Capital intensity gets a floor, a date, and a power bill attached
Capital intensity is now a scheduling question answered by utilities rather than by capital markets. The pass-through clauses protect AWS margin but shift the timing risk to customers, which makes energy contracts the place to look for the next capacity revision.
The capex conversation has shifted from how much to how it converts. Management now frames $142 billion as roughly one third of the eventual annual run rate with a decline after 2027, which is a destination map rather than a spend disclosure. The offsetting evidence is physical: gigawatts queued for interconnection and power purchase obligations itemized for the first time, with energy risk passed to customers through adjustment clauses.
- 2026 capital expenditure guidance was raised to approximately $142 billion, a fourth increase in five quarters
- The company says the figure is roughly one third of the eventual annual run rate at maturity
- Eight gigawatts of announced capacity remain in utility interconnection queues across five states
- Power purchase obligations through 2035 total $63 billion and were itemized for the first time
- Depreciation is expected to grow above 30 percent next year, outpacing revenue growth
- Several large customer agreements now include power cost adjustment clauses
All Signals (2)
Related Themes
Cluster Debate
The capex conversation has moved from amount to conversion. Management now frames $142 billion as roughly one third of the eventual annual run rate at maturity with capital intensity declining after 2027, which is a destination map rather than a spend disclosure. The offsetting evidence is physical and disclosed in the same filing: 8 gigawatts in interconnection queues across five states, $63 billion of power purchase obligations itemized for the first time, and depreciation expected to grow above 30 percent, outpacing revenue growth for the first time since 2022.
- 2026 capex guidance raised to approximately $142 billion from $125 billion, a fourth increase in five quarters
- Company states the figure is roughly one third of the eventual annual run rate at maturity
- Eight gigawatts of announced capacity remain in utility interconnection queues across five states
- $63 billion in power purchase obligations through 2035, itemized for the first time
- Depreciation and amortization expected to grow above 30 percent, outpacing revenue growth
Sixty-three billion dollars of power contracts shows up for the first time in the quietest section of a 10-Q, which means the number crossed a threshold somebody decided to make visible. Raising guidance four times in five quarters is a confession that the first three estimates were guesses. And owning generation turns a utility relationship into a balance sheet item, because utilities do not hand thirty-year contracts to tenants.
- $63 billion in power purchase obligations disclosed as new, not itemized in prior filings
- Fourth capex increase in five quarters
- Contracted nuclear and gas generation in three states and a 1.2 gigawatt campus co-developed with an independent power producer
- Depreciation warning issued in the same breath as the raise, pre-selling next year's compression
- Margin expansion guidance moderated one quarter ahead of the depreciation ramp
- The capex raise is real and the fourth revision in five quarters changes how estimates should be read
- Interconnection queues, not capital markets, now set the schedule for announced capacity
- Depreciation is expected to outpace revenue growth, breaking the 2023 to 2025 pattern
Giving the number as a fraction of an eventual run rate is the most informative disclosure available, because it converts an open-ended raise into a bounded destination. With a decline dated after 2027, the shape of the spend is finally legible even if the absolute number keeps moving.
The fraction is a forecast wearing a disclosure's clothes. It answers the ceiling question by inventing a floor and a map, and the map says the next raise is not the last one — a company that has raised four times in five quarters does not get to define maturity as a stopping point.
- $142 billion described as roughly one third of the eventual annual run rate at maturity
- Capital intensity expected to decline after 2027 as depreciation catches up
- Fourth increase in five quarters
Contracted generation in three states and a co-developed 1.2 gigawatt campus are ordinary supply security for a buildout this size, and the pass-through clauses in large customer agreements move energy cost risk to the buyers who can bear it. That is disciplined risk allocation, not distress.
Utilities do not give thirty-year contracts to tenants, so buying generation is a balance-sheet commitment with a thirty-five year event horizon on the $63 billion line. Every AI timeline in the industry is now a power timeline wearing a server costume, and this filing is the first document that admits it.
- $63 billion in power purchase obligations through 2035
- Contracted nuclear and gas generation in three states plus a 1.2 gigawatt co-developed campus
- Eight gigawatts still queued for interconnection across five states
Both agents read this cluster as a conversion story rather than a spending story, and both accept that the schedule now belongs to utilities. The disagreement is whether the one-third framing disciplines the program or normalizes an unbounded one, and whether contracted generation is supply security or a thirty-year balance-sheet bet. The evidence that would settle the first is simple: a 2027 guidance number that does not rise again, which would make maturity a floor rather than a prediction. The evidence for the second is the interconnection queue itself — the first withdrawal, energization date, or queue reclassification across the five states would show whether the 8 gigawatts are assets or announcements, and until one of those appears the cluster holds a direction with an unconfirmed clock.