The deal, in plain numbers
Google and Fervo Energy announced a power purchase agreement on September 1 covering 396 megawatts, with an option to expand by close to 600 additional megawatts, bringing the total toward nearly 1 gigawatt by June 2030. The power comes from Fervo's Cape Station enhanced geothermal systems project, a GeoCluster development in Utah, and Google expects it online in 2028. Fervo CEO and co-founder Tim Latimer framed the agreement as validation of the underlying technology: 'This agreement reinforces that EGS is ready to power the next generation of computing infrastructure.' Google's Lucia Tian, director of advanced energy technologies, tied the deal to a specific data center ambition, saying the project 'will drive meaningful economic benefit to the local community and help catalyze long-term energy cost reductions.'
The structure is worth noting on its own. Google built this as a floor with an expansion option attached, letting it scale the commitment up as its Utah data center plans firm up without renegotiating from scratch. That option structure has become a template across hyperscaler power deals this year, letting buyers lock in priority access to a developer's pipeline without overcommitting capital before demand is confirmed. For Fervo, it converts a single customer relationship into a multi-year revenue base that can be used to raise project financing for the drilling and plant buildout Cape Station still needs.
Why enhanced geothermal, and why now
Enhanced geothermal systems work by drilling into hot dry rock and circulating fluid through engineered fractures to bring heat to the surface, rather than relying on the naturally occurring steam reservoirs that limit conventional geothermal to a handful of geographies. The pitch to a hyperscaler is straightforward: unlike solar or wind, an EGS plant runs continuously regardless of weather, and unlike nuclear, it does not carry a decade-plus licensing and construction timeline. For a data center operator that needs power available around the clock, every hour of the year, geothermal sits in a narrow but valuable category alongside gas and nuclear as a genuine baseload option, without the emissions profile of the former or the regulatory friction of the latter.
Enhanced geothermal remains a young commercial technology, with far less large-scale operating history behind it than gas turbines or nuclear reactors carry. Cape Station's 2028 online date means Google is committing capital years ahead of any certainty that the drilling economics hold at the scale this deal implies. Fervo's track record makes the wager credible, and it remains a wager all the same. Enhanced geothermal has not yet been proven at gigawatt scale anywhere, and Cape Station's expansion toward that number over the next four years will be the first real test of whether the technology's unit economics survive the jump from pilot to grid-scale asset.
A pattern, not a one-off
This is Google's third publicly disclosed transaction with Fervo. The relationship started with Fervo's Project Red pilot in Nevada in 2023, a proof-of-concept scale deployment that validated the drilling approach at a small footprint. Google followed that with a 115 megawatt power purchase agreement in Nevada in June 2024, its first commercial-scale commitment to the technology. The September 2026 Utah deal is an order of magnitude larger and geographically distinct, signaling that Google has moved past evaluating whether EGS works and is now treating it as a repeatable procurement category it intends to scale across multiple states.
That progression matters more than the headline number. A single large power deal can be opportunistic, tied to one specific site or one favorable utility relationship. Three deals across three years, two states, and an order-of-magnitude increase in scale each time is a company building a supply chain, not chasing a press release. For competitors watching Google's power strategy, the message is that geothermal has graduated from experimental to strategic inside at least one hyperscaler's energy procurement function, and the other major cloud providers will be evaluating whether to follow.
The two-year gap enterprises should actually plan around
The 2028 online date is the detail easiest to skim past and the one that matters most for anyone reading this as a capacity signal. This deal does not add power to the grid this year, next year, or the year after. It reserves power that will start arriving in roughly two years, with the full expansion option not exercised until 2030 at the earliest. Treat this the same way you would treat a hyperscaler's real estate acquisition rather than its capacity announcement: it tells you where the company plans to build, not where it can run your workloads today.
For enterprise technology leaders evaluating Google Cloud region roadmaps, particularly in retail and commerce businesses that care about proximity to Mountain West or Western US customer bases, this is a legitimate signal to start conversations with your Google account team now about anticipated Utah capacity, pricing implications of geothermal-backed power costs, and whether committed use discounts in that region will reflect the capital Google is putting into securing baseload power years in advance. Getting ahead of that conversation before the region opens is worth more than waiting for the general availability announcement.
What this says about the power constraint, broadly
The fact that a company with Google's balance sheet is locking in power four years before it can use all of it tells you how binding the power constraint on AI infrastructure has become. Chip supply gets the headlines, but for the hyperscalers themselves, securing gigawatt-scale, always-on power has become the longer lead-time bottleneck, and deals like this one are evidence that the major cloud providers now plan power procurement on the same multi-year horizon they plan chip procurement. Enhanced geothermal is one of several answers to that constraint, alongside gas turbines, nuclear restarts, and utility-scale battery storage, and Google's willingness to bet on a less mature technology suggests the conventional options are not moving fast enough on their own.
For CIOs and CTOs, the broader implication is that cloud capacity availability in any given region is now downstream of power procurement decisions made years earlier by developers you have never heard of. Understanding your cloud provider's power supply chain, not just its data center footprint, is becoming a legitimate part of vendor risk assessment for any workload where regional capacity and pricing stability matter over a multi-year contract term.
The roadmap implication
If your infrastructure planning has any dependency on Google Cloud capacity in the Mountain West over the next three to five years, put this deal on your vendor risk calendar and revisit it annually as Cape Station's construction progresses. If you have no near-term dependency there, the more useful takeaway is procedural: start asking every major cloud vendor what their power procurement pipeline looks like in the regions you actually use, not just what capacity they claim is available today.
Geothermal, gas, and nuclear deals are all being announced at a pace that outstrips actual construction timelines right now, and distinguishing a real, financed, permitted project from an announcement designed to signal ambition to investors is a skill your infrastructure and procurement teams need to build. Fervo's track record with Google across three deals gives this particular announcement more credibility than most, but the 2028 delivery date is a reminder that even credible power deals are a multi-year bet, not a capacity guarantee you can build a near-term roadmap on.
