Hey guys, Non here, coffee in hand. Want to hear some interesting stories?
Last week, a major cryptocurrency exchange shifted over 200 million dollars from Bitcoin into decentralized AI compute networks. This move signals that the biggest players in digital finance are no longer just betting on a currency, but on the literal hardware required to run the next decade of technology. It matters because it shows where the world’s "easy money" is moving. We'll look at why Nvidia's revenue just doubled on this same demand, then see Nvidia's plan to buy Hugging Face for 13 billion dollars.
We'll also cover Australia's new ban on AI music and the passing of the GLM-5.3-Flash model.
Nvidia reported its second-quarter revenue on Wednesday, and the numbers were what many analysts call a monster set of results. The company brought in 96 billion dollars, which is more than double what it made in the same period last year. To put that in perspective, its data centre division alone—the part of the business that builds the heavy machinery for the internet—generated 89 billion dollars. That is a 117 percent increase over the previous year.
Because of this, Nvidia’s shares rose by 4.7 percent in after-hours trading, and the company now sits as the world's most valuable firm with a market capitalisation of more than 5 trillion dollars. The logic behind these numbers is actually quite straightforward. Right now, every major tech player, from Amazon and Meta to Google and Microsoft, is engaged in a massive, expensive buildout of artificial intelligence infrastructure. They are essentially trying to build the largest collective brain ever conceived.
To do that, they need massive amounts of computing muscle, and Nvidia currently owns the most efficient way to provide it. When a company like OpenAI or SpaceX wants to train a new model, they have to buy the chips that make the math happen. Nvidia isn't just selling a product; it has become the primary landlord for the physical space where the future of software is being constructed. The popular narrative you are seeing in the headlines is that Nvidia is "winning" an AI war, almost as if there is a clear winner and a finished race.
The media often frames this as a triumph of innovation. But there is a more practical reality at play here. For the biggest companies in the world, buying these chips isn't necessarily a choice about who has the best idea; it is a choice about who can get the hardware the fastest. If you are a trillion-dollar company and you don't have the chips to run your next model, you fall behind your competitors.
In that environment, you buy what is available, regardless of how much it costs or how much it inflates the company's own balance sheet. Someone will say that this level of growth is unsustainable, or that the competition from cheaper suppliers in China or from customers designing their own chips will eventually crash the party. And they might be right about the long term. However, the current numbers show those challenges are limited for now.
The more important question is the sheer concentration of this wealth. With about 40 percent of the US stock market concentrated in just ten companies heavily invested in AI, Nvidia’s success is no longer just a Silicon Valley story. It is a primary driver of the broader market's health. When the cost of the infrastructure is this high, the only thing that matters is whether the finished product eventually pays for the construction.
We just talked about Nvidia's revenue doubling because the world is desperate to build the brains for artificial intelligence. Now, we are going to look at the physical skeletons those brains live in, and how a completely different industry—Bitcoin mining—is providing the literal bricks and mortar for that construction. The BBC reports that companies that used to fill warehouses with computers just to earn Bitcoin are now repurposing that same computing power for AI. To put it simply, the gold rush in crypto is being sold off to fund the AI boom.
These mining firms aren't just changing their software; they are refitting their entire physical infrastructure. To understand why this is happening, you have to look at what a Bitcoin mine actually is. It is not just a room of computers. It is a massive, high-voltage industrial site that requires two things that are incredibly difficult and expensive to secure: a steady supply of cheap electricity and the ability to run large-scale data centers without melting the equipment.
Bitcoin mining companies like Riot Platforms and Core Scientific have spent years mastering those two specific hurdles. Here is the economic hinge. Bitcoin mining is a "variable" business. If the price of a Bitcoin drops, a miner can simply turn off the machines and wait for the price to rise again.
It is a low-margin, high-volume game where the reward is often just a bit of new digital coins. But in October 2025, Bitcoin hit a peak of about $124,000. Since then, it has slumped. While it has rallied recently to around $80,000, that is still a long way from the top.
For the big players, the rewards aren't high enough to justify the overhead of keeping those massive warehouses running at full capacity. AI, on the other hand, is a "fixed" business. It requires the same huge data centers and the same cheap power, but it offers something Bitcoin doesn't: long-term, contracted revenue. Instead of hoping the market price of a coin goes up today, these companies can sign a lease to provide computing power to an AI giant like Anthropic.
Watch what happens next. Riot Platforms just signed a nine-billion-dollar deal with Anthropic. That is a twenty-year commitment. Bitdeer, one of the largest Bitcoin miners in the world, just signed a sixteen-year deal to provide compute for the same company.
The reason this matters for the listener is the permanence of the move. If you own a warehouse and a power contract, you might think you can switch back and forth. But you can't. Once you retrofit a multi-gigawatt power site to host AI chips, or "GPUs," you are locked in.
You are essentially changing the plumbing of the building. As Wolfie Zhao from The Energy Mag points out, you can unplug from the Bitcoin network any time. But you cannot easily undo a twenty-year lease to a tenant who needs specific hardware and cooling. There is a real case that this pivot might actually weaken the stability of the Bitcoin network. If the big, reliable players—the ones with the most infrastructure—leave the pool, there might be fewer computers verifying transactions, which could make the network less secure.
That is a fair concern. It is the classic trade-off between a speculative asset and a functional utility. The logic of the miner, however, is driven by the math of the invoice. It is much cheaper to keep a facility running with a guaranteed contract than to gamble on the daily price of a commodity.
