Everyone is staring at the chip.
The smart money already moved one layer down — to the power socket.
In 2026 the world's four largest cloud companies — Amazon, Microsoft, Google, and Meta — are pouring an astonishing sum into artificial-intelligence infrastructure. Analysts put combined hyperscaler capital expenditure north of $600 billion this year, a jump of roughly 77% over the previous record. It is, by most measures, the single largest corporate capital-expenditure cycle in recorded history.
That number alone hypnotizes the market. But at Book Coupling we don't read headlines — we read structure. And the structure is quietly telling us something the headlines haven't caught up to yet. The bottleneck has moved. It is no longer the chip. It is no longer even the money. It is electricity.
1. The biggest bet in corporate history
Nvidia now captures an estimated 57 cents of every dollar hyperscalers spend on AI hardware, and its most recent quarterly guidance — around $91 billion — beat Wall Street's pre-announcement consensus by more than $12 billion. On paper, the supercycle looks unstoppable.
Hyperscaler AI capex → $600B+ (≈ +77% YoY)
Nvidia's share of each $1 → $0.57
Nvidia quarterly guidance → ~$91B
2. The Catch-22 investors keep whispering about
Here is the trap. If a hyperscaler pulls back on AI spending, investors may reward the discipline in the short term — but the company risks losing the AI race for a decade. If it keeps spending, margins compress and the market starts asking when, exactly, all this capex turns into profit. Hyperscalers, as one analyst put it, can't win with investors either way. That tension is the real reason the word "bubble" keeps surfacing — not because AI is fake, but because the return timeline is uncertain.
3. The real chokepoint: the grid can't keep up
Here is the part the headlines miss. You can order a million chips. You can raise half a trillion dollars. But you cannot conjure electricity out of a spreadsheet.
- New high-capacity grid connections in major data-center hubs now face 4–7 year wait times.
- US grid interconnection queues have exceeded 1,500 GW of pending projects.
- Up to half of planned data-center projects face delays from grid capacity and transformer shortages.
- Data-center electricity demand is on track to more than double, driven overwhelmingly by AI.
Read that structurally and the conclusion is blunt: the constraint on AI is no longer silicon or capital. It is power. And whoever controls the scarce resource in any supply chain controls the profit.
4. Big Tech's pivot to nuclear — the tell
Watch what the giants do, not what they say. Meta, Microsoft, Amazon, Google, and Oracle have all signed nuclear-power agreements — reviving retired reactors, contracting existing plants, and funding small modular reactors (SMRs).
The signal in the numbers:
- SMR pipeline tied to data-center operators grew from 25 GW (end-2024) to 45 GW (April 2026) — up 80% in 16 months.
- Meta alone signed three nuclear deals in early 2026 totaling over 6 GW (Vistra, Oklo, TerraPower).
- The industry is shifting from relying on the public grid to building its own dedicated power.
When the richest companies on earth start buying reactors, they are not making an environmental statement. They are telling you where the next scarcity — and the next fortune — lives.
5. The investor's map: follow the current
Most investors crowd into the most visible layer — the chip. But value in a supply chain tends to accrue to the scarcest link. Here is how the current actually flows:
- Compute — the crowded, obvious layer (GPUs, accelerators).
- Power generation — utilities, independent power producers, nuclear & SMR developers.
- Grid & equipment — transformers, switchgear, high-voltage cabling, the physical bottleneck.
- Fuel & cooling — uranium, natural gas, and the liquid-cooling systems AI density now demands.
You don't have to pick a single winner. You have to understand that the money is migrating down the stack — from the chip everyone sees toward the power almost no one is pricing correctly yet.
The AI era will not be won by whoever has the most chips.
It will be won by whoever has the most power.
Read the structure, not the headline — and you were early, not late.
Where do you stand?
Is the AI capex boom a supercycle or a bubble? And is nuclear the real winner of the AI age?
Drop your take in the comments — the sharpest one shapes next week's piece.
Sources:
Morgan Stanley — Powering AI ·
Futurum — AI Capex 2026 ·
Yahoo Finance — Hyperscalers & investors
Figures are as reported by third parties and may be revised. Educational content only — not investment advice.
GOLDEN KEYWORDS
AI capex 2026, hyperscaler capital expenditure, AI infrastructure spending, Nvidia AI chips, AI data center power, data center electricity demand, AI power bottleneck, grid interconnection queue, SMR small modular reactor, nuclear power data center, Meta Vistra Oklo TerraPower, AI bubble debate, uranium stocks 2026, nuclear energy stocks, power grid investment, AI energy crisis, AI 데이터센터 전력, SMR 소형모듈원자로, 원자력 수혜주, AI 거품 논쟁, 하이퍼스케일러 capex, 엔비디아, AIデータセンター 電力, 小型モジュール炉 SMR, 原子力 関連株, AIバブル, ハイパースケーラー 設備投資
Confession: when I first saw the $600B capex number, my gut said "bubble." Then I looked at where the giants are actually putting their money — reactors, not just chips — and I changed my mind. You don't sign 20-year nuclear deals for a fad.
But I could be wrong, and that's the point. So tell me:
Supercycle or bubble — which side are you on, and what's the ONE data point that would flip you?
If power really is the bottleneck, where does the smart money go: utilities, nuclear/SMR, grid equipment, or uranium?
And the contrarian take I want to hear: what's the strongest argument that this whole AI buildout disappoints?
Best counter-argument in the comments makes it into next week's piece. Come prove me wrong. 👇
고백하자면, 6,000억 달러라는 capex 숫자를 처음 봤을 때 제 직감은 "거품"이었습니다. 그런데 거인들이 실제로 돈을 어디에 넣는지 — 칩만이 아니라 원자로 — 를 보고 생각을 바꿨죠. 20년짜리 원전 계약은 유행 때문에 맺지 않으니까요.
하지만 제가 틀릴 수도 있고, 그게 핵심입니다. 여러분께 묻습니다.
① 슈퍼사이클인가 거품인가 — 당신은 어느 편이고, 당신을 뒤집을 단 하나의 데이터는 무엇인가요.
② 정말 전력이 병목이라면, 스마트머니는 어디로 갈까요 — 전력회사, 원자력/SMR, 전력망 장비, 아니면 우라늄?
③ 그리고 제가 가장 듣고 싶은 반대 의견 — 이 AI 투자 붐이 실망으로 끝난다는 가장 강력한 논거는 무엇인가요.
가장 날카로운 반박이 다음 주 글에 실립니다. 저를 틀렸다고 증명해 보세요. 👇