high convictionactive · updated 2026-05-30T00:00:00.000Z
AI capex sprint → power-supply gap → grid-component + materials bottleneck → nuclear/copper/transformer beneficiary cascade
Hyperscaler capex hit $725B-$830B for 2026 (+79% YoY). Power, not GPUs, is now the binding constraint — most announced gigawatts aren't being built because the grid components (transformers, turbines) are themselves on backlog. Nuclear emerges as the consensus baseload answer; copper deficit and US-China critical-mineral exposure compound. Beneficiaries split into three chains: nuclear (CCJ, CEG, BWXT), copper miners (FCX, SCCO), grid infra (transformer / cable makers).
The chain
1
Hyperscaler 2026 capex is $725B-$830B (+~80% YoY), with Microsoft $190B, AWS $200-230B, Google $180-190B, Meta $145B. AI capex is now >2% of US GDP and ~75% of GDP growth.
chamath-palihapitiya in 2026-05-01-all-in-podcast-openai-misses-targets-codex-vs-claude-elon-vs: "$725 billion in CapEx guidance in 2026 from but four companies. Amazon, Microsoft, Google and Meta. Amazon leading the pack with 200 billion, 190 billion each for Microsoft and Google, 145 billion for Meta. You add Grok, you add OpenAI and some other players to these plans... we are going to see a trillion dollars in build out over the next year"
david-sacks in 2026-05-01-all-in-podcast-openai-misses-targets-codex-vs-claude-elon-vs: "this year I think we were supposed to have 660 billion of hyperscaler capex up from 350 last year. I think there's now the new estimate is going to be over 700. So this is again, it's more than 2% of GDP... in the last quarter AI was 75% of GDP growth"
From 2026-05-11-autoresearch-macro-semis-ai-infrastructure-may-2026: "TrendForce (May 6) revised the combined 2026 CapEx of the top 9 cloud providers to $830 billion (+79% YoY)"
From 2026-05-15-autoresearch-csp-capex-2026-2027-durability: All four top hyperscalers RAISED 2026 capex in Q1 2026 earnings (Amazon ~$200B, Microsoft ~$190B, Alphabet ~$190B, Meta $125–145B). Cycle confirmed supply-constrained not demand-constrained: Pichai Q1 2026: "Cloud revenue would have been higher if we could meet demand." Alphabet enterprise backlog $462B (nearly doubled QoQ); Microsoft commercial cloud backlog $250B+. 2027 consensus: Evercore + BofA both >$1T combined capex; Goldman $1.15T cumulative 2025–2027.
2
Power, not GPUs, is the binding constraint. OpenAI's miss is explained by power-supply lag, not demand. Hyperscalers are signing 2x spot-rate forward energy contracts (e.g., Microsoft / Three Mile Island).
david-friedberg in 2026-05-01-all-in-podcast-openai-misses-targets-codex-vs-claude-elon-vs: "Everything in this market is power constrained. The reason that these folks may miss a number or a forecast have nothing to do with demand. It is entirely 100% due to the supply of the power necessary to generate the output token"
david-friedberg in 2026-05-01-all-in-podcast-openai-misses-targets-codex-vs-claude-elon-vs: "When Microsoft convinced the owners of Three Mile island to turn their nuclear site back on... their forward purchase agreement was for more than 2x the prevailing spot rate for energy"
andrew-feldman in 2026-05-21-odd-lots-why-cerebras-ceo-andrew-feldman-built-the-world-s: "Business today is constrained by data centers. And that's the grand irony... what are we all constrained by? Buildings. Data centers right now are everybody's constraint in the entire industry. Powered buildings. So real estate... that will not change for the next 15 or 18 months, for sure."
3
Power-side build-out is itself bottlenecked by grid components (transformers, natural gas turbines, reciprocating engines) — backlogs build, less than half of announced gigawatts is actually under construction. Most stuck on supply-chain delays and permitting.
