The Marginal Buyer of Energy Is a Data Center — and Baker Hughes Already Holds $39.4 Billion of Those Orders

Baker Hughes chairman and CEO Lorenzo Simonelli told CNBC on Nov. 5 that he expects energy demand to keep growing for a decade. Six days earlier, bp had reported third-quarter underlying replacement cost profit of $8.5 billion, up $2.7 billion from the second quarter — and its chief executive, Meg O'Neill, told the same network that bp must plan for a lower-price environment.

Both statements are true, and that is the entire argument. One describes a volume cycle; the other describes a price cycle. The panel disagreed about the barrel, not about the build-out, and the market has mostly priced the barrel.

The reason this cycle is not the usual boom-bust energy capex wave is the identity of the buyer. Two weeks before the Abu Dhabi panel, Venture Global signed power and grid frame agreements with GE Vernova and Baker Hughes to build up to 16 island-mode power units worth 11.52 GW behind the meter — enough to enable as much as 10 GW of data center compute, with the first purchase order expected by the end of 2026.

The buyer of record is not a car buyer, a shipping line or a petrochemical plant. It is a data center operator holding a 20-year power contract, and it is paying for the equipment with contracted cash flows rather than with a hope that AI capital spending keeps rising.

Demand converts into pricing power only if supply cannot respond. A new combined-cycle plant's lead time went from about 3.5 years in 2023 to roughly five years, and seven years for some heavy-duty frames. GE Vernova is now taking reservations for 2031 delivery, and expects to be more than halfway contracted for that year by the end of 2026.

That is the hinge of the whole trade. A 2027 pause in hyperscaler capital spending cannot un-book equipment whose revenue is recognized in 2028 through 2030 — and it cannot be cancelled by the company selling it, because the contract sits with an off-taker holding a 20-year power agreement.

Goldman Sachs' power-demand arithmetic shows how tight the gap is: US data center power demand climbs from 31 GW in 2025 to 41 GW in 2026 and 66 GW in 2027, while scheduled US capacity additions go from 8.5 GW realized last year to 13.6 GW in 2026 and 36.3 GW in 2027. Data centers alone take 8.5% of US peak summer demand in 2027 against 4.1% today. Added load is running roughly three times ahead of added supply — and the equipment that closes the gap is sold out to 2030 and beyond.

Baker Hughes is the cleanest listed expression of the cycle, and the disconnect between its numbers and its share price is at its widest. Industrial & Energy Technology orders rose 42% year over year to $5.9 billion in Q3 2026, and the segment's remaining performance obligations hit a record $39.4 billion — up 23% year over year, with $16.8 billion of it in services that are paid whether or not the tape cooperates.

The other half of the same quarter explains the share price. GAAP net income fell to $411 million from $609 million a year earlier, and the company paid $14.2 billion of net cash for Chart Industries, leaving net debt at roughly 2.8 times trailing EBITDA. Two business models now sit inside one ticker: a levered oilfield-services business that will feel any 2027 oil oversupply, and a power-equipment business with a contracted service annuity attached to 4.9 billion dollars of data-center awards.

The case that this cycle is fragile starts with the AI capital spending line itself. In late October, Microsoft trimmed its 2026 capex forecast to $170 billion from $175 billion, Alphabet held guidance at $195 billion to $205 billion while reporting negative free cash flow of more than $9 billion, and Meta raised the bottom of its range to $135 billion. Supply delays in third-party accelerators push some compute capacity into the first half of 2027.

Those are trims and delays, not a cancellation — and they land on the one part of the chain that was never the funding source.

J.P. Morgan's October baseline still has the five largest US hyperscalers spending about $900 billion on AI capex in 2026, raised from $736 billion, and $5.5 trillion by 2030, with data centers taking 9% to 17% of US electricity demand by 2030 against 3% to 4% today. The 2027 debate is about the slope of that line, not its direction. Meanwhile the equipment suppliers were marked down 30% to 42% while their order books set records.

Two balance-sheet moves already done should not be lost in the noise: the Chart Industries acquisition closed on July 16, 2026 for $13.6 billion with $325 million of annualized synergies promised by year three, and the sale of Waygate Technologies to Hexagon closed on Nov. 2, 2026 for about $1.45 billion in pre-tax proceeds. Portfolio surgery, not operating improvement, is what turns the record backlog into free cash flow — and it is why a levered balance sheet is less alarming than the 2.8 times net debt to EBITDA figure implies.

Over the long run the thesis is a volume-and-price call on gas-fired generation. Baker Hughes' own outlook has US data center capacity adding roughly 100 GW by 2035, which it estimates requires about 18 billion cubic feet a day of incremental gas supply, and global LNG capacity reaching about 800 million tonnes a year by 2030 before rising toward 950 million with new final investment decisions. That is a decade-long pull on gas molecules, not a one-quarter pull on equipment.