The AI Race Nobody Is Watching

GriftMatrix Substack Tech Oligarchy

I know a lot of people are spinning out over all of the “AI will be the end of us all!” talking points. But are you open to hearing a counterpoint? Things are not as dire as many think they are.

While Sam Altman is out here burning fourteen billion dollars a year — that’s billion with a B, in projected losses for 2026 alone — trying to convince investors that ChatGPT will somehow generate Nvidia-level revenue by 2029, Apple quietly figured out how to run massive AI models on your desk. No server farm. No cloud subscription. No sending your data to Sam Altman’s servers so it can be used to train the next model or, apparently, help plan military operations.

Let me explain…

In a normal computer, the brain that does the thinking (CPU) and the brain that does the graphics (GPU) keep their stuff in separate rooms. When they need to work together on an AI task, they’re constantly shipping data back and forth between these rooms. It’s like putting files on a bus and driving them across town in rush hour traffic. Slow. Wasteful. Your computer sounds like a jet engine and your electricity bill goes up.

Apple looked at this and said — what if we just put everyone in the same room?

That’s the M-series chip. The CPU, GPU, and a dedicated AI processor called the Neural Engine all share the exact same pool of memory. No bus. No traffic. No waiting. The M5 Ultra that’s about to drop can run a hundred-billion-parameter AI model — the kind of thing that normally requires a room full of Nvidia hardware — silently on a desk. The fan doesn’t even spin up.

One guy tested it. On his Nvidia rig, loading the same model took 45 minutes of Python scripting. The fans were roaring. The room temperature went up five degrees. On the M5 Ultra, he dragged the file into Apple’s native AI loader. It opened in four seconds.

Now here’s why this matters way beyond cool hardware demos.

Every single panic point about AI right now is built on one assumption: that AI has to run in giant centralized cloud data centers. The “AI is going to boil the oceans” warning? Based on cloud infrastructure. The “AI companies will own all our data” people? Based on cloud infrastructure. The “we need to give the Pentagon unlimited access to AI” drama? Cloud infrastructure.

What if that whole paradigm is already obsolete and most people just haven’t caught up yet?

Apple’s chips run these exact same models using a fraction of the electricity — less power than your kitchen toaster. Unified memory means less data movement, which means less energy wasted on what computer scientists call “memory bandwidth bottlenecks” and the rest of us call “why is my computer so loud and hot right now.”

The privacy implications are even bigger. Businesses can run powerful AI entirely on their own machines. Sensitive data never has to leave the building. Your medical records, your legal documents, your financial data — none of it needs to touch a server that some defense contractor might be running military operations on. Which feels particularly relevant this week.

And here’s the part that should really make you think about the long game.

OpenAI is projecting $14 billion in losses this year. They expect to hemorrhage $115 billion in cumulative negative cash flow through 2029 before they maybe turn a profit. Deutsche Bank said — and I quote — no startup in history has operated with losses on this scale. Their web traffic share dropped from 87% to 65% in a single year. Google Gemini ate 16 percentage points of that. Engineers are leaving OpenAI for Anthropic at eight times the rate of the reverse. They’re spending three dollars and thirty cents for every dollar they make. Every single query a user sends costs them more than they earn from it.

And their grand plan to fix this is… building more data centers and selling AI to the military.

This is not a successful business model.

Meanwhile, Apple’s approach doesn’t require any of that. No billion-dollar data centers. No Pentagon contracts. No circular financing where Nvidia invests in OpenAI which buys Nvidia chips. Just a chip that gets better every year, sold inside hardware that people already buy.

The M5 is 4x faster at AI than the M4. The M6 is already in development. Apple’s making their own AI server chips too — codename Baltra — for the stuff that does need cloud support.

A lot of people thought Apple missed the AI train — Siri still sucks — but as it turns out, they were thinking long tail this whole time and were just — fuck it — Siri can wait.

The pundits keep scoring the “AI race” like it’s about who can build the biggest server farm or who gets the fattest government contract. They look at Apple and see a company that makes pretty laptops. They’re missing the plot.

Apple isn’t trying to win the race everyone else is running. They’re running a completely different one — where AI is small, cheap, private, and energy-efficient enough to live on your desk. And when the cloud-dependent companies start collapsing under the weight of their own infrastructure costs — which, at three dollars per dollar earned, is a matter of when, not if — the company that figured out how to do AI without the cloud is going to look less like “losing” and more like the only adult in the room.

The existential AI crisis everyone is screaming about right now — the environmental destruction, the surveillance state, the military applications, the monopolistic control — those are problems created by a specific business model. Not by the technology itself. And that business model is already being disrupted by a company most people aren’t even watching.

Give it a few years. The things keeping you up at night about AI might not even be issues anymore. And ChatGPT might be a cautionary tale in a business school textbook about what happens when you mistake burning cash for building something sustainable.

Sources