david-friedberg in 2026-05-01-all-in-podcast-openai-misses-targets-codex-vs-claude-elon-vs: "backlogs build up of not just the access to the power, but then the componentry that's actually necessary. Not just recips and not just NAT gas turbines, but now you're talking about transformers and all the actual tactical grid infrastructure... If you look at the actual amount of gigawatts that are under construction. We have a huge mismatch now. People have announced all these projects, Jason, but less than half of it is actually being built"
From 2026-05-08-all-in-podcast-elon-s-anthropic-deal-the-next-ai-monopoly (Chamath): "you have about 9 gigawatts that are supposed to come online this year. Almost 50% of it now is being protested. More than likely if history holds, most of that will get turned off so they will get even more supply constrained." Adds a *political* dimension to the construction-shortfall claim that wasn't yet quantified in earlier sources — half of planned 2026 capacity facing organized opposition. The deployed-capacity number is the right thing to track on this mechanism; the announced-capacity number now overstates buildout by ~2× even on optimistic assumptions
From 2026-05-08-all-in-podcast-elon-s-anthropic-deal-the-next-ai-monopoly (Brad): "this is not like organic hyper local protests by people in a community that aren't being spurred on. This is highly organized activists that are moving across the country to stir up trouble in the exact same way they did to stop all fission reactors being built 30 years ago in America." Brad's funding-conspiracy frame is unverified — but the *empirical* claim that the anti-data-center movement is geographically coordinated rather than locally organic is testable; look for the named groups, their funders, their staffing overlap with anti-nuclear orgs of the 1970s-80s. If true, the same forcing function that produced the US nuclear deployment freeze is now operating on AI compute. Also reinforces the Texas-vs-NY/CA divergence: "Electricity bills are going up in the places that are not building data centers, New York and California, because they haven't built any supply on the grid. In Texas, where you're building the most data centers in the country, electricity costs are going down"
From 2026-05-08-all-in-podcast-elon-s-anthropic-deal-the-next-ai-monopoly (Chamath + Calacanis): Pulte Homes + Span partnership puts "many data centers with Nvidia GPU clusters beside every home" — distributed-compute architecture as an *adaptation* to the centralized-buildout constraint. Calacanis extends: "the powerwall with compute in it... distributed system from home to home... Starlink also gives him the ability to do distributed compute to people's homes." If the centralized buildout is permanently capped at ~50% of plan, edge / residential GPU deployment becomes a parallel deployment channel that routes around the protested grid additions. Watch Span (private), Pulte Homes (PHM), and the residential-grid intersection (Tesla powerwall + Solar) as derivative beneficiaries
From 2026-05-15-autoresearch-exec-capex-statements-may-12-15: Jensen Huang (May 14, 2026): **"30,000 truckloads to build a single 500MW data center"** (excluding the power plant). Called it "the single largest infrastructure buildout in human history." Told electricians and plumbers: "this is your time." Data center construction spending +138.6% YTD through November 2025 (to $53.7B). Huang's framing quantifies the physical material intensity of the bottleneck — the constraint is not just permitting and grid components but the physical logistics of construction at this scale.
christian-bruch in 2026-05-19-podcast-columbia-energy-exchange-speed-to-power-christian-bruch-on-siemens-energy: **"You will see in the next two to three years some shortfalls in the industry of just digesting this growth."** First-party CEO-level corroboration from Siemens Energy (one of three global gas-turbine suppliers) that the supply-chain compression is operationally locked in for 24-36 months. Bruch frames "speed to power" as a *new fourth dimension* of energy planning alongside sustainability/reliability/availability — "because that is determining at the end the dominance of AI or not."
christian-bruch in 2026-05-19-podcast-columbia-energy-exchange-speed-to-power-christian-bruch-on-siemens-energy: **"If you have an interest rate which is 4% instead of 2% then you have a problem to roll out that energy infrastructure."** Adds a financing-cost dimension to the bottleneck — capital-intensive renewables and grid carry more rate sensitivity than fossil, so policy can de-risk via backstops. Cross-link supply-shock-inflation-persistence: if central banks stay tight, the cascade itself is dampened by financing-cost pass-through. The mechanism is not just physical but capital-cost-sensitive at the deployment margin.
christian-bruch in 2026-05-19-podcast-columbia-energy-exchange-speed-to-power-christian-bruch-on-siemens-energy: "We push even more investments into the U.S... invest a billion in the US and it's probably not end there. It will probably continue in particular to build out factories for grid infrastructure and also obviously more factories for gas services." Siemens Energy operationalizing the demand thesis with $1B+ US factory expansion for grid + gas turbines. Tradeable signal that suppliers are now committing capex into the bottleneck (capacity addition, not just price rationing).
christian-bruch in 2026-05-19-podcast-columbia-energy-exchange-speed-to-power-christian-bruch-on-siemens-energy: **"You have thousands of gigawatts of capacity not connected to the grid, because they're waiting for the grid connection. So you have to use it smarter... we launched this year what we call Neuedra... we trying to create a capable mind off the grid, that the grid is capable at the end to think independently, autonomous, to really route more power through the lines which are existing."** Software-based grid optimization (Siemens Neuedra) is the optionality leg — if it works at scale it partially defers new physical buildout. Watch as a *tension* with the build-more thesis: not a contradiction, a complement.
