Eliminate the middlemen, and now you’re acting as the middleman for AI.

Written by: Cathy, Plain English Blockchain Section

In 2022, Galaxy Digital spent $65 million to buy a 180 MW Bitcoin mining farm. This year, the land has started generating rent for CoreWeave: a 15-year lease with average annual revenue of over $1 billion.

Crusoe was more direct. It packaged and sold its entire set of 425 Bitcoin data centers to NYDIG, then turned around and went to Texas to build Stargate for OpenAI.

Crypto.com spent $70 million on a domain. The domain is AI.com, the most expensive domain transaction in public records. The product ran ads during the Super Bowl.

Three companies—every one of them is heavily shifting toward AI.

And this isn’t another story about “adding blockchain to AI.” They don’t issue tokens, write white papers, or develop protocols. What they sell is compute power, server racks, and AI products. The customers are OpenAI, Microsoft, and CoreWeave.

The money is also moving in the same direction. In Q1 2026, about 6,000 startups worldwide raised $300 billion, and AI took $242 billion—about four-fifths. In the same period, only 8 new crypto funds were established, totaling $1.1 billion, the fewest since Q3 2020.

For an industry that makes money by tokens, why can it get a seat in AI?

01It’s not selling anything different

First, let’s look at Galaxy’s numbers.

Helios is a mining farm in West Texas. The purchase price in 2022 was $65 million. Galaxy stopped mining, invested $350 million of its own funds, and borrowed $1.4 billion in debt to renovate.

In this year’s first phase of delivery, the critical IT loads totaling 133 MW were sent to CoreWeave, and rent started to be billed. In the first three rounds combined, CoreWeave committed to 526 MW under a 15-year lease, plus two five-year renewal options.

An asset bought for $65 million became a business collecting $1 billion in annual rental revenue.

Crusoe’s path is even more decisive. The company started with oilfield-associated gas power generation; Bitcoin mining was simply its first use for consuming that electricity. In 2025, it sold its entire Bitcoin business—425 modular data center modules, leaving not a single one behind.

The freed-up capacity went to Abilene, Texas. That’s OpenAI’s Stargate Phase 1 campus, planned at 1.2 GW. In May 2025, Crusoe raised $11.6 billion in debt and equity for it.

In March of this year, Microsoft signed another 900 MW next door. In the end, the entire Abilene plan totals 2.1 GW.

Worth noting is the timing. Galaxy bought Helios in 2024, and Tether’s $420 million purchase of NVIDIA H100 was also in 2023.

Back then, ChatGPT had only just started to catch fire, and AI data centers were not yet something everyone was rushing to grab. These companies didn’t chase a wave of hype—they stood there first, and then the wind came.

By this point, the logic of the transformation is actually simple.

Over the past decade, the only real skill crypto companies have learned is how to turn cheap electricity and a piece of unwanted land into a row of racks that can be powered, can be cooled, and can run at full capacity all year round.

AI happens to be missing exactly this.

So the companies that already hold the power and the land can pivot with the least effort. The assets are ready-made, the customers are new, and the only thing in between is one round of renovation.

Others can pivot too, but they have to take a few extra steps first. They need to prove that besides issuing tokens, they can do something else.

02Not just selling land

Tether is taking a different route.

Its AI division is called QVAC. What it does has nothing to do with token issuance: an open-source, cross-platform SDK that lets billion-parameter-class models run directly on laptops, consumer GPUs, and even phones.

In March this year, QVAC released a LoRA fine-tuning framework for Microsoft’s BitNet 1-bit model; in April it released an SDK; in June it released an open-source implementation of Google Research’s VRAM compression algorithm TurboQuant. There’s also a local AI desktop app called Workbench.

A stablecoin company working on on-device inference optimization sounds absurd, but these things are real and live on GitHub.

Tether’s money also goes down another, more roundabout path. In December 2024, it made a strategic investment of $775 million in the video platform Rumble. This June, Rumble completed its acquisition of German listed company Northern Data, obtaining 85.2% of the shares.

