The Great AI Valuation Pump: Follow the Money in the IPO Window

Something is off about this AI boom.

Not the technology itself. That part is real. But the financial timeline around it is starting to look less like momentum and more like a coordinated exit.

Let me show you what I mean.

The clock is ticking on the biggest IPO window in history

SpaceX filed its S-1 on May 20. Target valuation: $1.75 trillion. Target listing: June 12.

OpenAI filed confidentially on May 22. Target valuation: $1 trillion. Target listing: September.

Anthropic is closing a $30 billion round at a $900 billion valuation right now. Bankers expect an IPO by October at roughly half that number.

Three companies. One quarter. Over $3.7 trillion in combined market cap announcements inside six days.

There is no precedent for this.

The numbers are moving faster than the business

Anthropic sat at $61.5 billion in March 2025. Nine months later it was worth $183 billion. Three months after that, $380 billion. Today: $900 billion.

That is roughly 15x in 14 months. Same products. Same customers. Same market.

Here is where it gets interesting.

Forbes reported that some early Anthropic investors who entered at $4.1 billion or $61.5 billion are skipping this $900 billion round entirely. They are waiting for the IPO.

Why would early investors pass on a round that doubles or triples their paper gains?

Because bankers privately estimate Anthropic will IPO at $400 billion to $500 billion. Roughly half the current private valuation.

When the people who know a company best say "this price is too high" and the round still closes, you have to ask who is buying and why.

The same money is on both sides of every deal

Sequoia, Dragoneer, Altimeter, Greenoaks. Those four firms are co-leading Anthropic's $30 billion round at $2 billion each.

They also hold positions in OpenAI.

The largest growth-stage funds are no longer betting on one winner. They are funding both sides of the duopoly simultaneously.

This creates an artificial floor under valuations. If one company stumbles, the same funds still own the other. There is no independent price discovery. The "market" is four fund managers setting prices on both sides of the same trade.

And the ultimate capital behind them? Pension funds. Sovereign wealth funds. Crossover hedge funds. Retirement money flowing through SoftBank and sovereign vehicles into a market where the same insiders are setting the terms.

Revenue is being engineered for prospectus purposes

Anthropic reported its first operating profit this quarter: $559 million.

Tech journalist Ed Zitron argued this is a non-GAAP one-time result tied to a temporary compute subsidy from the SpaceX infrastructure deal. Not an improvement in unit economics. A subsidy.

There is also an $8 billion gap between what Anthropic reports as gross revenue and what OpenAI argues the net figure actually is. The dispute is over whether full end-customer spend should be booked when models are accessed through AWS and Azure, with partner payouts recorded as expenses. OpenAI says it inflates the real number by about $8 billion.

This is not an academic debate. It becomes a disclosure issue at IPO.

Meanwhile, gross margins are compressing. Anthropic's margins sit at roughly 40% after inference costs ran 23% over projections. The company's own CEO said a 12-month delay in AI progress would make Anthropic bankrupt.

The consumption pricing shift is juicing short-term revenue

Cursor hit $3 billion ARR in record time. Faster than Slack, Zoom, or Snowflake ever did.

But look at how they got there.

Earlier this year, Cursor quietly changed premium request costs for several models from 1x to 20x. The change was not announced. Developers discovered it when their monthly quota ran out in two to three days instead of lasting the full month.

Reddit threads documented overage charges of $200 to $400 on what users expected to be $20 monthly plans. One heavy work session could generate $50 in overages.

A few months ago, multiple AI companies quietly converted from per-request pricing to token-based billing. Requests became credits. Credits became dollars. Companies that were paying predictable monthly fees suddenly saw bills jump 3x, 5x, even 10x.

This is the pattern. Consumption pricing produces higher short-term revenue for IPO-bound companies at the cost of predictable budgeting for customers.

The same playbook is sweeping the industry. SAP moved to consumption-based AI pricing. OpenAI tiered its plans with usage caps. Anthropic introduced credit systems. Claude Code went pure consumption too, discounting the $20 seat but requiring upfront spending commitments that keep total cost high.

Every company is making the same move at the same time.

The $45 billion compute deal is the most circular number in the room

Anthropic arranged a nearly $45 billion deal with SpaceX for GPU compute capacity. $1.25 billion per month through May 2029.

A few things worth noting about this.

Elon Musk's SpaceX merged with xAI in February. SpaceX owns Colossus, one of the largest GPU clusters in the world. SpaceX is selling compute to Anthropic. SpaceX has an option to acquire Cursor. SpaceX is going public on June 12.

Musk is on every side of this trade.

The $45 billion commitment is larger than Anthropic's entire lifetime revenue. The company has raised roughly $90 billion across all rounds combined. It is pre-committing half of that to a single compute supplier.

Here is the loop: raise money to buy compute. Use the compute deal to signal demand growth. Use the demand signal to raise more money at a higher valuation. Repeat.

The open source risk nobody on Wall Street is talking about

Every valuation thesis for OpenAI and Anthropic rests on three premises: proprietary models are irreplaceably better, the gap will not close, and customers will keep paying premium prices.

Open source is attacking all three.

DeepSeek V4 Pro, Kimi K2.6, Qwen 3.6-A3B. These models are competitive with frontier proprietary models at a fraction of the cost. The gap between open and closed models has been narrowing, not widening.

Cursor itself routes traffic to cheaper models including China's Kimi to improve margins. The paid AI companies' own customers are using open models underneath.

If a free model delivers 90% of the capability for zero cost, the pricing power that justifies $900 billion valuations does not survive. This risk is absent from every upbeat analyst note on the sector.

The chronology

April 29: CNBC reports Anthropic is weighing a $900 billion round.

May 18: Elon Musk's lawsuit against OpenAI is dismissed by a jury in under two hours.

May 20: SpaceX files its S-1. Target: $1.75 trillion.

May 21: Bloomberg reports Cursor hit $3 billion ARR.

May 22: OpenAI files its confidential S-1. Goldman Sachs and Morgan Stanley leading.

May 22: Bloomberg reports Anthropic will close its $30 billion round "as soon as next week."

June 12: SpaceX IPO pricing.

July 2026: Cursor-SpaceX acquisition window opens.

September 2026: OpenAI IPO.

October 2026: Anthropic IPO.

Nobody knows where this ends

The technology ecosystem being built is real. Agentic AI will reshape work. Open source models are getting better every month. The engineering is not the problem.

The problem is the financial engineering happening in parallel.

$135 billion in new equity supply hitting the market in a single quarter. Three companies racing to go public in a six-month window. Circular capital recycling between the same four funds. Revenue numbers engineered through accounting methodology and aggressive pricing shifts. A $45 billion compute commitment that locks in costs before revenue is proven.

And at the end of it: the earliest, most informed investors sitting out because they expect the IPO to price below the private round.

The bubble question is the wrong question. The right question is simpler.

If everyone is cashing out at the same time, what do they see coming that the headlines aren't telling you?