Energy Transfer Is Quietly Becoming One of the Biggest Natural Gas Suppliers to AI Data Centers

Energy Transfer signed multiple gas supply deals tied to AI data-center power demand, including ~900,000 Mcf/d to Oracle across three U.S. data centers (Bloom fuel cells), plus 150,000 Mcf/d to Nexus and 900 MW of Crusoe-related supply. It also secured long-dated utility and power-plant demand, such as a 20-year Entergy agreement for at least 250,000 MMBtu/d starting Dec. 2028 and 300,000 Mcf/d to four Oklahoma gas plants through 2028. The company expects at least 17.5% EBITDA growth this year and sees scope to lift its >6% distribution by 3%-5% annually, despite permitting-driven delays.

Analysis

The market should view this as a moat-expansion story for ET, not a broad “AI beneficiary” basket trade. When a pipeline network is already sunk and new demand only requires short laterals, incremental EBITDA is disproportionately attractive: low capital, contractual volume, and limited competitive encroachment. The second-order winner is ET’s upstream and processing system, because behind-the-meter generation pulls molecules through multiple toll points before they ever hit a campus.

The real bull case is time arbitrage. Grid interconnects and utility buildouts are multi-year, so gas-as-temporary-power can persist long enough to matter for cash flow visibility, but most of the large projects won’t translate into material EBITDA until 2028-2030. That means near-term upside is mostly rerating on backlog quality, while the fundamental compounding is a 12-36 month story. The main falsifier is execution: if permitting delays become a pattern or if developers secure faster grid capacity than expected, the “bridge fuel” thesis weakens quickly.

Contrarian take: the consensus is probably overstating the benefit to power equipment and data-center operators, and understating the economics of the pipes. For ORCL and META, gas is a cost of speed, not a profit driver; for BE, the upside is more project-specific and easier to displace if turbine economics improve. ET is the cleaner expression because it monetizes volume regardless of which power technology wins. The overhang is policy/local opposition, which can delay starts by quarters but is less likely to destroy demand outright.

Key Decisions for Investors

• Buy ET on pullbacks over the next 2-6 weeks; use a 6-12 month horizon. Risk/reward is favorable because most of the AI narrative is still not in 2026-2028 consensus EBITDA, while downside is cushioned by the existing distribution.
• Pair trade: long ET / short BE for a 3-6 month relative-value expression. ET captures fee-based molecule flow; BE depends on project-level adoption and can be displaced if customers choose turbines or grid power instead. Falsify if BE starts winning repeat large-scale campus contracts.
• Avoid chasing ORCL or META on this headline alone. Their energy cost exposure is immaterial versus total capex, so the trade is not in the hyperscalers; it belongs in infrastructure names with toll-road economics.
• Set an alert on ET’s next earnings and project updates: if management quantifies AI-related backlog conversion into fee-based EBITDA, add on confirmation. If permitting delays extend beyond one-off slips, take profits.
• If the stock rerates sharply on narrative before the backlog converts, fade strength rather than add aggressively; the catalyst path is months-to-years, not days.