Northern Data has the European cloud business with more than 22k GPUs, plus data center capacity of about 250 MW by 2027. It raised its revenue guidance for 2026 from €130–€150 million to €170–€190 million. It also signed a multi-year GPU cloud contract worth $270 million, with Together AI as the customer.

A stablecoin profit that detours past a video company ends up as a GPU data center in Europe.

Crypto.com’s $70 million isn’t just sitting there either. Marszalek started building a team after taking over AI.com in April 2025. The product launched after the Super Bowl and is an individual AI agent that can send messages for users, execute actions across apps, and trade stocks. He is also the CEO of two companies.

On the other side of the money, on July 8, Paradigm shut down its $1.2 billion fourth fund, explicitly writing AI and robotics into it. Paradigm manages nearly $12 billion in crypto venture capital and has already invested in drone delivery company Zipline and space defense company True Anomaly.

Most ironically, there’s OpenRouter.

Founder Alex Atallah is a co-founder and CTO of OpenSea. He left in early 2023 and built a “switchboard” for more than 400 large models: developers call its API once, and it decides whether the request goes to GPT, Claude, or some open-source model.

In May this year, OpenRouter raised $113 million in its B round at a $1.3 billion valuation. 8 million registered developers, processing 2.5 trillion tokens per week.

An industry that spent a decade arguing it “doesn’t need intermediaries”—the most successful company that emerged is one that brokers models.

Meanwhile, in crypto, the money is draining out at a visibly fast pace. In April 2026, global crypto venture capital invested only $659 million in a single month, the lowest in two years—down 75% year over year.

Why did the money go? Paradigm didn’t say much, but LPs’ books are clear. After FTX, Terra, and Three Arrows all blew up, institutional trust in crypto funds never truly recovered. In the last cycle, those high-valuation funding projects neither produced revenue nor built real users.

At least in AI, there are bills you can check.

The last time this industry collectively changed direction was after the ICO crash in 2018, when everyone shifted to building DeFi.

Back then, it was like rearranging rooms in your own house.

This time, they moved out.

03But the results are split

Not everyone has turned into that.

Canaan is an example that goes the other way. This company, which makes Bitcoin mining machines, also touched AI chips. In 2024, that business contributed about $900k in revenue. In the same year, the company’s total revenue was $269.3 million.

$900k versus $269.3 million. And this division consumed 15% of the company’s operating expenses.

In June 2025, Canaan shut down its AI chip business and went back to selling mining machines.

The mining firms are also layered. Those that have electricity, grid interconnection, and existing data center capacity got multi-year contracts from Microsoft and Google. Those with only mining machines and a press release—once the announcement goes out, there’s nothing more to show.

The same MW is a 15-year lease in a company that has handoff capabilities, but just a slide deck in a company that doesn’t.

The distinction really comes down to one thing: is anyone actually paying for it?

Galaxy has CoreWeave’s rent. Crusoe has Microsoft and OpenAI’s campuses. OpenRouter has 8 million developers using its API. Canaan has $900k in annual revenue—and then it settles the math itself.

04Summary

These companies aren’t doing “crypto-powered AI.” They’re just doing AI.

What actually gets carried over isn’t technology. It’s a group of people who know how to raise billions when nobody believes you, and how to get power connected to a barren plot of land within a dozen months. Those skills have nothing to do with blockchain; they were simply trained in the crypto industry by coincidence.

And we should be clear: going out doesn’t mean winning. Crusoe’s Microsoft campus in Abilene won’t be powered up until mid-2027. Crypto.com’s AI product just launched, and after Rumble takes Northern Data, it still has to prove it can operate the data center.

None of these are accepted yet. The only thing that can be confirmed is this: the bet has been placed—and it was placed on someone else’s table.

Ten years ago they said they wanted to eliminate intermediaries.

Now they’re acting as intermediaries for AI.