This is the part of the AI boom that people keep pretending isn’t happening: it’s not some light, magical “software revolution.” It’s a money-and-power land grab, and the price tag is starting to leak out in court.
A Microsoft executive testified in the Musk vs OpenAI case and said Microsoft has spent over $100 billion on its partnership with OpenAI so far. Not just “we invested in them.” The number includes direct investment, the Azure infrastructure, and the ongoing hosting costs for running models like ChatGPT. Public reporting also says this ramps up from Microsoft’s initial $1 billion investment in 2019, and later a commitment of around $13 billion more.
If that $100B figure is accurate—and it was said under testimony, which is not the same thing as a marketing blog post—it tells you something blunt: modern AI isn’t just about clever research. It’s about who can afford to run the machines, and who gets to sit closest to the control panel.
My reaction is pretty simple: this is both impressive and concerning. Impressive because Microsoft clearly decided that being “the place where AI lives” is worth a staggering amount of money. Concerning because once you spend that much, you don’t back away politely. You don’t say, “Well, that was fun, let’s keep it open and neutral.” You squeeze. You bundle. You lock in customers. You make sure the rest of the market has to route through you, even if nobody calls it a monopoly out loud.
And that’s the key tension for me. A lot of people want OpenAI to feel like a public-minded lab that happens to ship products. But a $100B partnership doesn’t behave like a lab. It behaves like a strategic asset.
Imagine you’re a mid-size company trying to build a product on top of these models. You want stability. You want predictable pricing. You want to know that the tool your product depends on won’t get turned off, rate-limited, or suddenly repriced because two giants are renegotiating behind closed doors. When one partner is spending this kind of money on compute and hosting, your “platform” starts to look less like an open ecosystem and more like a company town.
Now flip it. Imagine you’re Microsoft. You’ve eaten massive costs to host and run these models. Your incentive is to drive usage through your cloud, push premium tiers, and attach AI features to everything you sell. That might be great for Microsoft shareholders. It might even be great for many customers in the short term—easy integrations, familiar tools, fewer vendors to manage. But the long-term consequence is that “AI capability” becomes less like a common ingredient and more like something leased from one of a few gatekeepers.
People will argue: so what? Big infrastructure always costs big money. This is no different than building data centers for search, video, or cloud apps. And honestly, that’s the strongest counterpoint. If AI is as useful as claimed, then of course it takes heavy investment. Maybe this is just what it costs to build the next layer of computing.
But I don’t think it’s that neutral. The difference is that these models aren’t just infrastructure. They’re also decision engines that shape writing, hiring, customer support, education—stuff that touches real lives. When the ability to run those engines depends on $100B-scale spending, the winners are basically pre-selected.
There’s another consequence people gloss over: once costs are that high, the pressure to make the technology pay for itself gets intense. That doesn’t automatically mean “evil,” but it does mean trade-offs will show up in product choices. Maybe safety and quality improve. Or maybe “good enough” wins because it’s cheaper. Maybe tools get designed to drive engagement and dependency rather than accuracy. If you’re paying to host an ocean of compute, you want the ocean to be busy.
It also changes the labor conversation. If AI features get bundled everywhere because Microsoft needs usage, then lots of workplaces will adopt them by default, not because they have a careful plan. Picture a manager pushing AI summaries into performance reviews because it’s “included,” or a call center being told to rely on AI replies because it’s “the new standard.” Some people will be helped. Some people will get steamrolled. And the line between those outcomes will often be decided by budget and policy, not by what’s actually best.
One more uncomfortable angle: this testimony is happening inside a lawsuit context. We don’t know what else will come out, or how each side is framing facts. Courtroom revelations can be clarifying, but they can also be selective. Still, even with that uncertainty, the scale alone matters. It signals that the AI race is not just “who has the best model.” It’s “who can afford to keep the lights on.”
So yes, I’m impressed. But I’m also wary of the story we tell ourselves—that this is an open, competitive, innovation-driven moment. A $100B partnership is a bet that the future will be rented, not owned, by most people using it.
If AI really is becoming basic infrastructure for work and communication, should we be comfortable with it being shaped primarily by who can spend $100 billion to keep it running?