My essay this morning was titled, "Why is no one talking about oil?" Of course, everyone is talking about oil. Oil discourse gushed onto the timeline this weekend, and crude prices spiked to nearly $120 a barrel as a broadening war in Iran threatens both transportation routes and production. The geopolitical and economic analyses are flowing. But what does this mean for AGI timelines? And a lot of people in the AI world are sort of tuning all of this out because they see recursive self-improvement, AGI, ASI, the buildout as more important. And I just wanted to sort of reality check the AI supply chain to understand how does oil actually affect data center construction, AI production, token pricing — like, is there any effect? My conclusion was that it's very moderate, but there are some interesting effects in the financial markets that are probably the bigger takeaway. But it's still interesting to hear about, like, yes, oil actually is used in the production of AI, at least a little bit.
So, power has been at the forefront of the AI pushback. Like, everyone's worried about these local energy prices, these electricity prices increasing near the data centers, and it's become like a political issue. But pain at the pump might become a bigger story as gasoline prices spike, and that has been pain at the pump. Oh, it's been the most tangible sign of inflation. It moves so quickly. You know, one jitter in the economy...
And it's a huge component. You know, people on the coast, people in tech don't have a good sense for this, right? If gas — for a lot of people, gas could quadruple and they wouldn't really notice it.
But if you actually look at the average American and how gasoline fits into their budget, it's a meaningful component of their monthly budget. So they feel it super intensely.
Variable cost. There are so many different ways where a lot of Americans go on driving vacations that obviously is directly impacted by gas prices. And then also just psychologically, there's something about filling up at the gas tank where you see the number ticking up, and you're doing that on a every week basis or so, that it's just so visceral. It's this thing that you have to stop and then go experience and watch the money flow out of your account like in real time. It's very visual. It's very interactive.
Yeah. I remember — I must have been probably 18 at this point where I would just go and I would like — for a long time I'd just put — you know, prepay for a certain amount. Let's see how much I got. And I felt like really — I felt like the king of the castle. I'd just put my card down and let it run up and see how far it would go.
No, no, totally. So, pain at the pump's been like a big broad economic issue. So visceral — like the actual phrase has a lot of power in politics and in attack ads, like, "Your gas prices are going up under this person, that's why you got to vote for me." This is common. But in AI circles, the discussion's been much more focused on RSI now — is the new acronym that everyone's focusing on, not AGI. AGI is here. We know artificial general intelligence. We passed the Turing test. The models can do things generally intelligently. But can they recursively self-improve? Are they RSI? Are we in RSI now? Is it coming? Is this going to be a fast takeoff? Is this going to be a slow takeoff? Well, something's taken off. Dylan Patel said, "Being in SF is like being in Wuhan right before the pandemic. Something is happening. It's going to hit everywhere, but so few people know it." So he's sort of echoing that something big is happening mentality. This idea of like recursive self-improvement, being AGI-pilled. A lot of smart people agree there's definitely something happening. There's still this question of like what is sticky? Where will diffusion take longer? What will be sort of AGI-resistant or ASI-resistant for a variety of reasons, even if they're completely irrational and you could get a better thing from a computer, but people like the human version or whatever. The irony is that George Hotz hitting the timeline to raise money makes me more bullish on acceleration.
Right. Because if not, he's obviously not historically been a huge fan of venture capital.
Well, he's not raising from traditional VCs. He specifically said he wants to raise from like basically like friends and family type of round. He does want 20 million. We have this in the timeline somewhere. Where is this post from George? I'll pull it up if you want to keep it. Tiny Corp. He said that he's raising $20 million to buy a building. He's going to buy a building. Here it is. So he said — this is from Tiny Corp, the makers of Tiny and the Tiny Box. "If Tiny Corp was raising $20 million at $200 million valuation, who'd be interested? Business model is basically this: buy an 11.5 million building with 5 megawatts of power. Link in our Discord. Wait for AMD to launch the RDNA 5 96 gigabyte cards mid-2027. Pre-order 3,000 cards. Hopefully we can negotiate for $2,500 each. Build 500 $20,000 tiny boxes with six of the cards in each box. Run all the Chinese LLMs. Make $600,000 per month revenue selling tokens on OpenRouter. Market depth is there. This is 1% of OpenRouter. Improvements to tiny yield revenue improvements. Due to how power is priced in Oregon, it's only like $50,000 for the electric bill before the four megawatts — before they price for peak, not usage. We get like 3 cent kilowatt-hour power, three cents per kilowatt hour. We can also make $100,000 per month leasing colocation space to comma. Building and cards paid off in three years max. Investment made back. Low risk of being undercut since we're using consumer GPUs and running the cheapest colo you can believe. If someone chill wants in, I do it. I'm not going to fake hype. I'm not going to hype fake tech, but demand for tokens is going to skyrocket. Look at the OpenClaw install numbers. With crazy good optimizations, we could potentially get 3x more from the machines. And we have electricity for 3x more machines, 5.4 revenue per month, then continue to scale from there, custom chips, etc." He's starting a Neocloud. Are you starting a — Yeah, he's going to be serving tokens.
So that was, you know, incredibly bullish. You know, demand is definitely there. At the same time, he sort of took the other side of like, this is different, this is crazy. He says there's no AGI. There's no magic threshold. "You guys see auto research change the random seed from 42 to 137 and OMG, it's AGI, it's over. Yet you critique the junior engineer for the same stuff. The cost of development is falling. Overpaid engineering struggles to compete. That's the story." And he says, "Humanity has been recursively self-improving for centuries. So everything new is old."
I don't know. I obviously have a huge amount of respect for George, but at the same time, saying there is no artificial general intelligence when you're also arguing that engineers are struggling to compete with new technology that is replacing them — how is this not artificial intelligence competing with...