4
Materials side compounds the bottleneck: a 304kt copper deficit was forecast for 2025 (widening in 2026); AI DCs head toward 500kt/yr copper by 2030; only 70% of 2035 demand is covered by existing+planned mines. US depends on China for 100% of 15 critical minerals.
From 2026-05-11-autoresearch-macro-energy-critical-minerals-may-2026: "a 304,000-tonne copper deficit was forecast for 2025, widening in 2026, with only 70% of 2035 demand covered by existing and planned mines"
From 2026-05-11-autoresearch-macro-energy-critical-minerals-may-2026: "AI data centers are headed toward consuming 500,000 tons of copper annually by 2030"
From 2026-05-11-autoresearch-macro-energy-critical-minerals-may-2026: "the US depends on China for 100% of 15 critical minerals"
5
Therefore three investable cascades: (a) nuclear baseload (CCJ uranium, CEG / Vistra / TLN utilities, BWXT / OKLO SMR designers); (b) copper miners (FCX, SCCO); (c) grid infrastructure (transformer manufacturers, cable makers, high-voltage equipment).
From 2026-05-11-autoresearch-macro-energy-critical-minerals-may-2026: "These constraints produce at least three investable chains — copper miners (FCX, SCCO), nuclear (CCJ, CEG, BWXT), and grid infrastructure (transformers, cables)"
What would falsify this
- Step 2: If hyperscaler earnings beat without power-supply caveats and announced gigawatts come online on schedule, the power-as-binding-constraint framing is wrong.
- Step 3: If transformer / turbine backlogs clear within 12 months (suggests OEM capacity ramps faster than expected), grid-infra beneficiaries lose their scarcity premium.
- Step 4: If copper mine output growth surprises to the upside in 2026 (>5% YoY), the deficit thesis weakens.
Contradictions / tensions
- Sacks pushes back on the 'dot-com 2.0' framing: 'there's no dark GPUs today' — demand is real, not over-built. Chain remains a bull thesis.
- **Training cost commoditization (Chamath, 8090)**: chamath-palihapitiya in 2026-05-29-podcast-all-in-podcast-anthropic-s-digital-god-pope-vs-ai-job-loss: "There was an economic and capital moat to training that is going away...domain specific architectures at the silicon layer...Elon was like, we've rewritten the entire training complex in C and it's an order of magnitude increase...why would we stick to the $10 billion training runs when we can have the $10 million training runs?" If training costs collapse from $10B to $10M per run (domain-specific silicon + compiler efficiency), the *size* of step-1 capex in this chain is at risk — future builds may require less compute than the current $725-830B capex wave implies. Falsification test: track whether training compute per FLOP actually falls 10-100x by 2027 or whether scale remains the dominant competence signal.
- **Capex not a bubble — FCF still covers in aggregate (Chisholm, Fidelity data)**: denise-chisholm in 2026-05-29-podcast-the-compound-and-friends-what-if-it-s-still-early-with-denise-chisholm: at the 2000 bubble peak, corporate America spent 3.5-4x FCF on capex; today we're still under 1x in aggregate even including hyperscalers. The FCF-coverage signal argues current capex is sustainable rather than bubble-driven; corroborates Sacks and supports chain durability.
- Permitting reform (US permitting/Inflation Reduction Act follow-ons) could partly relieve the bottleneck.
- Hyperscalers' balance-sheet capacity to keep signing 2x-spot energy contracts is finite — at some point pricing power flips to the energy supplier rather than to hyperscaler beneficiaries downstream.
Implications
- The bottleneck is upstream of the GPU. INTC/TSMC/NVDA can be in supply, but if power isn't, the token-factory revenue doesn't materialize.
- Hyperscalers giving up free cash flow / buybacks to fund this cycle (Amazon FCF -97%, Google/MSFT/Meta -8 to -12%). Long-term valuation reframing follows (Friedberg: hyperscalers will look like 'big bulky industrials' in 5 years).
- Friedberg's actionable rule: 'follow the dollars going out of hyperscalers — those companies are underpriced.' Picks-and-shovels logic extends beyond semicap.
- Reinforces the orbital-DC chain (terrestrial-power-flat-to-orbital-dc-arbitrage) — power gap is real and structural, not a market head-fake.
Companies
noneConcepts
Nuclear as the only viable AI data center baseloadCopper supercycle: AI data centers and energy buildoutUS critical mineral independence
Open questions
none