Jungle Boogie
Time travel back to Autumn 2023. Some planning conversation somewhere inside Amazon, Microsoft, Facebook or Google went something like this, probably:
President of Data Centers: "Bruh, so this AI stuff is going to take a few more chips than I thought. I'm gonna need a bit more budget."
CFO, looking skeptical: "How much?"
PDC: "Look, I know we've been spending, like, $40BN a year on this stuff. It'll take a 50% step-up next year just to tread water."
CFO, eyes narrowing: "Excuse me?"
PDC: "OK…um…and we'll need to double that by 2026…"
CFO: "GTFO"
PDC: "…and that still won't be enough..."
CFO: [slams door]
Back in 2011, VC Marc Andreesen wrote "software is eating the world." It ate for the next 11 years. OpenAI's release of ChatGPT changed that. AI — in all its forms — LLMs, agents etc. — takes a lot more computing power than the software era — a lot more hardware.
Headwaters
In 2016, Amazon, Microsoft, Facebook and Google spent $30BN on capital investments — chips, cables etc... They increased that steadily. In 2023, their spending hit around $110BN. On the heels of OpenAI's release of ChatGPT in late 2022, spending really took off. In 2026 the group expects to spend more than $700BN. Next year? They could collectively pass the $1 trillion mark.
When investment spending began to ramp, Amazon and others changed a bit of their accounting. They said the new investments would have longer useful lives. They'd produce returns for four or five years rather than three. Three was the term they used before. That would flow through to profits. It would make the companies look better.
Investors pushed back. The pace of technological change had just leaped forward. They were investing to catch up to where technology was already, surely the investments would become outdated quicker!
Technology development did speed up. Today's accelerators can run circles around the ones those companies spent money on three years ago. Cerebras IPOed yesterday. Its accelerator is to three-year-old chips like comparing an anaconda to an ant.
What also happened was the bits — the tokens — processed by those chips became more valuable. That happened faster. It happened like a deluge after a storm — same river more water — same chips more intelligence. As a consequence, that hardware investors worried would become obsolete is more valuable now than they thought it would be.
In 2023, NVIDIA's H100 was the hot accelerator that everyone had to use for AI. NVIDIA released Blackwell in 2024. It’s the chip that powered the step change in AI agents. It’s much better than the H100s. Rubin, launched this year, is even better. Those H100s — now two generations behind — are now coming up for rent renewal on the open market. Those rental rates are higher today than back in 2024.
Delta
Amazon outspent everyone else. They spent on H100s, Blackwells, Rubins and they spent developing their own Trainium accelerators along with a bunch of other stuff. All of it, together, looks like it will deliver more value for longer than what they put into their books. As the biggest investor in this stuff, Amazon stands to benefit the most from that bonus value.
Eventually the hardware will wear out. But so far, AI models and the value they deliver have advanced quicker than the chips running them break. Amazon spent a lot — somewhere between $200BN and $300BN on AI-related investment over the last three years. Despite the flood, there's a computing shortage today. Prices are rising. That'll boost the return it gets on those investments. The market has been sympathetic to Amazon. Its price has gone up. But not enough. Its recent spending and increasing returns chart a course that ought to benefit shares.
Disclaimer: None of this is investment advice. It's meant to illustrate ways LCM thinks about investing. Things that LCM decides are good investments for LCM and its clients are based on many criteria, not all of which are covered here. Some or all of LCM's ideas may not be suitable for other investors. LCM does not recommend investing either long or short any position mentioned. LCM may own positions in some of the companies mentioned. Some of its ideas will lose money — investing entails risk. See full disclaimer here.