Google signed a 50-megawatt power purchase agreement with Quaise Energy for Project Obsidian in central Oregon, with deliveries as early as 2030, Quaise announced on November 17, 2026 (Business Wire release). Syndicated Wall Street Journal coverage the same day put the deal’s value at more than $500 million over 15 years. Deschutes County economic development staff materials describe Project Obsidian as a 50 MW net engineered geothermal facility in Newberry, Oregon, with a 15-year PPA and phased capital plan; county planning materials cite June 2030 as a commercial-operation target.
For AI builders, the headline is not charity — it is firm megawatts for Oregon data centers as PJM queue fights and monitor-reported capacity adders show grids struggling to keep pace with load.
TL;DR
| Field | Reported value |
|---|---|
| Buyer | |
| Seller / developer | Quaise Energy (Houston-based) |
| Capacity | 50 MW |
| Deal value | >$500M over 15 years (WSJ via syndicated wire) |
| Target COD | June 2030 (county staff report) |
| Total project capex | Quaise cites ~$350M project investment in trade press |
| Technology | Superhot rock + millimeter-wave deep drilling |
What Quaise is building
Quaise’s public description: drill conventionally to about 2.5 miles, then extend with a gyrotron-style millimeter-wave rig that vaporizes rock toward 3 miles depth where rock exceeds ~300 °C (572 °F). Water returns as steam for turbine generation.
That is different from shallow geothermal or Google’s Finland nuclear pairing. It targets 24/7 firm power without burning gas on site — attractive for training clusters that cannot follow solar curves alone.
Why Google needs more deals like this
Hyperscalers already buy wind, solar, nuclear uprates, and VPP flexibility. Geothermal adds capacity factor without shipping LNG. Lucia Tian, quoted in trade summaries of the WSJ story, framed Google’s need as building an ecosystem of advanced clean technologies because AI load scale exceeds what incremental grid imports alone can serve.
Same-day grid news on PJM Fast Energization shows interconnect studies, not electrons, remain the bottleneck in other regions. Oregon PPAs do not fix Virginia queues — they anchor regional portfolios.
County tax and jobs footprint
Deschutes staff analyzed a five-year RRED zone tax abatement request tied to roughly $173M taxable project value in year-one tables — a local jobs and revenue story adjacent to AI, not a token price driver. Construction timelines in the staff report run through permitting, interconnection, financing, and phased 20 MW + 30 MW capacity blocks.
Risks and what is unverified
- First-of-kind drilling at commercial scale — schedule slip is plausible.
- $500M+ PPA math depends on undisclosed energy price curves; we cite WSJ reporting, not a Google PDF.
- Regulatory geothermal permits and grid interconnection agreements are not fully public in sources we read.
What small teams should take away
You will not buy Quaise electrons directly. You will see Oregon region colo marketing clean firm power once COD nears. Compare contracts against PPA-heavy claims in Amazon Ohio builds and Google Malaysia spend.
How superhot rock differs from enhanced geothermal systems (EGS)
Traditional enhanced geothermal fractures rock at moderate depths. Superhot rock targets much hotter dry rock beyond conventional drill bits, which is why Quaise advertises millimeter-wave vaporization instead of only mechanical grinding. The company argues that once you reach sufficient temperature, energy density per well rises enough to justify extreme drilling R&D.
Google’s earlier Fervo partnerships cover a more familiar horizontal well + stimulation playbook in other states. Project Obsidian is Quaise’s bet that deep heat beats wide footprints for 50 MW blocks serving steady AI load.
Timeline from county documents to electrons
Deschutes materials outline phases: site control, interconnection, permitting, engineering, procurement, construction, and commercial operation with Phase I at 20 MW contracted capacity and Phase II at 30 MW. Even with a signed PPA, non-code risks — forest permits, water rights, transmission upgrades — can move COD.
AI capacity planners should treat 2030 as a planning anchor, not a guaranteed ramp for training schedules.
Economics: $500M PPA vs $350M capex
Trade press quotes Quaise on ~$350 million total project investment while the PPA exceeds $500 million over 15 years. The gap is normal: PPA dollars include energy payments across years, carrying costs, and potentially REC-style attributes, while capex is construction spend. Without public $/MWh, outsiders should not back-solve a levelized cost from headlines alone.
Oregon grid and data center load context
Oregon already hosts hyperscaler regions benefiting from hydro-heavy grids. Adding firm geothermal diversifies beyond intermittent renewables and reduces gas peaker exposure when training jobs run flat. It does not automatically lower interconnect queues in other ISOs — see PJM Fast Energization for how study speed differs by RTO.
