In the AI industry, a wealthy Tether has arrived

Written by: Curry, Deep Tide TechFlow

Tether made $13 billion in 2024.

You might not have a clear idea of this number. Let’s put it another way: OpenAI earned $3.7 billion in 2024 but lost $5 billion. Anthropic earned $1 billion but also lost $5 billion.

The combined losses of these two serious AI companies are still less than what Tether earned in a year.

Tether has a total of 150 employees, while OpenAI has over 3,000. The per capita output difference is approximately:

60 times.

How does Tether make money? When you buy 1 USDT, they receive 1 USD, which they use to buy U.S. Treasury bonds. The interest from these bonds belongs to them, and has nothing to do with you.

The essence of this is that Tether doesn’t pay interest. Banks need to pay interest to attract deposits, but Tether doesn’t. When you exchange your money for USDT and hold it, you earn no interest. They use your money to buy U.S. Treasury bonds, earning 7 billion in interest in 2024 alone.

150 people manage over 130 billion USD worth of Treasury bonds, doing nothing but collecting interest directly into their accounts.

Anyone would want to sit back and enjoy this kind of business.

But when you have a lot of money, you need to spend it. Tether chose a direction:

AI.

And it’s not just casually investing in a couple of projects to check the box.

First, computing power.

Running AI requires graphics cards; the more, the better, and the more expensive, the better. Tether lent over 600 million USD to a German company called Northern Data.

What does this company do?

Europe’s largest GPU cloud service provider. They have over 10,000 NVIDIA H100 graphics cards, the kind used to train GPT at OpenAI, costing around $20,000 to $30,000 each.

This cluster of graphics cards ranks 26th in the global supercomputer TOP500 list. The 600 million USD Tether invested is basically buying an AI training base in Europe.

Next, data.

Training AI requires feeding data. Last week, Tether released a dataset called QVAC Genesis, covering 19 disciplines including mathematics, physics, chemistry, and computer science. They claim it’s the world’s largest open-source AI training dataset.

It’s worth noting that the training data for OpenAI and Anthropic is not publicly available. Tether released theirs for free, accessible to everyone.

Then, there’s the more sci-fi part.

In April 2024, Tether spent 200 million USD to acquire a company called Blackrock Neurotech. The name includes Blackrock, but it’s not related to BlackRock.

This company develops brain-machine interfaces. They implant chips into the human brain, allowing paralyzed people to type, control wheelchairs, and operate robotic arms with their thoughts. It sounds like science fiction, but they’ve been working on it since 2008, eight years before Elon Musk’s Neuralink.

How impressive is this company?

Worldwide, 35 people have brain-machine interface chips implanted, 31 of whom use Blackrock’s technology. In 2016, a fully paralyzed patient used their device to control a robotic arm and fist-bumped Obama. The chip in his sensory cortex allowed him to “feel” Obama’s hand.

Last year, this brain-machine interface company helped a patient with ALS “speak” again. The chip in his brain translated his thoughts into speech at a rate of 62 words per minute.

Tether spent 200 million USD and became a major shareholder in this company.

And it’s not over. This year, Tether invested another 70 million euros in an Italian robotics company. They are also in talks with another German robotics firm, with an asking price of:

1.2 billion USD.

In total, Tether has invested nearly 3 billion USD in AI-related fields.

What does this mean?

Anthropic raised 3.5 billion USD in funding in 2024. Tether’s investment alone is almost as much as the total funding of a major AI company.

OpenAI spent 6.7 billion USD on R&D in the first half of 2025. Tether can be a benefactor in the AI circle with just a small fraction of its profits.

Why would a stablecoin company want to do AI?

We see two possible reasons.

First, anxiety. The Federal Reserve is cutting interest rates, and Treasury yields are falling. Earning 7 billion USD in interest in 2024 might not be as easy in 2025 and beyond. Even the printing press needs a new story.

Second, ambition. Everyone is talking about AI—investors, media, politicians. If you say you’re a stablecoin company, no one pays much attention. But if you say you’re working on AI, brain-machine interfaces, and humanoid robots—that makes you:

A tech leader.

What’s the most interesting part?

Tether promotes AI with slogans like “decentralization,” “local operation,” and “returning intelligence to individuals.”

But Tether itself is the most centralized company in the crypto world.

Issuing tokens is decided by them, the reserve amount is decided by them. It’s been ten years without an audit. Only they know where the users’ money is.

Such a company now wants to teach the world what “decentralized AI” really means.

It’s a bit like a casino boss running a class to teach people to quit gambling.

It’s not impossible.

After all, OpenAI is still losing money, expected to stop burning cash by 2029. Anthropic is similar, projected to break even by 2028. Sam Altman is raising funds everywhere, Dario Amodei is doing the same. Both companies have lost a total of 10 billion USD and are still spinning stories for investors.

Tether doesn’t need to tell stories. The money is already in its pocket.

What’s the biggest challenge in the entire AI industry? Business models.

How to make money? No idea. When to make money? No idea. Can they make money? No idea.

Tether doesn’t have this problem. Its business model is:

Not doing AI.

The profits from stablecoins are used to invest in AI. If the investment is right, it’s foresight; if wrong, it’s tuition. Anyway, it doesn’t affect their main business.

AI startups are losing money, while non-AI companies are making money. AI companies are raising funds, non-AI companies are investing.

By 2026, the best AI business model might be: don’t do AI.

First, get the printing press in order.