CoreWeave’s $39 billion AI bet says something bigger is happening

For most of the artificial intelligence boom, Wall Street has fretted that tech companies would one day build too many data centers, buy too many processors, and spend more money than artificial intelligence can reasonably earn back.

CoreWeave (CRWV) just gave investors the opposite problem.

The AI cloud company raised its 2026 capital-spending forecast to $35 billion to $39 billion, up from a previous range of $31 billion to $35 billion, after second-quarter revenue more than doubled. CoreWeave said its near-term computing capacity is effectively sold out, while its contracted revenue backlog reached $104.2 billion.

Shares rose more than 14% in after-hours trading on the results.

But the real story for investors is not just another AI stock exceeding quarterly estimates. CoreWeave has become a kind of bellwether in the market for whether the massive AI infrastructure spending is outpacing real demand.

Its figures imply demand is still out in front right now.

CoreWeave’s backlog was $104.2 billion at the end of the second quarter vs. $99.4 billion three months prior. Even more impressive is the corporation has locked in over $25 billion in net new client commitments year-to-date in the current quarter.

That goes a long way toward explaining why management is willing to expand an already enormous spending plan. CoreWeave is not developing billions of dollars of extra computing capacity with no clients in sight; a lot of that capacity is being developed against a massive book of committed business.

CoreWeave’s backlog may be one of the clearest AI demand signals yet

CoreWeave generated $2.58 billion in second-quarter revenue, more than double its year-earlier level and slightly above the $2.56 billion analysts expected. Its adjusted loss of $1.03 per share was also narrower than expectations for a $1.20 loss.

Those data explain some of the stock’s movement, but the backlog may tell investors more about what’s to come.

More than half of CoreWeave’s existing backlog relates to a contract that is already undergoing the process of being delivered to customers. That distinction matters because it transfers most of the company’s order book beyond predictions of future AI demand and to computer services actually being supplied.

The customers driving that demand are some of the greatest brands in technology. Microsoft (MSFT) is one of CoreWeave’s top customers, but its customers also include Meta Platforms (META), Anthropic, and Caterpillar (CAT). The tight integration with Nvidia (NVDA) has helped CoreWeave become a key source of computing capacity driven by the chips at the heart of the generative-AI boom.

CoreWeave is commonly referred to as a “neocloud,” a newer cloud provider that is geared toward serving the massive compute needs of AI developers and companies. Its business largely focuses on high-performance AI infrastructure, rather than competing with Microsoft, Amazon, and Google on every cloud service.

The specialization has proven valuable since the AI sector is faced with a fundamental constraint: There is only so much high-end processing power to go around.

Intrator said CoreWeave’s near-term capacity is effectively sold out and that the company is securing new compute agreements on “increasingly favorable terms.”

That might be one of the most important details for investors. Scarcity not only helps CoreWeave fill its data centers, but it can also give the company leverage when signing new contracts.

CoreWeave is spending $39 billion to keep up

But there’s another side to the tale, of course.

Capital expenditures were $9.4 billion in the second quarter, up from $6.8 billion in the prior quarter, and CoreWeave’s current full-year guidance implies as much as $39 billion in spending in 2026.

That’s a big commitment from a company that’s still producing adjusted losses, especially since AI infrastructure demands costly Nvidia chips, data centers, networking gear, and power.

More AI:

The bull case is that CoreWeave is investing against contracted demand and has increased its 2026 revenue and adjusted operating-profit estimates. It’s in the execution. Tens of billions are being invested, and construction delays, tighter financing, lower AI demand, or problems with big customers may swiftly erode the economics of that expansion.

CoreWeave just showed why AI capacity is becoming the new scarcity trade

Bloomberg / Getty Images

CoreWeave is benefiting from the AI bottleneck everyone keeps talking

CoreWeave is significant not just for its own stock but because it is positioned between some of the most crucial factors behind the AI trade.

Nvidia provides the computational engines. Those engines should be open to big IT corporations and AI developers. CoreWeave creates infrastructure around them and sells that compute power to customers who can’t get enough of it anywhere else.

So the startup provides investors with something like a real-time barometer of AI supply and demand.

CoreWeave numbers investors should know

$104.2 billion:Revenue backlog at the end of the second quarter.
More than $25 billion: Net new customer commitments secured so far this quarter.
$35 billion to $39 billion: New 2026 capital-spending forecast.
$9.4 billion: Capital expenditures during the second quarter.
$2.58 billion: Second-quarter revenue.
More than 14%: Initial after-hours stock gain following the results.
More than half: Share of backlog associated with contracts where delivery has begun.

For the typical investor, those numbers boil down to one question: Is the AI business developing too much infrastructure, or is it still failing to create enough?

CoreWeave’s most recent quarter suggests the latter in spades.

That does not automatically make the stock cheap, and a huge backlog does not alleviate the financial concerns generated by huge capital expenditures. But it does ease one of the most persistent concerns hanging over the AI trade, that infrastructure spending has grown untethered from real customer demand.

At CoreWeave, clients are pledging billions, capacity is tight, and management is spending quicker to keep up.

Eventually AI growth could face an overcapacity dilemma. CoreWeave’s recent data suggests it’s not there yet.