The gap here is between what the headlines say—that these companies are "giving up" on Bitcoin—and what is actually happening. They are moving toward a dual-purpose model. Bitdeer’s chief strategy officer, Haris Basit, says they will keep mining Bitcoin because it is "interruptible." They will run the Bitcoin miners when the power is cheap and the market is good. And they will run the AI workloads to provide the steady, contracted revenue that pays the electricity bill. They are moving from being gamblers to being landlords.
They are building the infrastructure for the next decade of the internet, and they are doing it by cannibalizing the old one. The cost of this transition is high; some companies have even had to sell off their own Bitcoin holdings just to pay for the refitting. It turns out that in the world of heavy industry, it is often cheaper to change your business than to keep your old one alive.
We just saw how crypto firms are abandoning Bitcoin to find better shelter in the AI gold rush. And now we see the literal infrastructure of that rush being consolidated under one roof. If Bitcoin was the old way of betting on the future of value, Nvidia’s move to buy Hugging Face is a bet on who owns the plumbing of the new economy. According to Business Insider, Nvidia is in talks to acquire Hugging Face for 13 billion dollars. The two companies haven't signed a deal yet, but the talks are happening now.
Hugging Face is the primary hub for open-source AI, a massive library where developers go to find and share the models and datasets that power everything from chatbots to image generators. To understand why Nvidia wants this, you have to look at the math of the hardware. Nvidia already has a relationship with Hugging Face, having participated in a 2023 funding round that valued the company at 4.5 billion dollars. But Hugging Face actually turned down a 500 million dollar investment from Nvidia last year because they didn't want a dominant owner swaying their independence.
By buying them now, Nvidia moves from being a supplier to being the landlord. If you own the platform where every developer builds their AI, you have a much stronger path to ensuring those workloads run on Nvidia chips rather than on competitors like AMD or Intel. Someone will say this destroys the neutrality of open source. They have a point.
Hugging Face’s biggest selling point is that it works for everyone, regardless of which chip they use. If the company becomes an Nvidia subsidiary, that "Switzerland" status evaporates. But here is the thing that survives that tension. The sheer amount of capital involved makes the status quo difficult to maintain.
Nvidia currently holds 47.9 billion dollars in private company investments and has another 18 billion dollars committed to equity for this fiscal year. They are not just buying a website; they are buying the gravity of the developer ecosystem. For the founders, who are French and originally joked about going public with an emoji as their ticker symbol, this is a 13 billion dollar exit that could fund a new generation of labs in Europe. For Nvidia, it is a move to make sure that as AI becomes the standard, their hardware remains the only way to build it.
While Nvidia is buying the infrastructure to build the future of AI, the music industry is trying to decide which parts of that future it actually wants to host. It’s a shift from the hardware of the machine to the soul of the output. The Australian Recording Industry Association, known as ARIA, has officially banned songs that are largely or wholly created by artificial intelligence from its music charts. This comes on the heels of a specific controversy involving an Australian DJ named Josh Fawaz.
He used AI to remix a cover of Madonna’s Like A Prayer, which topped the ARIA dance singles chart and hit number two on the overall national chart. That single alone has been streamed more than 48 million times on Spotify. To put it simply, the industry is trying to draw a line between a tool and a replacement. ARIA says they want to promote human artistry in what they call the most crowded market in history.
The new rules allow for AI assistance—think drum machines, auto-tune, or mastering—but a human must still write the song and perform the lead vocals and primary instruments. They are specifically targeting music generated wholesale by services built on existing artists' recordings. Someone will say that these rules are just protectionist noise. And that if a song sounds good and millions of people stream it, the "humanity" of the creator shouldn't matter to the listener. It’s a fair point.
If the listener enjoys the song, the chart is technically doing its job of reflecting popularity. But the industry's argument is about the underlying incentive. ARIA’s chief executive, Annabelle Herd, argues that rewarding unlicensed AI output undercuts the very basis of the recorded music industry. If a machine can generate a hit by scraping the work of humans without permission, the financial reason for humans to create music in the first place starts to evaporate.
To address this, ARIA is now requiring artists to declare AI use upfront. And they have even threatened to retrospectively strip chart positions or demand the return of awards if a song is found to be mostly AI-generated.
Before the last story: if you are getting something out of this, subscribe. The World in Twenty Minutes is on Spotify, on Apple Podcasts, on iHeartRadio, and anywhere else you already listen. It is free, it lands every morning, and subscribing is the whole reason it keeps finding people.
Australia is trying to police the ghost in the machine by making artists sign a declaration. But in the world of software, the machine is moving so fast that the declarations are becoming obsolete before the ink dries. We just saw the music industry try to put a fence around AI-generated melodies. But while they are arguing over chart positions, the underlying technology is moving from a laboratory curiosity to a raw commodity. According to a thread on Hacker News, a new model called GLM-5.3-Flash was released recently by a Chinese lab. This is the part worth stealing: the speed of the progression.
In just four weeks, this specific model achieved performance levels comparable to previous top-tier models while cutting the number of parameters—the internal variables the model uses to process information—to a third. Then, in just twelve days, they released a "Flash" version that cut the price to a fifth of the original cost. The incentive here is simple. If you can provide the same intelligence for one-fifth of the price, the old winners lose their monopoly.
Someone will say that these models are just high-tech parrots, but the math says otherwise. If a company can match the performance of a "Pro" level model at a tiny fraction of the cost, they aren't just making a better product. They are making the previous version a financial liability. For the person running a business, the choice isn't between "good" and "bad" software anymore. It is between a high price tag and a functional tool that is getting cheaper every week.
When the cost of intelligence drops this fast, the gatekeepers of the old industry have a very hard time keeping the door closed. Goodnight for now.