The Hundred-Billion-Dollar Quest for a Better Toaster

There is a certain kind of ambition that can only be expressed in eleven figures. It is the sort of ambition that looks at a perfectly functional global social media empire and decides that what it really needs, more than anything else, is a 'personal superintelligence.' And to achieve this, one must apparently go shopping for silicon with the same reckless abandon a toddler displays in a sweet shop, provided the toddler has a credit limit roughly equivalent to the GDP of Morocco.

Meta, the company formerly known as a place to argue with your aunt about sourdough, has recently struck a deal with AMD that could reach the dizzying heights of $100 billion. This is not, as one might hope, a very expensive way to ensure that the 'Poke' button finally does something useful. Instead, it is a strategic pivot into the world of high-end chipsets, a move designed to ensure that Mark Zuckerberg’s digital offspring have enough processing power to contemplate the universe, or at least generate a very convincing image of a golden retriever wearing a tuxedo.

One must admire the sheer scale of the logistics involved. A hundred billion dollars buys a lot of things. It buys several islands, a fleet of moderately used aircraft carriers, or, in this case, a mountain of HBM3 memory and GPU cores. It is a diversification strategy that feels less like a business move and more like a polite, yet firm, eviction notice served to Nvidia. For years, Nvidia has been the only game in town, the sole purveyor of the digital shovels required for the great AI gold rush. Meta, it seems, has decided that relying on a single shovel-maker is a bit like relying on a single brand of tea: risky, potentially disappointing, and ultimately un-British.

(I once spent three days trying to find a specific brand of Earl Grey in a remote village in the Cotswolds. The resulting existential crisis was not unlike the current state of the semiconductor supply chain, though with significantly less venture capital involvement.)

The deal also involves a 'warrant' for 160 million shares, which is a delightful piece of financial wizardry. It essentially means that Meta might end up owning a significant chunk of the company that is making the chips it is buying. It is a recursive loop of corporate ownership that would make a Victorian bureaucrat weep with joy. You buy the chips, you own the maker, you use the chips to build an AI that tells you to buy more chips. It is the circle of life, if life were made of etched silicon and quarterly earnings calls.

But what, one might ask, is a 'personal superintelligence' actually for? The term itself is wonderfully vague, suggesting a digital companion that is both infinitely wise and deeply concerned with your personal schedule. One imagines a being that can solve the Riemann Hypothesis while simultaneously reminding you that you’ve run out of semi-skimmed milk. It is the ultimate butler, a Jeeves for the algorithmic age, capable of navigating the complexities of quantum physics and the even greater complexities of a Tuesday afternoon in Slough.

There is, of course, the slight problem of power. Not political power, though that is certainly a factor, but the literal, humming, electricity-bill-inducing kind. Running a hundred billion dollars' worth of chips requires enough energy to make a small sun feel inadequate. We are building digital cathedrals of logic, vast data centers that hum with the collective effort of billions of transistors all trying to decide if a picture of a muffin is, in fact, a chihuahua. It is a monumental achievement of human engineering, dedicated to the pursuit of the perfectly curated feed.

(I recently asked a moderately intelligent AI to help me organize my sock drawer. It suggested I buy more socks. I suspect the 'super' version will simply suggest I buy the factory.)

The irony of the situation is not lost on those of us who remember when the internet was mostly about sharing low-resolution photos of cats. We have moved from the era of the 'Like' to the era of the 'Large Language Model,' and the cost of entry has risen accordingly. To participate in the future, one must now possess the financial clout of a mid-sized nation-state. It is a high-stakes game of musical chairs, where the music is composed by a neural network and the chairs are made of rare earth metals.

As Meta and AMD embark on this hundred-billion-dollar honeymoon, the rest of us are left to watch from the sidelines, wondering if our personal superintelligence will be polite enough to say 'please' before it inevitably takes over the household chores. One can only hope that with all that processing power, it might finally be able to explain why the printer never works when you’re in a hurry. But perhaps that is a problem even a superintelligence cannot solve.

In the end, we are witnessing the industrialization of thought. We are building the infrastructure for a world where intelligence is a utility, piped into our homes like water or gas. And if it costs a hundred billion dollars to ensure that the digital pipes don't leak, then so be it. Just don't be surprised if your new superintelligent friend spends most of its time trying to sell you a slightly better toaster.