Environmental and community lens
Geothermal still disturbs land, uses water in some designs, and requires induced seismicity monitoring programs. Local tax abatement debates in Deschutes mirror other data-center tax deals nationwide. Communities may support jobs while resisting transmission line expansions — watch county hearings beyond the PPA press cycle.
Comparison table: Google clean firm options
| Source | Firmness | Novelty | Status |
|---|---|---|---|
| Nuclear uprates | High | Moderate | Operating plants |
| Fervo geothermal | Medium-high | Moderate | Projects under build |
| Quaise superhot | High if tech works | High | COD ~2030 |
| Grid imports | Varies | Low | Always available |
Questions energy teams will ask Google partners
- Balancing: Does geothermal output pair with existing hydro flexibility?
- Additionality: Is this capacity new build versus purchased credits only?
- Curtailment: Who backs down if BPA transmission limits export?
- Price pass-through: Will colo tenants see PPA costs in $/kW-month quotes?
We lack public answers in sources reviewed; flag them for your next sustainability diligence call.
Builder takeaway without a PPA
If you rent 8 GPUs in Oregon, you will not sign the Quaise deal — but you should ask hosts whether 2030 firm supply is in their renewable portfolio narrative. If not, your carbon accounting may still rest on grid-average factors after 2030 growth.
Historical Google geothermal bets
Google’s clean energy team has iterated on enhanced geothermal for years, treating each PPA as R&D plus procurement. Fervo projects proved horizontal drilling techniques; Quaise bets on depth and temperature instead. None replace immediate grid imports where memory fabs strain timelines, but they reduce long-run gas reliance for always-on training halls.
What could delay June 2030 COD
Drilling risk is the obvious delay: bit failure, unstable boreholes, or regulatory pauses after seismic events. Less obvious: interconnection for 50 MW in Central Oregon may need transmission upgrades if regional load jumps. Tax abatement appeals can also slow financing closes even when a PPA is signed.
Due diligence checklist for colo buyers
Ask providers: (1) contracted vs grid-mix renewables after 2030, (2) additionality documentation for Quaise, (3) curtailment rights during maintenance, (4) PPA pass-through clauses, (5) backup fuel if geothermal trips. Without contract exhibits, treat marketing slides as unverified.
Federal permitting context
National debates on who pays for transmission and overseas hub siting fights show electricity is geopolitics. Geothermal PPAs help one region; they do not replace PJM queue reform elsewhere.
Related context on explainx.ai
- Google Voltus 1 GW VPP on PJM
- PJM Fast Energization filing
- SK Hynix memory supply chain
- OpenAI Grid Protectors flexible demand
- Pax Silica Philippines hub protests
- Nvidia infrastructure supply chain testimony themes
Scenario planning for GPU cluster operators
Model three 2030 Oregon capacity scenarios in your capacity model: (A) Quaise on time at 50 MW firm, (B) two-year slip with continued BPA hydro imports, (C) partial derate forcing gas peakers online during training peaks. None of these scenarios change today’s spot GPU prices, but they change five-year lease renewals when hosts renegotiate power-adjusted rates. Document assumptions in internal runbooks the same way you document failover regions for inference.
Why explainx.ai tracks power and silicon together
AI news splits into model drops and megawatt deals, but rent rolls merge them. A team that ignores PPAs may overbuild VRAM while under-provisioning rack power. Cross-reading geothermal headlines with Nvidia supply testimony keeps planning holistic.
Glossary for searchers
PPA means a power purchase agreement: a long-term contract to buy electricity at agreed terms, not a one-time grant. COD means commercial operation date, when the plant can sell power under the PPA. Superhot rock means geothermal heat at unusually high temperature accessed by deep drilling. Newberry, Oregon is the volcanic field hosting Project Obsidian in county planning documents we reviewed.
Retail readers sometimes ask whether a Quaise headline means cheaper ChatGPT. It does not. Consumer prices track competition and margins, not a single Oregon PPA. The builder link is indirect: lower stranded-power risk for Google regions may stabilize API rate cards years out, but nobody has published that causal chain in filings we read. Watch Google’s sustainability report cycle for first-party confirmation beyond trade press. We will update this article when Google publishes one.
Primary sources: Quaise Energy Google PPA announcement, Deschutes County staff report on Project Obsidian, and Quaise Energy, accessed November 17, 2026. >$500M deal value via syndicated Wall Street Journal reporting the same day.
