Wednesday, November 4, 2026

@DeepSailCapital Nov 4

Deep Sail Capital on X

Just a bit of my thoughts on AI: If AI buildout is really materially slowed down it’s a death sentence to American Economy / our AI pole position in the world. AI datacenter buildout saved us from a recession a few years ago, it is the only thing holding us back from one right now. An extremely bad outcome is China takes the lead in AI, US businesses use said Chinese AI to cut US jobs and the money saved is shipped to China, so we are net negative as a country. Now on to why I don’t think AI build cadence will be materially changed and it’s all posturing / the easy trade for tourists. Texas has already stopped new data centers back in August, and that's the most important fact the market missed today. 1/ Abbott paused it in August. ERCOT stopped energizing new data centers of 75 MW or more and shelved a ~200 GW study batch. TCEQ stopped issuing permits. Hinojosa's moratorium will keep the existing pause in place on New permits instead of lifting it. She also said it targets new permits, not projects already under construction. So that just means next project goes to a friendly locale, not Texas. 2/ Ohio was already moving. DeWine froze the data center sales-tax exemption for new projects back in June. Acton's conditional moratorium is a pay your own power costs, use union labor, build on brownfields. That's a cost hit, not a ban outright. 3/ Congress can't ban anything. The Sanders/AOC moratorium bill would face a Trump veto through Jan 2029. 4/ What's already built or nearly built keeps going: • Stargate Abilene: 4 of 8 buildings live, the rest due this month • Fairwater Atlanta: ~636 MW running • Rainier Indiana: ~1 GW running • Hyperion Louisiana: not in play at all About 56.7 GW is under construction nationally and largely outside the pledges. 5/ Demand didn't change this week: • $ANET Q3 revenue $3.66b, +58.5% y/y, $360m above guide • $CLS guiding 2027 to $40b, roughly double 2026 • $MOD data center sales +95%, record backlog, a $4b+ deal for 2027–29 • Hyperscalers are on pace for $803b–$828b of 2026 capex and they're guiding 2027 higher 6/ The market half-figured it out. Texas-heavy builders and power names fell 7–23% ($HUT -23%, $VST -15%). Suppliers with national backlogs barely moved ($ANET -0.4%, $NVDA -0.8%). $NUAI is a zero now for sure. 7/ Where it's real: new 2027–28 capacity in TX/OH/GA gets slower and pricier, so it will shift to different locations. New projects go to friendly locales, TX/OH/GA lose out. 8/ AI component suppliers are still maxed out for the next few years, that demand doesn’t change, just where they ship it to does. 9/ Neoclouds still suck. Ain’t nothing gonna change that.