While most of the world slept, Google lost half a trillion dollars. Yes a Trillion, with a T. Some are already calling it Nokiafication, the fear that Google could lose its grip as people stop Googling and start asking AI.
But the bigger worry was the bill. Search printed money for two decades, and now that money is going into data centers. Alphabet expects to spend up to $205 billion this year, and for the first time since going public, it has burned more cash than it earned for two quarters running.
That raises a question few on Wall Street are asking. Computing power is fast becoming a commodity, and a GPU running an open AI model in Iowa does much the same work as one in Singapore. So why does almost all of it still come from a handful of giant cloud providers, while millions of capable GPUs in gaming rigs, workstations and university labs sit idle for much of the day?
XDC believes it can help close that gap. Linking spare GPUs into a shared network is technically straightforward, but paying the people who own them has proved far more difficult, since a single AI request can be worth less than a cent and the buyer and seller may be on different continents. Settling small payments between parties who have never dealt with each other is the kind of work XDC already does. The Layer 1 blockchain has spent years processing trade finance deals and cross-border payments for banks and businesses, and institutions including MoneyGram and Clearstreet now run validators on the network.
Computing power is starting to be bought and sold like any other commodity, and XDC wants to provide the network those trades are settled on. It is preparing to launch an open compute marketplace where developers can access open-source AI models through a single connection and pay for each request by card, without a subscription or a crypto wallet. Owners of capable GPUs will be able to connect their hardware with a single command and get paid for the work their machines carry out, with payments settled on XDC. The settlement system has already been tested on XDC’s public testnet, and the company is preparing a wider rollout.
“The world is going to need far more computing power than any one company can afford to build,” says Ritesh Kakkad, Co-Founder, XDC Network. “Much of that capacity already exists, sitting idle in machines around the world. The hard part is making it dependable and paying its owners fairly, one request at a time. That is what we have been building, and we look forward to opening it up very soon.”
The design reads less like a crypto project than a utility meter. Every request shows its own cost, down to fractions of a cent, so developers know exactly what they are paying and hardware owners know exactly what they have earned.
Decentralised compute networks have always found execution harder than the idea, and XDC will face the same test. Every connected machine has to download and run the AI models itself, and a GPU that goes offline in the middle of a job leaves a developer waiting for an answer that never comes. To compete with a data center, XDC will need thousands of scattered machines to perform reliably. Crypto markets learned a similar lesson with tokenized assets, where creating the token turned out to be easy and keeping it liquid across dozens of trading platforms proved far harder.
None of this rescues Google. Alphabet still has the cash, the chips and a cloud business that more than doubled last quarter. But the sell-off made one thing clear. Intelligence has become expensive to produce, and even its biggest producers aren’t sure how they will earn the money back.
The next time someone asks an AI instead of Googling it, the answer may come from a data center that cost billions. Or it may come from a GPU in someone’s spare room, paid by the request and settled on a blockchain built to move value between strangers. The market has only just started working out which is cheaper.
