Hey guys, Non here, coffee in hand. Want to hear some interesting stories?
I want to start with a number you might not expect on a list of global powers. China now produces and refines more oil than any other country in the world, and it does it largely through the complex network of pipelines running through Xinjiang into Central Asia. That is the story I want to walk through this morning. We will also look at what Elon Musk’s new Grok model is actually doing to speed up your options, take a quick look at Google’s latest attempt to shave down the wait times, and explain why the biggest companies in Tokyo are still waiting for the technology to land before they try to use it.
And finally, we will remember Nirmal Purja, who climbed the world’s highest mountains with the specific intent of rewriting the records. Let’s talk about the oil pipeline first.
You are standing in front of a massive, blinking machine at a trade show. The salesman tells you this is the only engine that can move the future. He shows you a price tag that makes you blink. Then, your phone buzzes with news that a rival company has just rolled out an engine that does exactly the same thing for pennies on the dollar, and they are giving it away for free to anyone who wants to try it.
According to The AI Daily Brief, that is the situation with Grok 4.6 this week. xAI released a model that is described as fast and capable, and, critically, dramatically cheaper than the current leaders. The headline framing is all about "massive funding rounds" and "booming infrastructure demand" — the idea that this is a zero-sum war where only the biggest, richest player survives. The popular narrative is that AI is becoming a natural monopoly, a utility like electricity or water, where one provider eventually wins all the customers and you either pay their price or do without.
That reading depends on the idea that there is a ceiling to how good AI can get, so you need a giant infrastructure to reach it, and only giants can afford that infrastructure. The more important question is not how big the machine is, but what it costs to run it. When Grok 4.6 shows up and performs at the level of the expensive models but at a fraction of the price, the economics of AI flip. If a single engineer with a high-end laptop can run a model that replaces a team of analysts, the value of a massive data centre drops.
The "infrastructure boom" is a bubble if the marginal cost of intelligence keeps falling faster than the price of the computer you need to run it. Someone will say that the infrastructure is the bottleneck, that we simply do not have enough chips to run all these models at once. But the data is already suggesting the bottleneck is not power or hardware, but attention. The "41 stats" in this week's briefing note that a majority of American workers use AI, but the gap between the frontier labs and the rest of us is widening.
The "supermodel" gets the headlines and the funding, but it is mostly idle while the smaller, cheaper models do the actual work. We are looking at a fork in the road. Either AI becomes a high-priced luxury, reserved for enterprises that need to lock in their data and their compute, or it becomes a commodity, a utility that is so cheap and ubiquitous that it changes how we organize society from the bottom up. You can tell which way it is going by looking at the people leaving the big labs.
When the people designing the systems realize they can get the same output for a fraction of the cost by using an open-weight model, they stop building the megacenter and start building the network. The race is no longer about who has the biggest server farm. It is about who can deliver the most intelligence for the least friction. And right now, the price of intelligence is falling faster than the hype.
We just talked about how the cost of intelligence is dropping as AI models get cheaper and faster to deploy. It’s a story about software becoming a utility. But now we have to shift gears from bits to barrels—from the digital flow of information to the physical flow of energy. If AI is the new electricity, then oil is still the blood supply that keeps the existing world running while we build the next one.
According to reports from Al Jazeera and The Economist Finance, China has officially cemented its position as the world’s leading oil power. This isn't just a matter of who owns the most land; it is about who controls the movement, processing, and ultimate destination of the world’s crude. While Western powers have spent decades trying to diversify away from fossil fuels, China has doubled down on the infrastructure required to dominate the global supply chain. To understand how this works, you have to look at the plumbing of global trade.
Oil isn't just a commodity you buy; it is a logistical marathon. It requires massive refineries, specialized shipping fleets, and "tanker hubs"—strategic ports where oil can be stored and redirected depending on where prices spike or shortages occur. China has systematically built these assets. They aren't just buying oil from the Middle East or Russia; they are building the pipes that ensure that if a shipment is disrupted in the Strait of Hormuz, there is an alternative route through their own ports and refineries to keep their factories running.
