Oracle Trucks In Gas to Beat the Grid

Forty minutes. That is how long one large trailer of compressed natural gas would keep an Oracle data center running, according to Ellie Holbrook, an energy analyst at SemiAnalysis. Her math: if Oracle powered just 100 megawatts with CNG deliveries, about 4% of its ultimate 2.45-gigawatt footprint, the trucks would have to arrive on a metronome.

Oracle is doing it anyway. Bloomberg reported that the company is using a "novel power strategy to keep several data centers on track: trucking natural gas directly to the server farms," The report, by Brody Ford and Julian Hast, was carried by Transport Topics. Trucked gas kept a data center on the outskirts of Salt Lake City moving for more than a year while Oracle waited for a pipeline to be built and tied into its machinery. It is also doing early work at a campus in Shackelford County, Texas, being built for OpenAI.

That is the story of AI infrastructure this autumn. Chips are not the scarce input. Electrons are, along with the permits to move the molecules that make them. Morgan Stanley estimated last month that US data-center developers face a 34% net power shortfall through 2028, equivalent to 32 GW, even after counting behind-the-meter generation and fuel cells, Reuters reported via Yahoo Finance.

The pipeline that missed its date

Project Jupiter in New Mexico shows how the squeeze works. Energy Transfer's Green Chile pipeline, sized for up to 400 million cubic feet per day, now targets February 1, 2027, after slipping from August 15, 2026, according to Blockspace as relayed by Yahoo Finance. The reason is land. The New Mexico State Land Office rejected the company's rights-of-way applications in March and again in July, and Energy Transfer rerouted a section onto federal Bureau of Land Management land.

Oracle has said the campus would use up to 2.45 GW of gas-fueled Bloom Energy fuel cells in an onsite microgrid. Bloom's BE shares closed at $280.50 on Friday. Oracle sent the developer a force-majeure notice seeking payment relief if the $165 billion campus misses its planned 2028 opening. Oracle is now weighing trucked gas for Jupiter too, people familiar with the project told Bloomberg.

The trucks are not cheap. Jack Weixel of East Daley Analytics said the delivered cost of trucked CNG runs well above the price of gas at a major pipeline hub once labor, specialized equipment and fuel are counted. Certarus, which supplied the Utah site, says delays in securing pipeline and grid connections "have created a growing need for rapidly deployable and scalable energy solutions." That is a vendor describing its own market, but the invoices bear it out.

Investors start to flinch

The money is getting pickier. Oracle shares dropped 5.5% to $135.69 at the close in New York, the worst single-day decline since July 16, Bloomberg reported. Its free cash flow is negative and expected to stay that way until more of its AI data centers are finished.

Far from Redwood Shores, Firmus, the Nvidia-backed Australian operator, shelved a $5 billion IPO and will raise privately instead. Reuters reported it would have been the second-largest new share sale in Australia's history, but demand was lukewarm. Firmus had planned to sell at A$11 a share, an equity valuation of $30.6 billion, nearly triple the $10.5 billion it won in a fundraising round at the start of August. It has two AI factories online, in Melbourne and Singapore. "the board therefore concluded that proceeding with the offer was not in the best interests of the company and its shareholders." the company said.

Brookings, as summarized by Semiconductor Engineering, estimates the US buildout could need $10.3T through 2032. It warned that increasingly complex financing could obscure financial risks.

Cities say pause

Local governments are adding friction. San Francisco's Board of Supervisors this week approved a 45-day moratorium on new data centers, and planning officials must report back within 25 days. The moratorium could be extended for up to two years, the San Francisco Examiner reported. Supervisor Shamann Walton, the sponsor, called it a "first step toward permanent restrictions on the facilities." Oakland adopted its own 45-day pause on Tuesday. Walton said a growing list of cities, counties and states have moved to pause or prohibit new development this year.

CBS San Francisco listed the complaints: utility use and costs, environmental effects and thin community benefits. It also reported that Gov. Gavin Newsom signed a series of bills regulating data-center development late last month.

Wall Street is split on how much this matters. Goldman Sachs expects limited near-term impact from political pushback, while Morgan Stanley sees labor, power and political challenges, Reuters reported. Morgan Stanley also said Nvidia and Broadcom look relatively insulated. Makers of memory, optics, power-management and analog parts are more exposed if customers push out deliveries or cancel orders.

Ways around the wall

Not everyone is waiting on a pipeline. The Electric Power Research Institute says US data-center electricity use could climb from about 177 terawatt-hours in 2024 to between 383 and 793 TWh by 2030, according to a Reuters explainer. EPRI's Arin Kaye said surveyed facilities reported peak-power reduction potential of 10% to 30%, depending on type. Duke University's Nicholas Institute estimated that data-center flexibility could save between $40 billion and $150 billion in capital investment over the next decade.

OpenAI has agreed to cut its grid draw by up to 1 gigawatt from a planned 3.2-gigawatt Georgia facility during periods of grid stress. Federal regulators ordered grid operators in June to consider new rules for large power users, including faster connection for those that offer flexibility. BMI analyst Alexander Kheder cautions that scaling such deals across hundreds of facilities "will demand significant capital expenditure and coordinated policy frameworks."

Another route is to build where the power is. In a Cisco-sponsored piece, The Register reported that Microsoft has linked AI data centres in Wisconsin and Georgia into one distributed supercomputer. Cisco's Itamar Gold put the timeline plainly: "We are seeing early deployments now, but it is a process and I think it will take a few years."

Until then, someone is counting trailers.