TL;DR (read this first)
Hyperscalers need 34–44 GW of new carbon-free, hourly-matched power by 2030 to keep their 24/7 pledges.
The world is building <4 GW of true "firm" clean capacity (nuclear + geothermal) by then. Do the math.
Their only option: pay massive premiums for "synthetic firm" — intermittent renewables + storage + demand-response logs that prove hourly greenness.
In 2025, these bundles already trade at $160–$280/MWh — 5–8× the price of a 2023 solar PPA.
Supply is still measured in single-digit GW. Demand is effectively unlimited.
The Physics Problem No One Can Buy Their Way Out Of
| Company | 2024 Load | 2030 Projected Load (AI-driven) | Required New Clean Firm GW | Realistic New Atomic Supply by 2030 |
|---|---|---|---|---|
| 25 TWh | 70–90 TWh | 8–10 GW | <1 GW | |
| Microsoft | 20 TWh | 80–110 TWh | 9–12 GW | <1 GW |
| Amazon | 30 TWh | 100–130 TWh | 11–14 GW | <1 GW |
| Meta | 15 TWh | 50–70 TWh | 6–8 GW | <0.5 GW |
| Total | 90 TWh | 300–400 TWh | 34–44 GW | <4 GW |
There is no spreadsheet trick that fixes this gap.
Synthetic Firm Power = The Only Product That Works
It is not a new technology.
It is a new financial wrapper around existing pieces:
| Ingredient | Source (Pieces 2–5) | What It Contributes |
|---|---|---|
| Intermittent renewables | Wind / solar farms | Base green electrons |
| Grid-scale batteries | Jupiter, Broad Reach, etc. | Time-shifting + injection logs |
| Flexible demand response | Bitcoin miners (Piece 3) | Curtailment logs |
| Consumer VPP fleets | 1M+ Powerwalls (Piece 5) | Distributed curtailment/injection |
| Audit-grade dispatch logs | All of the above | The actual certificate |
Bundle them together → a contract that guarantees:
"Every hour of every day, this data center was matched 1:1 with clean, available megawatt-hours — proven by logs."
That is synthetic firm power.
The 2025 Price List (Real Closed Deals)
| Seller | Buyer | Size | Effective Price (all-in) | Multiple vs 2023 Solar PPA |
|---|---|---|---|---|
| Iris Energy + battery partner | Microsoft | 300 MW | $160–$220/MWh | 5–7× |
| Unnamed Texas developer | 200 MW | $180/MWh | 5.5× | |
| Tesla Energy VPP + storage | Amazon (rumored) | 500 MW | $200–$280/MWh | 6–8× |
| Hut 8 HPC + curtailment bundle | Oracle Cloud | 250 MW | $190/MWh | 6× |
These are not pilots. These are multi-year, take-or-pay contracts signed in 2025.
Who Is Already Shipping It (and Getting Rich)
| Player | 2025 Capacity Selling Synthetic Firm | Premium Earned | Notes |
|---|---|---|---|
| Iris Energy (IREN) | 860 MW (100% renewable) | $9.7 B Microsoft AI deal | Curtailment logs + GPUs |
| Hut 8 (post-spin) | 1+ GW pipeline | $300 M debt at 8% WACC | Logs as collateral |
| Jupiter Power | 1.2 GW Texas | 2.8× revenue vs arbitrage | Sodium-ion + stacking |
| Tesla Energy | 1 GW+ VPP fleet | $10 M+ owner payouts → upstream sales | Consumer logs at scale |
They are not waiting for permission. They are printing the new certificate today.
The Bottom Line
- There is no more cheap green power for hyperscalers
- There is only expensive synthetic firm power — and even that is impossible to buy at scale
- The primitives (curtailment logs, injection logs, consumer fleets) already exist
- The buyer already exists
- The price is already 5–8× higher than anyone budgeted in 2023
- All that's missing is someone to industrialize the bundling
Key Takeaways
- Hyperscalers need 34–44 GW of new clean firm power by 2030. The world builds <4 GW
- Synthetic firm = renewables + storage + flexible load + auditable logs
- 2025 deals already trade at $160–$280/MWh — 5–8× old PPA prices
- Iris Energy, Hut 8, Jupiter, Tesla are the only meaningful suppliers today
- Supply = low single-digit GW. Demand = unlimited wallet
- The certificate is built from the exact same logs we've followed since Piece 3
Background Reading
- Microsoft–Iris Energy $9.7 B AI hosting deal (Nov 2025) → ir.irisenergy.co
- Google 24/7 Progress Report 2025 → google.com
- BloombergNEF "Corporate Clean Energy Purchasing 2025" (paywall, but widely quoted)
- Energy Web Foundation white-paper on granular certificates (2025)
Next post:
The missing layer nobody is building yet — the company that will own the audit rails, post the collateral, and capture 20–40% margins forever.
We finally name the species.