Alphabet paid $4.75 billion for an energy company, while Meta signed agreements for 6.6 gigawatts of nuclear power with Vistra, Oklo, and TerraPower. The Trump administration insists on having the PJM system operator conduct an emergency capacity auction exclusively for data centers.
This is not energy news — it is news about artificial intelligence (AI). And it shows where the bottleneck of the industry has shifted.
Chips were the first bottleneck in the development of AI and brought huge profits. Now the limitation has shifted to more basic things: electricity and metals. While large tech companies face constraints from power grids and multi-year equipment delivery timelines, the cost shifts to companies that can quickly deliver electricity — and to supply chains that can tie it all together.
Gas producers are monetizing 'isolated' molecules.
The most telling AI projects are not in Silicon Valley. They are located in areas where there is a lot of gas, but transmission capacity is lacking.
Chevron is advancing a 2.5 GW autonomous gas power plant project in West Texas, designed to serve data centers. The first electricity is expected to be generated by 2027 with a possible expansion to 5 GW. The company is collaborating with Engine No. 1 and GE Vernova on a broader model aimed at 4 GW in several regions of the U.S.
ExxonMobil now has over 2.7 GW in its data center power supply portfolio. The initial concept for 1.5 GW — a gas power plant with carbon capture built exclusively for hyperscale load — remains in the pre-design stage. But in December 2025, Exxon announced an additional 1.2 GW project with NextEra Energy for a data center campus in the southeastern U.S.
The business model has changed. Gas is not just fuel. In the deployment of AI, it becomes a contract service product — sold not as molecules, but as reliability. Chevron is in 'exclusive negotiations' with an unnamed 'leading' data center operator. The buyer is not acquiring gas — they are acquiring schedule certainty.
Hyperscale companies are moving towards vertical integration.
The deal flow in January 2026 shows a strategic shift: large tech companies are moving from buying electricity to owning generation.
Alphabet's acquisition of Intersect Power for $4.75 billion (announced on January 2) represents the first case of a hyperscale company fully acquiring a large clean energy developer. Intersect's portfolio — 3.6 GW of solar and wind energy, 3.1 GWh of battery storage — gives Google direct control over generating assets instead of relying on power purchase agreements.
Meta's nuclear initiative is no less aggressive. On January 9, 2026, the company announced nuclear agreements for 6.6 GW: 2.1 GW from existing Vistra plants in Ohio and Pennsylvania, a campus of modular reactors with a capacity of 1.2 GW with Oklo, and two Natrium reactors from TerraPower with rights to six more. Meta is now 'one of the most significant corporate buyers of nuclear energy in American history.'
The logic is defensive. The counterparty risk in power purchase agreements increases when each hyperscale company competes for the same megawatts. Ownership eliminates this queue.
Where capacity markets exist, shortages manifest in bills. The December 2025 PJM capacity auction produced record-high prices for the 2027/28 delivery period at $333.44 per MW per day — the maximum allowed by the FERC price cap. The market observer later calculated that data centers accounted for $6.5 billion (40%) of the total auction value of $16.4 billion.
Copper is the physical layer of AI.
Even if AI becomes more efficient, deployment remains a wiring issue.
According to the latest estimates, data centers could add about 500,000 tons of copper demand annually by 2030. But transmission and distribution are a big story. One detailed analysis predicts that the demand for copper from transmission and distribution could reach 7.1 million tons per year by 2040.
Copper prices on the LME hit a record high of $13,387 per ton on January 6, 2026, although they subsequently retreated to $12,800 in mid-January. The 42% increase in 2025 marked the best annual result for copper since 2009.
The challenge of mining is not only geology, but also time: large new projects often require a decade or more from discovery and permitting to meaningful production. Supply lags behind demand shocks for years, not quarters.
The new reality of infrastructure investments.
The PJM auction and the market observer's analysis have turned the conversations about the AI energy crisis into real financial obligations. Alphabet's acquisition of Intersect turned the strategy of owning its generation into a completed deal.
The main thesis is that electricity and wiring have become the new bottlenecks in the development of AI. Companies that can solve these constraints — gas producers with fast projects, copper suppliers, and integrated energy solutions — gain an advantage in a market where demand exceeds supply.
AI Opinion
Analysis of historical patterns shows a striking similarity to the railroad boom of the 1840s: back then, too, the first profits went to locomotive manufacturers, while the real money was made by owners of coal mines and steel mills. Machine data analysis reveals an interesting pattern — every technological supercycle ends in a 'battle for infrastructure' when digital giants begin to acquire physical assets.
From a macroeconomic perspective, the AI boom creates structural inflation. Copper is used in 65% of all industrial products — from cars to refrigerators. Gas is needed for heating homes and producing fertilizers. Data centers compete for resources with the basic needs of the economy, which could trigger a price spiral far beyond the technology sector.
#AI #AImodel #energy #Write2Earn