By controlling the mid-stream—the refining and transport—they can dictate terms more effectively than those who simply sit on the oil in the ground. Someone will say that this focus on oil is a sign of failure, a move backward in an era defined by the green transition. They might argue that China is tethering its future to a dying industry while the rest of the world moves toward renewables and battery storage. It’s a fair point; it looks like an admission that they can't leapfrog the carbon age as quickly as we hoped.
But there is a different way to read this move, one that might be more pragmatic for their survival. China faces a massive "dual-track" reality. They need to build the green grid of the future because it provides energy security and manufacturing dominance, but they cannot turn off the internal combustion engine overnight. Their current industrial base—the steel mills, the heavy chemicals, and the shipping containers that make them the world's workshop—still run on high-heat processes that require massive amounts of oil and coal today.
By becoming the global oil power, China is essentially buying itself a longer runway. They are ensuring that even as they transition to electric vehicles, the very machines used to build those vehicles stay powered by affordable, reliable crude. The more important question is what this means for your own costs at the pump or on your heating bill. When one country becomes the primary "hub" for oil, it creates a bottleneck of influence.
If China controls the majority of the world's refining capacity, they have a significant lever over global price stability. We are seeing a shift where the power isn't just in who has the most barrels—which is still spread across many nations—but in who decides which country gets to use them first. To put it simply, China is moving from being a customer of the world’s energy system to being its primary manager. They are positioning themselves so that any future shortage or price spike becomes a conversation they have to host, rather than just a problem they have to endure.
It's a massive scale shift: one nation now moves and processes such a significant portion of the globe's oil that it rivals the combined logistical reach of most Western economies. We don't yet know exactly how much this will suppress prices in the short term, but we do know that when one person owns the warehouse, they eventually decide who gets to shop there.
The story we just left was about oil and who gets to shop in the warehouse. Today, the question is who gets to read the receipt. In the last story, a few hands owned the pumps and the pipes. Today, the question is whether a single prompt can write a whole year’s worth of software.
According to Google, they just released a model called Gemini 3.7 Flash, and they’re selling it cheaper than the previous version. Three weeks after the 3.6 Flash dropped, they’ve shaved another 50 percent off the price per million tokens. That is, the cost of processing every million words of text has dropped by half. They claim this model is better at coding, law, and bioscience, and they back it up with benchmarks: 43.6 percent accuracy on fixing bugs versus 34.4 on the last model, and it scored 1588 on the Arena.ai leaderboard, versus 1538.
The price change is real: $0.75 per million words of input and $3.75 per million words of output through the end of the year. Here is how the incentives line up. If you are a developer building a product, you are paying by the token. Every time the model writes a line of code, you spend a fraction of a penny.
Google wants you to use this model more. The faster you can iterate, the more you pay. So they lower the price and raise the intelligence. It is the same logic as the oil story: one entity owns the infrastructure, and they adjust the gate so more traffic flows through it, hoping you don’t notice that they’re also the ones selling the fuel.
Someone will say the price drop is a trap. They will note that the introductory rate expires on December 31, 2026, and the new rate doubles the price. Five months from now, Google might hike the cost just as you’ve trained your whole team to rely on the tool. Someone will look at the benchmarks and say the new Flash model is still behind GPT-5.6 Luna on the DeepSWE 1.1, and that Luna’s context window does not bloat as fast when you crank up the reasoning level.
They will point out that the previous model, 3.6 Flash, was already good enough for many tasks. But you do not need to win every benchmark to change how you work. I looked at the Hacker News thread. One user mentioned that when they asked for an image of a pelican on the default setting, they got an ambitious bird, but the bicycle it was riding had two wheels on one side.
That is a failure of execution, not a failure of imagination. The real value is not a perfect pelican; it is the ability to say, "Write a landing page, generate the layout, create the interactive charts, and summarize the annual report," and get five distinct, working tools back in fewer than thirty seconds. The more interesting number is not the Elo score. It is the cost.
Before this model, you paid a premium to treat the AI like a junior engineer. Today, you can treat it like a library of scripts. You can run it on a laptop that costs ten thousand dollars instead of a server rack that costs a million. That is the gap between owning the warehouse and owning the drill.
