LIVE · TAO
TAO$— SUBNETS VALIDATORS256
Bittensor intelligence updates
Home / AI News/ Engy (SN53) Is Earth’s Cheapest…
AI NEWS

Engy (SN53) Is Earth’s Cheapest Way to Run the World’s Biggest AI

Engy runs the world's biggest open AI model, Kimi K3, on ordinary gaming cards instead of costly data-center chips, making it one of the cheapest places on Earth to use frontier intelligence.

Engy (SN53) Is Earth’s Cheapest Way to Run the World’s Biggest AI

Somewhere out there is an artificial intelligence called Kimi K3, and it is one of the smartest ever built by anyone, anywhere.

Until this month, only a handful of companies could afford the enormous machinery needed to switch it on and keep it running.

A Bittensor subnet called Engy (SN53) just changed that by getting the whole thing running on ordinary gaming computers, the same kind teenagers use to play video games.

That single hack makes Engy one of the cheapest places on Earth to use frontier AI, and here is the story of how they pulled it off.

What Makes Kimi K3 So Hard to Run

Kimi K3 is the most powerful open model

Kimi K3 is what people call an open-weights model, meaning anyone is allowed to download it and run it themselves rather than renting it from one company.

Kimi K3 is very gigantic, with 2.8 trillion params and a memory that can hold a million tokens at once, so most computers simply cannot hold it.

To power this AI model, you need rows of specialized data-center chips costing a fortune. That is why owning this kind of intelligence stayed a rich-company privilege, and why what Engy did next surprised so many people.

The Gaming-Card Trick

Engy’s GLM-5.2 is Earth’s cheapest inference right now

Engy managed to run the full Kimi K3 on 80 RTX 5090 cards, which are ordinary gaming graphics cards sold in shops anywhere in the world.

The project’s founder mentioned that on day one, before any tuning, it produced 20 tokens per second, and he expects that speed to keep climbing as they refine it.

They had done the same warm-up act the week before with another top model called GLM-5.2, which they sped up from 30 tokens per second to 110 on that very same fleet, roughly three times faster overnight, an efficiency you can try yourself at engy.ai.

The clever part is that they used the abundant, affordable chips instead of the scarce, expensive ones, which is exactly where the cost savings come from.

Why Cheap Chips Change Everything

HBMs are AI’s premium memory

There is one very precious ingredient inside the fanciest AI chips called HBM memory, and it is the hardest and priciest piece of silicon in the whole industry.

Engy served its powerful model with none of that special memory at all, using only the standard gaming-card memory and the model’s official compressed weights with nothing watered down.

That means a small startup, a university, or even a curious lab in a far-off town can now own and run this intelligence for themselves, as the founder pointed out in his announcement.

Where Bittensor Comes In

Engy is subnet 53 on Bittensor, a network that rewards people around the world for contributing intelligence, much like Bitcoin rewards people for their computing power.

Those rewards act as a subsidy, so the folks running Engy’s machines can offer their service far cheaper than a normal company that has to charge full price to survive.

Engy also adds a layer of honesty called verification, a mathematical receipt that proves the AI you paid for is the exact AI that ran your request, described in their GitHub notes.

TaoDaily covered how that verification works in an earlier piece on how Engy tackles AI inference fraud, and it is a big reason people can trust the low price is also a fair one.

Anyone Can Plug In

Providers of inference on Engy

The real surprise is who runs the machines behind Engy, because it is not one big company but a crowd of independent people around the world.

Their live providers page shows the network at 64 clusters, 499 machines, 1,985 GPUs, and 63 separate providers, all pooling their computers into one service.

The mixture of hardware is the interesting part, since expensive data-center chips like the NVIDIA H200 and H100 sit right alongside ordinary gaming cards like the RTX 4090, competing on the same incentive.

Engy opened this mining up so that anyone with the right machine can plug in and earn rewards, which is how a scattered crowd of strangers ends up serving frontier AI cheaply.

The growth has been quick, and the numbers keep climbing as more people join. That growing pile of distributed machines is exactly what lets Engy keep pushing the price of intelligence down toward the cheapest on Earth.

The Beast Is Awake

For years, the most powerful AI stayed locked behind the walls of a few wealthy companies, and everyone else had to peer in from outside.

Engy took the most capable open model on Earth and put it on the most common computers on Earth, which quietly moved the power into far more hands.

The speeds will keep improving, and the costs will keep falling, so what looks impressive today is likely just the starting line.

The beast is awake; it runs on cheap gaming cards, and almost anyone can now afford to have a taste of Earth’s best inference.

Enjoyed this article? Join our newsletter

Get the latest TAO & Bittensor news straight to your inbox.

We respect your privacy. Unsubscribe anytime.

The Daily Dispatch

Enjoyed this article?
Join our newsletter

Get the latest TAO & Bittensor news straight to your inbox — every morning before markets open.

IA
Ige A
Senior Editor

Be the first to comment

Leave a Reply

Your email address will not be published.


*