Anthropic's $3.4 Billion Queue-Jump Puts Caterpillar's Backlog at the Center of AI

The deal: an AI lab bought a two-year-old queue ticket

On Sunday, November 8, the Financial Times reported that Anthropic, the developer of Claude, agreed to pay about $3.4 billion to take over commitments for 1.7 gigawatts of heavy-duty gas generators from OnePWR Solutions, a San Antonio, Texas-based equipment supplier. It is, per the FT, the largest known single power-equipment purchase by an AI lab.

The structure is the interesting part. Anthropic did not buy a power plant, sign a power purchase agreement, or acquire OnePWR — no employees transferred. What it bought was a purchase order: OnePWR had ordered those generators from Caterpillar roughly two years ago, before equipment prices surged, and could not secure the financing to close. Piper Sandler, which advised OnePWR, confirmed a $3.4 billion transaction selling a "power generation platform" to a frontier AI lab. So a private AI company now owns a 2024-priced queue position, with some units delivering soon and the rest within two years.

Why it matters: AI capex is no longer a chip story first. The binding constraint has moved downstream to heavy rotating equipment, and the companies that make it are collecting the rent. This piece answers one question — which listed US names actually sit where the money is flowing, and what would prove them wrong.

Figure 1 — The deal's headline numbers: $3.4 billion for 1.7 GW implies roughly $2,000 per kilowatt including installation, against Caterpillar's Q3 2026 power-generation sales of $3.42 billion, up 30%. Sources: Financial Times, Nov 8 2026; Caterpillar Q3 2026 results.

The queue behind the deal

Why pay $3.4 billion for equipment that was already ordered? Look at the order books. GE Vernova's gas equipment backlog plus slot-reservation agreements has climbed from about 83 GW at the end of 2025 to roughly 100 GW in Q1 2026, 116 GW at Q2 and 122 GW at Q3, and management has guided to at least 125 GW under contract by the end of 2026. It is now accepting slot reservations for 2031 deliveries, the CEO said in July. In Q2 2026, GE Vernova reported $24.2 billion in orders, up 88% organically, and a total backlog of $176 billion.

A 1.7 GW fleet, for scale, is a rounding error against that queue — but that is exactly the point. A new order placed today would land years out at today's prices. Anthropic's move converts waiting time into a balance-sheet asset: it now controls generation capacity it can schedule itself, bought at pre-surge pricing.

Figure 2 — GE Vernova's gas backlog plus slot reservations, end-2025 through the year-end 2026 target: 83 to at least 125 GW in nine months. Sources: GE Vernova Q3 2026 disclosures; Zacks; GE Vernova management remarks, July 2026.

Everyone is buying power — but nobody bought it this way

The direct-purchase playbook is now established. Alphabet acquired energy developer Intersect in late 2025 for a reported $4.75 billion plus assumed debt. xAI spent roughly $2 billion on mobile gas turbines, disclosed in SpaceX's IPO filing. Anthropic's deal is the third of its kind by value — but the first, as far as the reporting shows, where an AI lab took over another customer's existing equipment order rather than placing its own.

That distinction has consequences the market has not fully priced. If frontier labs start buying placed orders from distressed or under-financed owners, a secondary market in delivery slots effectively opens up — and every order on Caterpillar's or GE Vernova's books becomes an asset someone will pay a premium to inherit, one more bid under backlog quality.

Figure 3 — Disclosed values of direct power-equipment and energy-developer purchases by AI companies: Alphabet/Intersect at $4.75 billion plus debt, Anthropic/OnePWR at $3.4 billion, xAI turbines at about $2 billion. Sources: Financial Times, Nov 8 2026; SpaceX IPO filing via FT.

The economics: $2,000 per kilowatt, and a compounding compute bill

The implied price — $3.4 billion for 1.7 GW — works out to roughly $2 million per megawatt including installation and construction commitments. Expensive per kilowatt, cheap per year of waiting: Wood Mackenzie reported in April 2026 that gas turbine prices have risen about 195% since 2019, and Bernstein research reported this month put GE Vernova's gas equipment order prices more than 20% above end-2025 levels, with per-kilowatt pricing up nearly 50% versus 2025.

For Anthropic, owning generation is a hedge against a compute bill that keeps compounding. Its payments to SpaceX for leased compute started at $1.25 billion per month and are expected to reach $3.1 billion per month by year-end, per the FT. The company has disclosed nearly $900 billion in computing commitments through 2045, and Reuters reported in September that it expects to spend more than $500 billion over the next decade on AI infrastructure. Anthropic even pledged in February to bring "net-new power generation online" to match its data centers' needs. Buying them is how you keep that promise when the grid queue is five years deep.

Figure 4 — Anthropic's monthly payments to SpaceX for leased compute, from $1.25 billion at the start of the leases to an expected $3.1 billion by year-end 2026. Sources: Financial Times via VKTR, Nov 2026; Reuters, September 2026.

Bull Case

1. Backlog quality is bankable: Caterpillar's power-generation sales rose 30% year over year to $3.42 billion in Q3 2026, on a record total backlog — and labs inheriting placed orders removes the biggest conversion risk, financing.

2. GE Vernova's guidance is falsifiable and near: if gas backlog plus slots reach the guided 125+ GW by year-end, the pricing-power thesis confirms on a dated checkpoint.

3. Direct procurement spreads: each new "lab buys the equipment" deal validates the pattern; Duke Energy alone holds 7.8 GW of data-center electric service agreements and a 15.4 GW late-stage pipeline that converts by H1 2027, supported by a framework agreement now covering 26 GE Vernova turbines.

4. Pricing sticks: even if unit demand flattens, the 195% cumulative price rise since 2019 means multi-year backlogs reprice higher as legacy contracts roll off.

