Listen to the audio version of this article (generated by AI).
Hello, Reader.
Nearly three decades ago, Amazon.com Inc. (AMZN) bet that consumers would embrace online shopping. In 2004, the company now known as Alphabet Inc. (GOOG) bet that search would become the gateway to the internet. Tesla Corp. (TSLA) bet in 2010 that EV manufacturing would scale.
These wagers were laid out in each company’s prospectus – a document that tells potential investors about a company and its plans. We won’t get too hung up on the jargon here, but most IPO prospectuses tell investors what a company hopes to build.
But Anthropic’s prospectus, leaked earlier this week, is different. It tells investors not just what the company hopes to build, but what it has already committed to buying.
Anthropic expects to buy $518 billion in cloud, computing, and infrastructure commitments over the next decade.
This isn’t just a staggering number – it’s a revealing one. It shows just how much Anthropic expects compute to matter to the future of AI. But that doesn’t mean the infrastructure will actually be available.
So, in today’s Smart Money, I’ll explain how Anthropic’s massive infrastructure commitments are highlighting new bottlenecks across the AI supply chain — and which companies may benefit.
Let’s dive in…
Money Can’t Buy Compute
Roughly 80% of Anthropic’s future computing and infrastructure commitments must be paid – even if it ends up not needing or using all of that capacity. That means Anthropic can’t simply walk away from most of its deals if AI demand falls short of expectations.
The biggest commitments are:
- Alphabet Inc. (GOOGL): $111.1 billion
- Amazon.com Inc. (AMZN): $110 billion
- Microsoft Corp. (MSFT): $31.4 billion
- Broadcom Inc. (AVGO): $161.2 billion
Anthropic is willing to lock itself into contracts of this size because that compute is becoming a bottleneck for the entire AI industry.
Indeed, the AI infrastructure boom is increasingly being driven by companies locking up massive amounts of computing capacity.
And this isn’t just an Anthropic story.
In August, Nvidia Corp. (NVDA) said shortages of memory components were limiting how quickly it could expand, even as it forecast 70% revenue growth for its next fiscal year. CEO Jensen Huang put it bluntly: “Compute is revenue.”
But money alone doesn’t create compute. Anthropic can commit hundreds of billions of dollars, but it can’t instantly manufacture the infrastructure required to turn that money into computing capacity.
The problem is that compute is only as available as the components that go into it.
For instance, AI accelerators need enormous amounts of high-bandwidth memory (HBM). So, you can have demand for more AI compute and still hit a wall if you don’t have enough memory to feed the processors.
And that’s exactly what we’re now seeing in the memory market…
The Bottleneck Spreads
Micron Technology Inc.’s (MU) latest earnings report, released this week, shows why compute is becoming harder to secure.
Micron’s customers have increased their long-term supply commitments to $32 billion, from $22 billion in June. Its remaining performance obligations have jumped to $150 billion, from roughly $100 billion. Micron also says memory supply-demand conditions should be substantially tighter in fiscal 2027 and 2028.
That’s the same scramble for future capacity we’re seeing with Anthropic, just one step down the supply chain. Anthropic is locking up future computing capacity, while Micron’s customers are locking up the memory needed to power it.
Why? Because both are concerned that the capacity won’t simply be available when they need it. So, Micron has become a real-world example of the warning signs inside Anthropic’s prospectus.
And that’s the bigger story: The bottleneck isn’t confined to compute.
In fact, it’s spreading across the entire physical infrastructure required to run AI.
Own What AI Can’t Live Without
Compute is AI’s umbrella problem, but the industry needs more than chips. It needs power, data centers, networking equipment, cooling systems, and optical connectivity – and all of those building blocks have their own supply constraints.
So, here’s my actionable advice: You want to own the bottlenecks, not the companies scrambling to overcome them.
In my FutureProof 2026 presentation, I detail the three major bottlenecks affecting the AI buildout: memory, raw materials, and energy. And the companies I’m watching most closely in those spaces are those that own the scarce physical assets AI can’t live without.
That’s why I share 15 tickers – free of charge – that I believe are positioned to benefit from the bottlenecks emerging across the AI infrastructure buildout.
Anthropic’s multibillion-dollar commitments show just how much money is chasing AI infrastructure. But as the company’s prospectus makes clear, money alone can’t create the compute and other infrastructure AI companies need.
That’s the opportunity I see: As companies like Anthropic race to secure compute, the companies that own the scarce infrastructure behind it could benefit.
Regards,
Eric Fry
