Editor's note: AI companies are desperate to obtain energy amid growing demand...
Some companies are recycling power from their competitors to meet their needs. And according to Joel Litman – chief investment officer at our corporate affiliate Altimetry – one Big Tech behemoth could gain an edge over the rest of the competition... despite negative market sentiment.
In today's Masters Series, originally published in the August 13 issue of the free Altimetry Daily Authority e-letter, Joel explains how one company could earn billions from its AI rivals...
The $10 Billion Proposal Reshaping AI Infrastructure
Big Tech companies, like Microsoft (MSFT) and Alphabet (GOOGL), invested billions of dollars into AI startups...
Those startups, in turn, spent money on Big Tech's cloud infrastructure and chips.
It was the first phase of AI's circular economy. Now, another loop is forming... AI competitors are starting to recycle one another's computing power.
AI research firm Anthropic recently approached social media giant Meta Platforms (META) about leasing its AI data-center capacity.
The potential agreement could be worth as much as $10 billion over two years.
At the same time, Meta is spending heavily to build AI models that can compete with Anthropic's Claude Code.
This is fueling concerns that Meta is falling behind in the AI race... and preparing to hand valuable resources to a fierce rival.
But, as we'll explain, Meta's excess computing capacity could actually strengthen its AI strategy.
Scarcity is turning competitors into customers...
AI companies need enormous amounts of computing power to train large language models and run them for millions of users. The most popular models are consuming capacity faster than developers can secure it.
Anthropic's fastest-scaling enterprise product, Claude Code, has fueled a surge in demand. The company is now seeking infrastructure wherever it can find a reliable supplier.
It reportedly signed a three-year agreement with SpaceX (the parent company of another competitor – xAI) worth $45 billion. Under that agreement, Anthropic would pay $1.25 billion per month for computing power.
Now it's in talks with Meta.
The Facebook owner plans to spend as much as $145 billion this year, much of it on AI. That's more than double the roughly $72 billion it spent on the technology last year.
The company doesn't care if its servers belong to a rival. Meta just needs reliable computing power... and it's prepared to pay a lot for it.
Investors don't think Meta can earn an adequate return on its aggressive build-out...
In addition to developing models, Meta is expanding its compute capacity. Management acknowledged that the company may build more capacity than its own AI products need.
Leasing that spare capacity would turn a potential weakness into a revenue-producing asset. And because it's such a precious asset, Meta can charge top dollar.
That's an important point because investors expect Meta's returns to deteriorate. We can see this through Altimetry's Embedded Expectations Analysis ("EEA") framework.
The EEA starts by looking at a company's current stock price. From there, we can calculate what the market expects from the company's future cash flows. We then compare that with our own cash-flow projections.
In short, it tells us how well a company has to perform in the future to be worth what the market is paying for it today.
Meta's Uniform return on assets ("ROA") has been at least 26% every year since 2021. That's more than twice the 12% corporate average. (We use Uniform Accounting at Altimetry to cut through the noise and distortions that as-reported accounting methods provide.)
Even if the company sells some of its new computing power at top dollar, investors expect its Uniform ROA to fall to just 18% in the next few years. Take a look...
Their outlook assumes that Meta's infrastructure will become less productive as spending rises. But if Meta can lease its excess capacity at a premium, its returns may not fall as far as the market expects.
AI infrastructure is more than just the models it supports...
Companies like Meta are selling computing capacity to competitors before their own AI products need all the capacity they've built.
Meta's infrastructure also supports internal products, advertising tools, and enterprise services.
That creates several paths to returns from the same asset base.
Right now, the market is concerned about Meta's infrastructure growth. And it's underestimating the company's AI business model.
But management has planned ahead... Meta can continue developing advanced AI models while collecting revenue from its competitors.
Regards,
Joel Litman
Editor's note: During SpaceX's IPO in June, all of the headlines were about its sky-high valuation and the $75 billion it raised. Lost in the shuffle were Elon Musk's plans for those funds.
But Joel has been paying attention. Soon he'll be going on camera to reveal how Musk is likely to target key bottlenecks in memory chips, computing infrastructure, and more... He'll also share how Musk's goals could have a major impact on certain industries.