Bear Case

1. Slots may not convert: GE Vernova's own 10-Q risk language warns that slot reservations can lapse into orders that never materialize, exposing it to idle capacity, inventory write-downs and supplier penalties.

2. Cancellation contagion: Elon Musk claimed on X on November 6 that SoftBank is "defaulting on their turbine orders" — an unverified claim no party has confirmed, but if any large sponsor walks, the market re-rates backlog risk across the sector.

3. The financing crunch is real: the week's rout erased over $1 trillion from 20 top AI and semiconductor names, as Treasury yields climbed above 5%.

4. Owning generators has its own risks: fuel contracts, air permits and local opposition — xAI's Memphis turbines already drew an NAACP lawsuit over air quality.

5. Anthropic's own balance sheet: Blackstone has reportedly warned that Anthropic may receive a junk credit rating after its IPO, which would make debt-funded infrastructure more expensive at the worst possible moment.

Figure 5 — The cost of the queue: gas turbine prices up about 195% since 2019 per Wood Mackenzie, GE Vernova order prices up more than 20% versus end-2025 and per-kW pricing up nearly 50% versus 2025 per Bernstein. Sources: Wood Mackenzie, April 2026; Bernstein research reported November 2026.

Strategic Slam

$Caterpillar(CAT)$ : the cleanest expression. Anthropic's generators are Caterpillar machines, its power-generation segment just grew sales 30% to $3.42 billion in a single quarter, and the record backlog now carries an implicit put from AI labs willing to inherit orders. The condition to watch: power-generation revenue growth and backlog conversion through the next two quarters. If growth decelerates while the narrative peaks, the premium is wrong.

$GE Vernova Inc.(GEV)$ : the purest turbine exposure, but priced accordingly and carrying the slot-conversion risk its own filings disclose. The condition: year-end backlog of at least 125 GW, and evidence that 2026's record orders carry 2026's prices rather than 2024's.

Catalyst Calendar

Figure 6 — The dates that test the thesis: the late-September deal and its November 8 disclosure are confirmed; the Anthropic Nasdaq listing and Duke's pipeline conversions are expected, not confirmed. Sources: Financial Times, Nov 8 2026; Duke Energy Q2 2026 disclosures via Zacks.

- Late September 2026 — Anthropic takes over the 1.7 GW of Caterpillar orders from OnePWR for about $3.4 billion (Financial Times).

- November 8, 2026 — FT discloses the deal; largest known power-equipment purchase by an AI lab.

- November 2026 (expected, unconfirmed) — Anthropic's Nasdaq listing, reported as soon as this month; watch the S-1's power-supply risk disclosures.

- H1 2027 (expected) — Duke Energy's 15.4 GW late-stage data-center pipeline converts to executed agreements.

- Early 2027 (expected, dates unconfirmed) — Caterpillar and GE Vernova Q4 results: backlog and price data confirm or break the thesis.

Figure 7 — The 2×2 that decides the trade: turbine execution (rows) against AI demand (columns). Sources: GE Vernova Q3 2026 10-Q risk disclosures; Wood Mackenzie, April 2026.

Verdict

The AI trade has migrated from GPUs to the equipment that electrifies them, and this deal is the loudest confirmation yet: a frontier lab paid $3.4 billion not for chips, not for a power plant, but for a place in a queue. Caterpillar's order book just proved it is strategic infrastructure. Favor $Caterpillar(CAT)$ as the cleaner risk-reward, hold $GE Vernova Inc.(GEV)$ to its year-end backlog checkpoint, and treat any sign of slot cancellations as the thesis's kill switch.

Key Takeaways

1. Anthropic agreed to pay about $3.4 billion for 1.7 GW of gas generators, taking over Caterpillar orders that OnePWR placed two years ago and could not finance (FT, Nov 8 2026).

2. It is the largest known power-equipment purchase by an AI lab, implying roughly $2,000 per kilowatt installed.

3. GE Vernova's gas backlog plus slots rose from 83 GW at end-2025 to 122 GW at Q3, guided to at least 125 GW by year-end.

4. Gas turbine prices are up about 195% since 2019 (Wood Mackenzie), so buying a 2024-priced queue ticket is itself the trade.

5. Watch GE Vernova's year-end backlog, Caterpillar's Q4 power-generation growth, and any slot cancellation — the kill switch.

Is the next multi-billion-dollar AI purchase going to be a chip order — or someone else's turbine delivery slot?

Tickers: $Caterpillar(CAT)$ · $TICKER(GEV)$

Sources: Financial Times (Martha Muir), November 8, 2026; Reuters, November 8, 2026; CNBC and The Guardian via TechNewsReel, week to November 6; Elon Musk post on X, November 6 (unverified); Caterpillar Q3 2026 earnings release; GE Vernova Q2/Q3 2026 disclosures and Zacks coverage; Duke Energy Q2 2026 disclosures via Zacks; Wood Mackenzie, "US gas turbine market," April 2026; Bernstein research as reported November 2026; Piper Sandler transaction advisory; SpaceX IPO filing via FT; VKTR, November 2026.

Disclaimer: For information and discussion only; not investment advice. Figures come from the named sources above. Past performance does not indicate future results.

# 💰Stocks to watch today?(9 November)

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Report

Comment(2)

  • Top
  • Latest
  • IreneWells
    ·14:33
    TOP
    The 1.7GW angle matters, but CAT's real tell is whether that backlog converts to free cash flow cleanly through Q4
    Reply
    Report
    Fold Replies
    • Xaddy_Analyst: 
      Exactly the right lens — the backlog only matters if it converts. The Q4 print is the cleanest check: power-generation growth and cash conversion will show whether this is real demand or just reserved capacity.
      17:14
      Reply
      Report