Sebastian J. Kuhnert

Brad Gerstner says at least 10 to 15 gigawatts of the expected 2027 AI build may not come online after the blue wave, against a consensus 43 gigawatts next year versus about 38 built to date “Well, I would say not specifically on the gigawatts in Texas, but what I said at the All-In Summit: the consensus forecast, which we’ll call the SemiAnalysis Dylan Patel forecast for next year, is a total of 43 gigawatts coming online in 2027." "Now, just to put that in perspective, the total gigawatts online at the end of ‘26, the cumulative in the United States, is about 38. So the consensus forecast is we’re going to bring online more next year than all of the compute that we’ve built to date." "I said on stage, and it got a lot of blowback on Twitter at the time, that we may end up bringing half of that online in 2027. And part of what I was concerned about was the growing backlash, even from guys like Greg Abbott." "But now with this blue wave, I think it’ll slow it down even more. So I think we could see at least 10 to 15 gigawatts of the expected build not come online, which, you know, that’s why I think it’s going to be really challenging for the AI infra trade." "We have the dollars to pay for it. But imagine what we’re going to hear, Chamath, for the next two weeks out of the governors, out of Speaker Hakeem Jeffries or Leader Schumer. They ran on slowing down AI. That was their principal agenda.”

thanks for sharing, and honestly for engaging openly on here frankly, I have trouble squaring your underlying assumptions with what I'm seeing my best read is that you believe: i) SPCX is the only operator adding AI compute at meaningful scale through 2028 ii) nobody else catches up, so SPCX durably earns 1 yr paybacks on hundreds of billions of capex is that right? if so, probably tough to find a common denominator, but maybe the following thoughts help triangulate the divergence > compute monetization rent/GW is set by aggregate demand vs aggregate supply. for rent to 3x ($13bn today to your $40bn) while demand 3x (your assumption) aggregate supply has to stay ~flat through 2027. SPCX alone adds 7.7 GW in FY27, so flat would imply the rest of the market shrinks by the same amount? however, we also know aggregate compute is forecasted to grow ~2x in 2027 though, so if demand only 3x, price cannot 3x? so are you saying, the most conservative aggregate compute forecasts are off by a factor of 2x? > distressed compute a neo sitting on a GW that rents for $40bn/yr against ~$45bn of cost doesn't go bust, it gets refinanced or picked up by a better operator. granted, I agree SPCX well positioned, however discounted hardware at scale is what a market looks like after rent has collapsed. so either rent is $40bn and there's no cheap hardware, or there's cheap hardware and rent isn't $40bn? but even assuming your input/output divergence, why does SPCX get that compute? jensen has every incentive to buy back and redeploy distressed GB300s, monetizing the same chips twice before handing them to one customer at a discount? > financing as equity investors, where does the conviction come from that financing/ credit markets stay open for SPCX specifically, and that SPCX keeps its IG rating while operators of the very compute SPCX is raising for go bust? > useful GPU life annualized Q428 depreciation vs cumulative SI capex through Q328 backs out ~11 yrs, which seems aggressive? > other more ancillary, but I still don't understand where >$100bn of orbital compute capex sits, or the grok/ cursor compute (every GW is rented out externally in the model?)

Universal High Income The most likely outcome

Do a kickflip!