Two things landed on the same day. Anthropic closed a funding round that values the Claude maker at $965bn, putting it ahead of OpenAI. Salesforce raised its full year revenue guide and announced a $25bn accelerated buyback. Investors who buried software last quarter are now writing checks to buy it back.

A $965bn valuation for a five year old company

Anthropic finalized a $65bn funding round this week. Including the fresh capital, the company now carries a $965bn valuation. That figure is larger than the combined market cap of most European banks and sits in the same neighborhood as the headline valuations Meta and TSMC have printed in the last year.

This is the cleanest signal so far that the AI capital cycle has not finished its second leg. The first leg was Nvidia plus the hyperscalers. The second leg is private market money pricing foundation model labs above publicly traded incumbents. OpenAI was the previous reference point. It is no longer.

A few things follow. Private markets are doing the heavy lifting because public investors will not pay the same multiples without revenue and unit economics they can verify. Banks underwriting any debt against these valuations need to think carefully about what the loan is really collateralized against, since most of the value is goodwill plus a bet on future inference margins.

The other point worth stating plainly: at $965bn, expectations are baked in. Anthropic now has to deliver a generational shift in software economics for that price to look sensible. The gap between price and current cashflow is the largest in any sector an investor would normally call early stage.

Salesforce, Snowflake, and a sudden floor under software

Salesforce posted Q1 numbers that broke the recent gloom. Non GAAP earnings per share came in at $3.88, beating consensus by $0.75. Revenue was $11.13bn, $70m ahead of estimates. EPS jumped roughly 50% year over year.

The market still flinched on initial guidance until Salesforce raised the FY27 revenue guide to a band of $45.9bn to $46.2bn and announced an additional $25bn in accelerated share buybacks. That combination, raised revenue plus capital return, was enough to flip sentiment.

Snowflake also popped on its own report. Analysts who spent the spring writing about a software apocalypse are now sounding less certain. The short version: companies that can package AI features into existing enterprise contracts are getting paid for it, and companies that cannot are not. The sector is bifurcating, not collapsing.

For banks, the second order effect is exposure to commercial real estate occupied by smaller software firms that did not get the AI uplift. That is where some quiet pain still sits, and it does not get unwound by one strong Salesforce print.

Amazon kills the AI usage leaderboard

A separate AWS internal story matters more than it sounds. Amazon scrapped an internal AI usage leaderboard after a senior executive told staff not to use AI just for the sake of using AI. Costs were rising in ways that did not tie back to outcomes.

This is the first major hyperscaler to publicly acknowledge that AI usage metrics are easy to game. If you measure people on prompts per week or tokens consumed, you get prompts per week and tokens consumed. You do not get better software or lower costs.

The reason this matters for the broader market: enterprise AI revenue lines at the big cloud providers are partly self consumption. When Amazon admits internally that not every prompt is a good prompt, the read across to revenue projections is uncomfortable. Cost discipline is starting to come for the AI demand curve too.

The Anthropic data center contract that may not be one

Elon Musk publicly contradicted SpaceX claims about a multi year Anthropic data center deal. The SpaceX IPO prospectus described the arrangement as a three year agreement. Musk said it runs for 180 days. Six months and three years are not the same number.

This kind of disclosure tension is exactly what regulators read filings for. If the structure is a 180 day base term with optional extensions, that is a different commercial commitment than a hard three year contract. The valuation a market gives to recurring infrastructure revenue depends on which one is actually true.

For banks underwriting the SpaceX IPO, the immediate question is how the contract is described in the prospectus risk factors. For Anthropic, at $965bn, every reported data center deal is part of the moat narrative. A 180 day commitment is not a moat.

What to watch

Three things are worth tracking over the next few weeks.

• Public market reaction to the next batch of software earnings tied to AI. If the Salesforce and Snowflake bounce holds across smaller names, the AI premium spreads. If it does not, this was a top two story and the rest of the sector is still in trouble.

• Any update on the Anthropic to SpaceX contract length. The IPO timeline will force clarity. The number that appears in the final prospectus is the number markets will price.

• Whether other hyperscalers follow Amazon in tightening internal AI usage. If Google and Microsoft do the same, the implied internal demand for AI compute is lower than the current quarter’s earnings calls have suggested.

The market right now is paying for two stories at once: AI labs at private valuations larger than most mature banks, and incumbent software vendors at recovery multiples. Both stories cannot be fully true. One of them will get repriced before the year ends.