Why Hyperscalers Are Pouring $200B Into Nuclear PPAs
Why Hyperscalers Are Pouring $200B Into Nuclear PPAs
Last year, 92% of US GDP growth came from a single line of spending — data center construction. AI hyperscalers quadrupled their capex in two years and are now burning roughly $400 billion annually. Strip that line out and the rest of the economy grew 0.1%.
TL;DR: When gas prices push up wholesale electricity, nuclear revenue and margins rise automatically while fuel costs stay locked. Hyperscaler-signed PPAs already exceed $200B in commitments. This is a 5-to-10-year structural setup, not a 3-month trade.
It's not $20B — it's over $200B
Hyperscalers stopped waiting and started signing. Microsoft, Amazon, and Meta have all inked deals. Add up the long-term power purchase agreements, equity stakes, and co-development arrangements, and there's over $200 billion of nuclear-linked contracts on the table. The market hasn't fully priced this — that's the part I keep coming back to.
When I covered the four-tier nuclear stack last November (mining → fuel/enrichment → innovation reactors → operating utilities), PPAs were a trickle. Six months later, the flow changed.
The merit-order machine
The whole thing hinges on how wholesale electricity prices get set. The marginal power plant clears the market, and in most US regions that plant runs on natural gas. Gas is the swing producer.
Natural gas generates 43% of all US electricity. The on-again, off-again situation around the Strait of Hormuz has already sent gas prices vertical, and wholesale electricity follows. Layer summer cooling demand on top and you get the highest wholesale prints we've ever seen.
For nuclear, this is asymmetric upside. Fuel contracts are locked for 20+ years, so costs don't move when gas spikes. Revenue and margins rise automatically while the cost base sits still. That's what "all ships rise with the tide" actually means in this stack.
What the numbers say
- Spot uranium: ~$90/lb — already above the level needed to incentivize new mine development.
- Centrus Energy backlog: from effectively zero to $2B+ in two years.
- BWXT commercial backlog: up 85% YoY by end of 2025.
- Cameco's share of Westinghouse profits: +26% in 2025 alone.
Risks I'm tracking
First, grid connection delays. Microsoft's Three Mile Island restart PPA slipped from 2027 to 2031 — the contract holds, but revenue recognition pushes right.
Second, litigation. Talen Energy is waiting on a Q3/Q4 2026 ruling that touches its 1.9GW Amazon colocation structure. The outcome reshapes the model.
Third, NuScale's securities fraud class action. The lead plaintiff deadline passed April 20, 2026. I stay conservative until settlement.
What to watch
- How operator earnings respond during the summer wholesale price peak
- Centrus's next federal contract awards
- Oklo's criticality milestones (target: July 4, 2026)
Supply constraints hit first. Demand follows. The companies that own the leverage in between are the ones that print over the cycle.
More in this Category
February 28, 2022: The Day the Dollar Got Switched Off, and the Sovereign Gold Rush That Followed
February 28, 2022: The Day the Dollar Got Switched Off, and the Sovereign Gold Rush That Followed
Since Russia's $300B reserves were frozen, central bank gold buying has run at roughly 60 tons a month versus 17 tons pre-2022. I trace the four-year shift that almost nobody is pricing in.
The $29 Trillion Gold Stock and Why New Supply Is Effectively Captured by Central Banks
The $29 Trillion Gold Stock and Why New Supply Is Effectively Captured by Central Banks
About $0.5T of new gold is mined each year, but central banks buying ~60 tons a month are absorbing nearly all of it. I unpack the price-discovery mechanics of an asset whose supply cannot scale.
Fortinet (FTNT): Why the ASIC Moat Becomes the 2026 Cybersecurity Story
Fortinet (FTNT): Why the ASIC Moat Becomes the 2026 Cybersecurity Story
Fortinet's in-house ASIC chips give it structural cost-per-watt leadership in firewalls, while AI-driven attacks are forcing Fortune 500 security budgets up double digits. The post-pullback breakout to all-time highs is the classic reaccumulation pattern institutions love.
Next Posts
The AI Datacenter 4-Tier Framework — How to Allocate Capital Across a $7T Buildout
The AI Datacenter 4-Tier Framework — How to Allocate Capital Across a $7T Buildout
McKinsey projects $7T in cumulative global datacenter spending through 2030 — roughly 23% of 2025 US GDP. The way I make sense of it is to break that spend into four physical layers and own names in each.
Two Paths to AI Cooling — Modine vs Carrier
Two Paths to AI Cooling — Modine vs Carrier
NVIDIA's next-gen AI racks pull 132 kW. Air cooling can't handle that, and Modine and Carrier are the two most-cited liquid cooling plays. They are very different bets.
Three Hidden AI Infrastructure Plays — Belden, Mueller, Bloom Energy
Three Hidden AI Infrastructure Plays — Belden, Mueller, Bloom Energy
Fiber, copper, and behind-the-meter power. One or two steps removed from the GPU and largely absent from analyst models — but without them, the datacenter is just an empty building.
Previous Posts
Four Mega-Caps in 48 Hours: The Hinge Week for 2026
Four Mega-Caps in 48 Hours: The Hinge Week for 2026
Over 20% of the S&P 500 reports next week, and Microsoft, Meta, Amazon, and Apple are clustered into a single 48-hour window. Here's why this is the hinge week for 2026 — and what to listen for in each print.
How to Bet on AI Without Single-Stock Risk — SMH, DTCR, and Three Names
How to Bet on AI Without Single-Stock Risk — SMH, DTCR, and Three Names
How to get AI infrastructure exposure without single-stock risk — SMH (+27% YTD), DTCR (+30% YTD), and three individual names (APLD, IREN, NBIS) broken down one ticker at a time.
Beats Aren't Enough — The Week Guidance Beat EPS
Beats Aren't Enough — The Week Guidance Beat EPS
Texas Instruments rallied 10% this week. IBM and ServiceNow dropped double digits. All three met or beat consensus. Here's why the outcomes split — and the framework I'm taking into next week's mega-cap prints.