Did META Just Expose The First Crack In the AI CapEx Boom?
If you're wondering why the Nasdaq is suddenly tumbling this morning, wonder no more...
Nasdaq moves lower after Meta announces it has capitulated in the race to build a leading frontier LLM it is to build a cloud business to sell its excess AI compute, weighing on cloud peers like AMZN, ORCL, MSFT, neoclouds like Coreweave and Nebius (who will now be racing to the bottom for customers) and chip and memory names like NVDA, MU, INTO, as demand for their products is now likely to be much less thanks to the excess META capacity on offer.
As Bloomberg reports:
Meta, which has been rushing to secure expensive data centers and other infrastructure to fuel its own artificial intelligence ambitions, is forming a business to generate revenue from excess computing power sold to outside customers, according to people familiar with the matter, who asked not to be named as the details aren’t public.
One potential plan includes selling access to various AI models that are hosted on Meta’s existing AI infrastructure, an approach similar to AWS’s Bedrock offering, the people said.
Meta would run the data centers and chips that power the models, including its own Muse Spark models, and charge developers to access them.
The report also notes that the company is considering selling access to “raw” computing capacity, taking a chunk out of the business of neoclouds like CoreWeave. Ironically, META just signed multi-billion deals with CoreWeave and Nebius, and now it is turning around to compete with the very suppliers it is paying.
Development of these new business lines is part of Meta Compute, an internal initiative to build and manage the company’s AI infrastructure efforts, according to a person familiar with the plans. Meta Compute is led by Santosh Janardhan, Meta’s head of infrastructure; Daniel Gross, a leader inside the Meta Superintelligence Labs AI unit; and Meta President Dina Powell McCormick.
Despite the complexities, Meta Chief Executive Officer Mark Zuckerberg has signaled to investors that he’s open to selling excess computing infrastructure, or even a so-called API service where customers would pay for AI usage — a business that’s usually measured in “tokens,” or the amount of data used and generated for a customer query.
“It’s definitely on the table,” Zuckerberg said during a call with shareholders in May.
“Almost every week there are different companies that come to us from the outside asking us to both stand up an API service or asking if we have compute that they could buy from us at some premium to what we’ve bought it at.”
This move comes after SpaceX started leasing its 'excess compute' (which is struggling now that it has competition in selling 'compute'):
...raising questions about the potential for cutting CapEx which has perhaps overshot token demand...
...did META just shatter the market’s central premise has been that compute is scarce...
As Goldman Sachs 1-Delta desk-head, Rich Privorotsky, has been warning:
“The market’s central premise has been that compute is scarce.
If scarcity persists, prices should remain firm and justify continued capex.
If supply rises and rental prices continue to drift lower, that is a direct challenge to the shortage narrative.
The first place that pain shows up is hardware.
ORNN H100 index rolling over last couple days worth watching.
The beneficiaries are the companies selling the complete platform and monetizing usage rather than simply selling picks and shovels. My working conclusion remains that hyperscalers are the structural winners through this phase.
The first moment they demonstrate they can deliver equivalent output with lower spend, the market will reward them.
The bigger risk sits further upstream in the hardware and infrastructure stack where expectations remain built around persistent scarcity."
Simply put, confessions of 'excess capacity' will crush the hyperbolic dreams of the CapEx cycle that underpins so much of the market's recent incredible performance.
And the pivot to rewarding CapEx cutters begins...
"Lots of underperformance in hyperscalers. Everyone still appears convinced they must keep spending simply to remain competitive, while token cost compression/advent of neoclouds puts pricing pressure on core business. If token prices continue to compress alongside falling compute costs, the benefits may accrue to users faster than providers.
Ironically, the first hyperscaler to signal that it can slow the pace of spending will likely see its share price rewarded.
If that happens, others will take notice.
That is the reflexivity that ultimately stalls the capex cycle… not a lack of demand, but investors deciding that incremental returns on the next dollar of spend are no longer attractive.
Watch hyperscalers share price as leading indicator."
Don't say you weren't warned.
META shares are notably higher on the news...
Chipmakers are hurting...
The writing had been on the wall...
...and Premium Subscribers can read the full notes we have published over the past month here:
'How Far Can The Rubber Band Stretch?': Goldman 1-Delta Desk Says 'This Is The Breaking Point'
Will 'Rampart' Wreck The Hyperscaler 'To The Moon' Narrative? Goldman 1-Delta Desk Deep-Dives
Buckle Up!