Fervo Energy announced on September 1, 2026 that it signed a 396-megawatt geothermal power purchase agreement with Google for the Cape Station GeoCluster in Utah. The company says the power is expected to come online in 2028 and is designed to support a potential Google data center, with an option for Google to expand the agreement by about 600 MW for a total near 1 GW by June 2030.
That is a hyperscaler-sized deal, not a purchase order most small businesses will ever sign. But it makes a practical point visible: AI, cloud storage, analytics, automation, and data-heavy applications depend on physical infrastructure. The cloud still has a power bill. It also has data centers, grid connections, cooling, regional capacity, and contracts behind it.
Why this matters beyond Google
The International Energy Agency has been tracking the same pressure from a broader angle. Its 2026 analysis says large technology companies are driving a surge in data center investment, and that AI-focused data center electricity use grew faster than overall data center consumption in 2025. The report also notes that more energy-intensive AI use cases, including reasoning and agentic tasks, can use far more energy than simple text generation.
For owners and operators, the takeaway is not to panic about every AI feature or cloud invoice. The better takeaway is to treat AI and cloud expansion as an operating decision. If a vendor proposes a new AI workflow, larger data archive, analytics platform, backup expansion, or cloud migration, the question is not only whether the tool works. The question is what demand it creates and who is watching the assumptions.
The business decision hiding in the infrastructure
Many organizations approve technology work in pieces: one SaaS renewal, one AI pilot, one storage upgrade, one integration, one reporting dashboard. That can make the total infrastructure footprint hard to see until the bill changes, performance degrades, or a provider says a different tier, region, or architecture is now required.
A useful review connects the proposal to usage. How much data will move? How often will AI features run? Which systems will depend on the provider's region, identity platform, API limits, or uptime? What happens if the service slows down, changes pricing, or moves a workload to a different plan?
This is where the Google and Fervo announcement becomes more than energy news. If the largest providers are securing dedicated capacity years ahead, smaller customers should be more careful about accepting vague cloud and AI forecasts. A low-friction feature can still create long-term cost and dependency.
Questions to ask before approving the next AI or cloud project
- What usage assumptions drive the estimate? Ask for expected users, transactions, storage growth, API calls, AI runs, retention periods, and peak demand.
- Which costs can change with demand? Look for usage-based compute, storage, backups, egress, AI tokens, premium support, add-ons, and compliance logging.
- What region or provider dependency is being created? Confirm where the service runs, what happens during a regional issue, and whether the provider supports your continuity needs.
- Who reviews the bill after launch? Assign ownership for monthly variance review before the service becomes part of normal operations.
- What is the exit path? Ask how data can be exported, how integrations can be unwound, and what contract terms matter if the tool disappoints.
A practical next step
Before the next cloud or AI approval, ask your IT provider, MSP, software vendor, or internal team for a one-page capacity and cost note. It should list the expected workload, main cost drivers, owner of the monthly review, continuity assumption, and trigger points for revisiting the design.
That note does not need to be fancy. It just needs to make the hidden infrastructure visible before the organization gets comfortable depending on it. AI plans can move quickly, but the meter is still running.
Sources and further reading