The price is low, but the habit is permanent.
The price is low, but the habit is permanent. That was the lesson from the cheap, fast model that is now rewriting how we think about software. Now, step off the chip and onto the factory floor. Japan is sitting on a paradox: a country that invented the robot is running a version of the AI experiment at walking speed.
According to a report from the OECD released at the end of last year, only 8.4 percent of Japanese workers use AI as part of their job, compared with peers in the United States and the United Kingdom. Austin Xu, co-founder of the US start-up Kuse AI, opened a Tokyo office recently to sell his systems, and he finds the resistance baffling. "There are organisations where process and consensus culture genuinely slow things down," Xu says. "Where tolerance for AI mistakes is close to zero, especially in anything client facing.
Some would rather leave a role unfilled than let a machine handle it." This is not just shyness; it is the structure of the firm. Prof Parrisa Haghirian of Kyoto University agrees. Japanese companies, she argues, are highly sensitive to error, uncertainty and reputational risk. Since generative AI is still not fully reliable, it is mainly used for low-risk tasks—writing, summarising, or information gathering—rather than for core operations or decision-making.
The gap between the headline and the reality is stark. The Japanese government last year passed the AI Promotion Act, a light-touch regulation intended to encourage investment, and Tokyo has pledged to become "the world's most-friendly country for developing and utilizing AI." Yet, the Ministry of Finance claims that 75 percent of companies now use the technology, up from 11 percent five years ago. Prof Yasushi Ogasawara at Meiji University cuts through the optimism: "Although the Japanese like playing with gadgets such as smartphones, digital literacy is low here." He points to a report from earlier this year showing Japan has almost 800,000 IT professionals missing from the workforce. When a country is short the very people needed to keep the lights on, the choice to slow down looks like survival, not laziness.
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.
The Japanese companies were slow because they were running on fumes, and now they are out of fuel. When a country is short the very people needed to keep the lights on, the choice to slow down looks like survival, not laziness. Today we are talking about a different kind of fuel, and a different kind of survival. Nirmal Purja died on July 30th, aged 43.
The Gurkha mountaineer and Netflix star performed momentous feats in the Death Zone. Those are the words of The Economist. That is all I have. I do not know what he ate for breakfast, or which mountain he fell on, or who was standing next to him.
I only know that a machine was built to endure the most punishing environment on Earth, and that machine ran out of power in a rockfall. To put it simply, the Death Zone is above eight thousand metres. There is not enough oxygen there for a healthy person to think straight, let alone climb. The body turns inward.
It shuts down the things you need to survive the climb and keeps the ones you need just to stay alive. The more important question is why a human being would do this on purpose. You might be right that it is madness. Someone will say that when you are paid by the hour to walk up a hill, you walk up the hill; when you are paid in adrenaline to walk up a hill, you walk up the hill until you fall off.
The obvious objection is that he knew the risk and he took it anyway, and that is exactly what makes the news report odd. According to The Economist, he died in an avalanche. That is the sum total of his story. The headline tells you he was famous and strong, and it tells you the mountain killed him.
It does not tell you that for years, he held a list of records that the Nepalese government had no record of. It does not tell you that he was asked to go home and told to wait in line. It does not tell you that he decided he would not wait. The line you were given does not say he was obsessed, or that he was trying to prove a point, or that he was trying to save his country.
It only says he performed momentous feats in the Death Zone. That is a weird phrase to put on a man who died doing what he did. It implies the feats were the point, not the death. Fine.
Perhaps the feats were the point. But if you perform a feat in the Death Zone, you are not just climbing a mountain; you are asking a machine to do a job it was not built for. You are asking the body to carry the machine up, and the machine to carry the body down. There is a real case that the Death Zone is not a place for machines.
There is a real case that it is a place for gods. And there is a real case that no human being is a god. Nirmal Purja was thirty-three years old when he set his first record. He was thirty-eight when he set his last.
He died at forty-three. I do not know which record broke first, or if he ever saw the view. But I do know that he was in the Death Zone, and that he did not come down. Be good out there.