires buyers to believe a premise that has not been validated.

Dylan Patel, founder of semiconductor analysis firm SemiAnalysis, put it more bluntly: Anthropic and OpenAI will control most of the world's compute by 2028. That is a bullish narrative for the issuer but a risk warning for buyers — concentrated compute means concentrated capex, and whoever controls most of the compute carries most of the depreciation. Buying Anthropic at $1.5 trillion essentially means buying a bill for years of future compute capex, and the depreciation schedule, residual-value assumptions, and free-cash-flow inflection point on that bill have not been disclosed in any prospectus.

Compare another financing line and the contrast is clearer. The first tranche of OpenAI's latest round has closed, raising $11 billion at a pre-money valuation of $1.25 trillion, with a total target of $20 billion for the round. Private markets are still voting with real money for a $1.25 trillion pre-money valuation, while public markets have marked Anthropic's $1.5 trillion down to $1.1 trillion. Same week, same sector, two markets, two prices. This is not about who is right or wrong; it is about different liquidity constraints and exit cycles — private capital has long lockups and can tolerate narrative discounting, while public capital has to answer to quarterly scorecards.

The boundaries must be clear: the calculations above are third-party or our own estimates. Anthropic has not publicly disclosed revenue, gross margin, net retention, or free cash flow, nor the specific terms attached to the $1.1 trillion bid (such as the proportion of secondary shares and lockup periods). A single quarter's book-building is not final pricing, and bids may still adjust with market swings during the roadshow.

Buying Anthropic at $1.5 trillion is not buying profit; it is buying a capex bill with no depreciation schedule yet.

Beyond the valuation dispute, two capital moves landed in the same week and point in the same direction. On Nov 6, Microsoft decided not to renew its 2023 cloud server lease with Oracle, cut the number of neocloud partners, and shifted focus to building its own data centers. Leasing is operating expense; building is capital expense — the former is flexible and can shrink with demand, while the latter, once started, is depreciated over a decade. By moving from rent to build, Microsoft is shifting compute costs from the income statement to the balance sheet, trading heavier assets for lower unit costs.

Google is on the same path, only earlier and heavier. On Nov 5, Google announced a $17 billion investment to build Wyoming's first data center and to cover power and infrastructure costs. The same day, securities filings showed Google bought nearly 3 million Qualcomm shares worth about $548 million, a stake of less than 0.5% — a financial investment small enough to ignore, buying exposure to AI data center chips, not control. Big money goes into self-built facilities; small money goes into the chip supply chain. That is a classic vertical integration play: hold compute in your own hands and outsource uncertainty to suppliers.

These two moves are a hidden negative for Anthropic's IPO pricing. If the biggest cloud providers are shifting from 'renting compute' to 'building compute,' then model companies that rely on external compute procurement have their unit cost curve in someone else's hands. What buyers are marking down is not just the valuation multiple; it is 'I can't see your costs, and I can't calculate your depreciation.' What has not been disclosed: how much of Anthropic's compute procurement contracts are locked in at long-term prices versus spot prices, which determines its exposure to compute price swings.

The roadshow starts next week, and pricing will not be settled in a day. Over the next two quarters, there are three observable signposts. First, where the final offering valuation lands within the $1.1 trillion to $1.5 trillion range — near the low end would show that buyer pricing power is established and that subsequent AI IPOs must recalibrate to this new anchor; near the high end would show that the roadshow narrative persuaded long-only money and that the markdown was merely a book-building tactic.

Second, watch whether Microsoft's and Google's self-build progress accelerates. Microsoft ending its Oracle lease and Google breaking ground on the $17 billion Wyoming project are both heavy-asset commitments. If both companies keep expanding self-build and cutting neocloud partnerships over the next two quarters, it means cloud providers see compute demand as rigid enough to bet the balance sheet on. If self-build slows and leasing returns, it means demand expectations are cooling, and Anthropic's compute cost assumptions would have to be revised down as well.

Third, watch whether market sentiment converges. The selloff on Nov 5-6 spanned US, Korean, Taiwanese, and Japanese stocks: KOSPI hit a circuit breaker, Taiwan stocks posted their fourth-largest closing point drop on record, and the Nikkei fell below 60,000. The Magnificent Seven lost about $1 trillion in combined market value, and Tesla and SpaceX together lost more than $420 billion. Commentator Jim Cramer warned that the AI stock selloff could continue but argued the current situation differs from the 2000 dot-com bust. That judgment can be falsified: if the roadshow prices near the low end and the stock breaks issue on day one, Cramer's 'different' thesis takes a hit.

A note on conflicting figures: Wall Street's single-day market value loss has been reported as both $3.2 trillion and $2.5 trillion, with no clear explanation for the difference; a headline claiming '$25 trillion wiped out' even circulated on Hacker News, off by an order of magnitude from official figures. We do not use unverified extreme figures; data in this article follow FT, CNBC, and Bloomberg reports. A single-day plunge does not establish a trend, and first-day book-building does not equal final pricing.

Where the roadshow pricing lands within the range says more about who controls AI equity pricing than any bullish research note.