Interesting...

Vals AI CEO @RayanKrishnan tells @JTLonsdale on @AmOptimistShow that America's AI lead over China comes down to compute. The US is drowning in chips compared to China, which lets it train stronger models and run inference at scale. The risk to that lead is China reaching recursive self-improvement first and running away with it. Joe Lonsdale: "If Chinese labs can distill all these capabilities and they're doing it well, how do we preserve the advantage in the US?" Rayan Krishnan: "There's two areas I keep coming back to. One is chips being a considerable asset that we have. I think it surprises most people, but we're drowning in compute compared to what China has. As long as that's the case, we will continue to train more capable systems, and we'll be able to diffuse the models much more effectively because we can host and run inference. The other one is really talent. If we can consistently attract and keep top talent, it will allow us to keep pushing ahead while human advantages exist. Once we cross into the space of recursive self-improvement, where models are responsible for producing the next successor version, they just get ahead much faster. So that would be the point to be much more cautious of. If China is showing an indication of getting to RSI before the US, it's possible they may run away with it in a way that we can't really control.”

Vals AI CEO @RayanKrishnan tells @JTLonsdale on @AmOptimistShow that Anthropic is on track for recursive self-improvement, AI building its own successor models, by July or August 2027, with Gemini on the steepest curve. Once that happens, AI progress speeds up and the only limits left are chips and energy, the same bottlenecks behind the AI capex trade on Wall Street. Rayan Krishnan: "We have our RSI index. We're collecting a set of proxies of the work that researchers do within these labs on pre-training, post-training and harness-level engineering, and comparing to human baselines to extrapolate how far away we are from the human frontier on these research tasks. Based on our last estimates, it seems like Anthropic will get to full RSI by July or maybe August 2027." Joe Lonsdale: "And if you look back six months ago, how far out did that seem?" Rayan Krishnan: "Actually, the pace has been fairly uniform. It's almost a linear trend over the last year. So six months ago, the trend line was still July, June. The steepest curve has been with Gemini. I think something has changed where they've prioritized RSI, and the most recent model release is a rapid jump. There are still considerable resource constraints around chips and energy. But I think what it'll look like is that model progress will speed up, and it will maybe even move into places where we understand the models less. There's less human intervention required to build the models, and so we're not as clearly dictating how that intelligence emerges.”

Elon named the shutoff. H100 spot rental is down about 57% from 2H23. Operators still depreciate servers over five to six years. The loan fails on the refinance date, not on demand. Recoverable value sits in energized megawatts. The GPUs mark like inventory.

Interesting.

Private credit is underwriting the AI buildout against chips that age out fast. Someone ends up holding that risk when the cycle turns.

Most tankers crossing Hormuz now switch their trackers off. The oil is still getting out, about 11.6m barrels a day per Kpler, roughly two thirds of the 17m that flowed before the war. The backup is a single Saudi pipeline across the desert to Yanbu on the Red Sea, pumping 5.8m b/d of its 7m maximum. Today the Houthis claimed a strike on the Aramco facility at Yanbu, where that pipeline ends. So far the oil is still flowing, and Brent closed Friday at $116.72, up 6% on the week. If that pipe holds, Saudi Arabia keeps cushioning the shock for everyone. If it breaks, there is no backup to the backup.

On SpaceX's Q3 call, @elonmusk's team said power, cooling and infrastructure now limit AI more than chips do, and SpaceX casts its own turbine blades because outside supply runs 12 to 18 months late. The BLS data shows the same thing one layer down. Assembly costs, not the chips themselves, are the real bottleneck right now. +130% on loaded boards vs +1.8% on microprocessors tells you where the margin pressure (and opportunity) actually sits. Is board assembly next on the list to bring in house?

SpaceX is the most vertically integrated company out there, and I do not think it is even really close.