Houston, 23–27 March. S&P Global’s CERAWeek returned to the George R. Brown Convention Center under a theme — “Convergence and Competition: Energy, Technology and Geopolitics” — that read on the printed programme like it might be a slogan and turned out, by day two, to be an accurate description of the room. The most talked-about arrivals on the guest list were not the usual national-oil-company chief executives or the ministers-of-energy corridor. They were the semiconductor and chip-company leaders who, for the first time in CERAWeek’s history, sat on the panels as first-class participants rather than adjacent observers.
That is the tell. Energy and technology have converged. The AI capex build-out is now the pacing constraint on the US and European electricity systems. The oil and gas industry has read the load curves and rewritten its own capital allocation to serve them. And the two most consequential single announcements of the week — TotalEnergies terminating its US offshore wind book and reinvesting the money in Gulf and shale oil and gas, and Fervo Energy filing for an IPO on a geothermal cost curve — were, in different ways, both stories about how energy capital is repricing for an AI-electricity decade.
Chip companies walk into the oil conference
The presence of semiconductor and chip-company leaders on the CERAWeek main stage was the single most-remarked-on shift of the week. The frame was not academic. The rise of AI and data-centre infrastructure is reshaping electricity demand, capex, and the geography of new build in the US, Europe, the Gulf, and East Asia. Hyperscaler datacentre load is now the fastest-growing block on many US utility forecast plans; the semiconductor players are the ones sitting between the model developers and the grid.
The conversation was more concrete than in 2025. Capacity-firming contracts, long-duration PPAs, behind-the-meter generation, on-site nuclear (small modular reactors), geothermal baseload, and the very expensive economics of new transmission — all showed up in the technology-and-energy tracks as procurement problems with named counterparties, not as future-of-energy panel abstractions.
TotalEnergies fires its offshore wind book
The single loudest news of the week was TotalEnergies announcing, on Monday, that it will terminate roughly $1 billion of US offshore wind projects and redirect the capital into upstream oil and gas. The redirection is specific: $928 million into upstream conventional oil in the US Gulf and shale production, plus continued development of the Rio Grande LNG plant in Brownsville, Texas.
The strategic message the majors are collectively sending is that US offshore wind, under the current political and regulatory environment, is not underwritable at the return thresholds their boards will approve. That is a different message from the one European majors were sending three years ago. The retreat from US offshore wind is not universal — there are still committed developers finishing projects that were far enough along not to unwind — but the fresh-capital story has visibly changed.
The follow-through, less noisy but arguably more consequential, is what happens to the offshore-wind supply chain (turbines, foundations, cable-laying vessels, ports) that had scaled to serve the US pipeline and is now overcapacity relative to the near-term project book. Some of it will redirect to European projects. Some of it will absorb the pain.
Fervo Energy files to go public
The other headline announcement was that Houston-based Fervo Energy — the enhanced-geothermal company that has spent five years demonstrating that horizontal drilling, closed-loop reservoir engineering, and dispatchable geothermal power are commercially viable — filed for an IPO. Fervo’s rapid rise from startup to unicorn to public-market candidate is the shape of the story clean-technology investors have been waiting for: a company with a real cost curve, real projects generating real electrons, and a technology thesis that gets more credible as data centres shop for firm, low-carbon, 24/7 baseload.
The reception on the CERAWeek panels was warm and, notably, technical. The residual scepticism was about drilling costs at scale, subsurface risk in less-well-characterised basins, and the availability of rigs and crews trained for horizontal geothermal wellwork. None of those is a small problem. All of them are the kind of problem that has answers, which is why Fervo’s IPO filing lands with the credibility it does.
Data centres, nuclear, and the “AI electricity” story
The recurring subject of CERAWeek 2026 was electricity — specifically, where the incremental terawatt-hours of AI-data-centre load come from through the back half of the decade. The choices are limited and each has its costs. Natural gas peakers and combined-cycle build gets you predictable capacity and unattractive emissions. Nuclear — legacy fleet uprates, SMRs, and (further out) advanced reactors — is the cleanest firm answer but has schedule and cost risk. Enhanced geothermal, of which Fervo is the flagship, is the most credible new baseload category to emerge in a decade. Long-duration storage, transmission investment, and behind-the-meter renewables round out the portfolio.
The hyperscalers, notably, are participating in the answer directly. Microsoft, Google, Amazon and Meta are named on power purchase agreements across almost all of the above categories, and are increasingly writing cheques for merchant generation projects that would previously have depended on regulated-utility offtake. The old separation between energy buyer and energy financier has collapsed at the top of the AI market.
Energy security, affordability, and decarbonisation — the trilemma with sharper edges
The energy trilemma — reliability, affordability, decarbonisation, choose all three — was, as at every CERAWeek, the framing device for the policy panels. The 2026 edition sharpened the edges of the argument. Heightened geopolitical risk, supply-chain disruption in critical minerals and long-lead equipment, and the run of extreme-weather grid events across three continents have collectively elevated energy security to the top of the policy stack in most of the participant countries. Decarbonisation has not fallen off the agenda. It has been forced to solve for security and affordability as constraints rather than as trade-offs.
Houston’s own energy ecosystem — an integrated stack of hydrocarbons, LNG, petrochemicals, and the emerging clean-energy transition build — is unusually well-positioned to argue for the pragmatic version of the trilemma. That was the tone of the closing-plenary conversations, and — for an industry that spent much of the 2010s speaking past itself — the alignment was noticeable.
What CERAWeek 2026 actually moved
- Chip and semiconductor leaders are now first-class participants in the energy conversation. AI electricity demand is repricing capital allocation.
- US offshore wind lost a major sponsor. TotalEnergies’ pivot signals a re-underwriting of the sector.
- Enhanced geothermal has an IPO candidate. Fervo’s filing is a milestone for the category.
- Firm low-carbon baseload is the AI-data-centre procurement problem. Nuclear and geothermal are the credible answers.
- The energy trilemma is now security-first. Decarbonisation solves for it, not around it.
Open questions
Do other majors follow TotalEnergies on US offshore wind, or is this a company-specific pivot? Does Fervo’s IPO clear at a valuation that pulls the rest of the geothermal cohort forward? And do hyperscaler power purchase commitments actually field the terawatt-hours the AI capex build-out will need, on the timeline the compute buyers are planning around?
Bottom line: CERAWeek 2026 was the year the energy conversation became indistinguishable from the AI conversation, and the AI conversation had to grow up about grid reality. The capital is repositioning — into hydrocarbons where the majors underwrite them, into geothermal where the cost curve now works, into nuclear where the schedule can be lived with, and out of categories where the political and regulatory risk has become uninvestable. The follow-through is measured in megawatts delivered and permits cleared — not in pledges.
