Here’s the dirty little secret behind “AI magic”: it runs on cash and compute. And the people who control the compute, Microsoft, Amazon, and Apple, are tightening the screws.
OpenAI and Anthropic may be the poster children for generative AI, but they’re also the industry’s most expensive tenants. Training and running these models burns through billions a year. So the cloud giants are doing what they always do: bankroll the future, then write the rules.
The math is brutal. OpenAI raised$6.6 billionin October 2024, yet its training costs are already north of$2 billion a year. Anthropic’s internal projections say it needs another$8 billionby the end of 2026. That kind of tab doesn’t get paid without strings. And Big Tech doesn’t do “no strings.”
Microsoft’s $13B OpenAI bet comes with handcuffs, and a revenue skim
Microsoft’s relationship with OpenAI is the cleanest example of modern tech power: “We’ll fund you, we’ll host you, and we’ll take a big bite of whatever you sell.”
Since 2019, Microsoft has poured about$13 billioninto OpenAI and locked down exclusive rights to integrate GPT into its commercial products. That’s not a side project. It’s the core of Microsoft’s pitch to every office worker and IT department on Earth: AI everywhere, powered by Redmond.
That’s how you getCopilotstuffed into Microsoft 365,GitHub Copilotaimed at developers, and more AI features headed toward the next Windows cycle. And Microsoft doesn’t just get bragging rights. Under the deal described here, Microsoft takes75%of OpenAI’s revenue until its investment is paid back.
Then there’s the infrastructure leash: OpenAI is required to run its compute onAzure. That exclusivity is estimated to be worth about$1.8 billionin cloud revenue for Microsoft in 2024 alone. So when Sam Altman talks independence, remember: independence is hard when your entire brain lives in someone else’s data centers.
And OpenAI’s internal turbulence isn’t helping. The article points to recent departures of senior leaders, including CTOMira Murati, as a governance hit that effectively strengthens Microsoft’s influence over the company’s direction. When a startup wobbles, the deep-pocketed “partner” doesn’t exactly loosen its grip.
Amazon’s $8B Anthropic deal: “Welcome to AWS, now switch to our chips”
Amazon took a different route with Anthropic, but the destination is the same: dependency.
In September 2024, Amazon announced an$8 billioninvestment tied to a major technical condition: Anthropic has to move training workloads onto Amazon’sTrainiumchips, gradually stepping away from Nvidia GPUs.
This isn’t charity. It’s strategy. Nvidia’s AI chips have been the tollbooth for the entire industry since 2023, and the prices have been ugly. Amazon wants out from under that, and it wants everyone else to follow. If Trainium becomes “good enough,” AWS gets to sell the picks and shovels instead of renting them from Nvidia.
Amazon’s bigger prize is the enterprise AI market, pegged here at$150 billion by 2027. And it’s already pushing Anthropic’s flagship model, Claude, deeper into the AWS machine: Claude is now running about60%on AWS infrastructure, up from20%at the start of 2024.
That migration is estimated to generate$2.2 billionin additional cloud revenue for Amazon, while making it painfully hard for Anthropic to ever pack up and leave. Once you optimize your whole operation around someone’s chips and tooling, “switching providers” becomes a fantasy you tell investors.
Anthropic does get something real in return: access to Amazon’s massive business customer base, roughly100 millionprofessional clients, via integration intoAmazon Q, Amazon’s enterprise AI assistant. That’s distribution you can’t buy off the shelf.
Apple’s quieter power play: distribution control and the 30% toll
Apple doesn’t need to write an $8 billion check to control the game. Apple controls the device. And in consumer tech, the device is the throne.
Instead of massive equity investments, Apple is leaning on selective partnerships and its famously closed ecosystem throughApple Intelligence, built intoiOS 18andmacOS Sequoia.
The OpenAI deal signed in June 2024 is classic Apple: ChatGPT gets front-row access through Siri across an estimated1.4 billionactive Apple devices, but only on Apple’s terms. The article describes restrictions like no user data collection, a limited interface, and Apple’s standard30%commission on Premium subscriptions sold through the App Store.
Apple’s control extends down to silicon. It’s pushing its own AI-focused processing via theM3 Neural Engine, reducing reliance on outside suppliers while forcing developers to play by Apple’s tooling and standards if they want top performance on iPhones and Macs. That’s how ecosystems become cages: not with locks, but with “best practices.”
The payoff is leverage. The article claims65%of mobile AI requests in the U.S. now flow through Apple devices. If you’re a model maker, that’s not a “partner.” That’s a gatekeeper.
The AI arms race is turning startups into dependents
Here’s the part that should make anyone allergic to monopolies sit up: the hottest AI startups in the world are starting to look like extensions of the cloud giants.
According to the figures cited here, OpenAI relies on Microsoft for about95%of its infrastructure needs. Anthropic is said to rely on Amazon for about80%of its compute. That’s not a healthy market. That’s a supply chain with two or three choke points.
And the next choke point is chips. Microsoft is developing its own processors (named here asAthena), Amazon is pushingTrainium 2, and Apple keeps beefing up its Neural Engine. Control the chips, control the costs. Control the costs, control who gets to compete.
So yes, 2024’s AI story isn’t just about smarter chatbots. It’s about who owns the pipes, who sets the tolls, and how quickly the “independent” AI darlings are getting folded into the empires they were supposed to disrupt.
