Power & SitesUnited States

Senate investigation says that some AI data center claims are misleading

Credited to Tom's Hardware · tomshardware.com3h ago

Useful

The U.S. Senate has been conducting a year-long investigation into AI data centers, and its findings suggest that some developers have been misleading the public about the costs and benefits that these projects bring to their communities. According to Time, data center developers have routinely pointed to the number of construction jobs that projects bring to neighborhoods, but many have allegedly refused to share the number of permanent positions these projects bring to the neighborhood once they’ve reached operational status. A few companies that did give numbers point to a ratio of one permanent position for every megawatt of capacity, meaning a 100-megawatt development, which consumes around the same amount of electricity as 100,000 homes, would only hire 100 people.

The data center cooling state of play The custom AI ASIC state of play America’s AI chip rules keep changing — and the rest of the world is paying the price GTC 2026: Ian Buck press Q&A transcript — VP of Hyperscale and HPC speaks out on shelving CPX and shipping LPU decode this year Demand for data center CPUs has surged, and AI agents are responsible

“Congress must hold Big Tech accountable so these companies pay their fair share,” Senator Elizabeth Warren (D-Mass.) told the publication. Senator Chris Van Hollen (D-Md.) also said, “This report lays bare what we have long known: working Americans and local communities are footing the bill for Big Tech’s massive expansion of data centers, while these companies continue to operate without transparency.”

Aside from the lack of permanent jobs, the authorities also looked at the tax incentives these companies have been getting from local and state governments. The investigation revealed that although property-tax breaks catch critics’ attention during the application process, sales-tax exemptions on computer equipment are reportedly more lucrative, especially as these sites maintain, replace, and upgrade their hardware. The report estimated that 39% of a 1GW data center’s spending is allocated towards expensive GPUs, so if the data centers do not deliver the jobs and economic benefits they’ve promised, then the state is losing out on a lot of revenue.

The investigation also revealed that the surveyed companies — Amazon, Google, Meta, Microsoft, CoreWeave, Digital Realty, and Equinix — don’t want to cover the costs of the additional infrastructure needed to serve their data centers even though they previously promised to “pay their own way.” While the companies said that they were willing to pay for direct costs, they argued that they should not be responsible for the larger investments that could potentially benefit other customers, such as new power plants and transmission lines, even though these were primarily driven by their power demands.

There have already been efforts to legislate the ratepayer protection pledge that data center developers signed earlier this year, but it has failed because senators thought that it was “toothless.” This investigation shows that Congress is not giving up and that it’s conducting an even broader effort to regulate the infrastructure behind AI, especially with regard to its massive power consumption and other issues that residents living around them have been protesting against.

Original · Tom's Hardware

Front Method

Power & Sites · United States

Modular Data Center Company Harnesses Excess Solar Power

Credited to IEEE Spectrum · spectrum.ieee.org · 5h ago

Photograph · IEEE Spectrum

As billions of dollars are poured into building data centers, many projects face a constraint that goes beyond getting the facilities constructed: securing enough electricity to power them. San Francisco-based Rune Energy is flipping the script: instead of building data centers and waiting for grid power, the startup is putting data centers where there’s excess power the grid can’t absorb.

The company’s modular data centers, called Renewable Energy Linked Intelligent Compute (RELIC), connect directly to the direct current (DC) electricity generated by a solar array. Each RELIC module is a self-contained ecosystem with its own network equipment, cooling system, power electronics, and GPUs, all within a container about eight feet long, two feet wide, and five feet high. A module can draw up to 100 kilowatts of power and weighs about 2,000 pounds.

“It’s our estimate that there’s 50,000 gigawatt hours of energy wasted by utility-scale solar every year,” says William Layden, Rune’s co-founder and CEO. A solar farm can produce more electricity than the grid can absorb at particular times, especially when generation is high and demand is relatively low. Curtailing that generation means reducing output even though the panels would otherwise be producing electricity.

“We’re already in an energy abundant future,” says Varun Palivela, Rune’s co-founder and CTO. “But the energy abundance is stranded in time and space.” The power constraints facing large data centers, he adds, are “not a real bottleneck, but a resource allocation issue.”

The direct current advantage

In addition to curtailment, electricity is wasted due to equipment outages, conversion losses, and an issue called inverter clipping. Inverters typically convert DC electricity produced by solar panels to alternating current (AC) for delivery to the grid. When a solar array produces more DC power than its inverter can handle, the power gets clipped by the inverter.

Since Rune taps into the DC power before this conversion, they get ahead of some of the potential losses. Constance Crozier, an assistant professor at Georgia Tech’s School of Industrial Systems Engineering who has studied data center energy demand, says Rune’s main advantage won’t come from cutting out this conversion, though. “DC to AC conversion is typically quite efficient—95 to 97 percent—so I don’t expect the efficiency savings to be exceptional,” she says. “I would imagine the main advantage is reduced component cost.”

The company signs power-purchase agreements with solar-farm owners, buying electricity directly from them (and in the process, giving the solar facility income for power that would otherwise go unused). Utility-scale solar has a standardized voltage around 1500, Palivela says. While Rune’s technology doesn’t require an inverter to convert DC power to AC, it does use DC to DC converters to step down the voltage to a lower range. “You need to get the power in a conditioned state to allow the GPUs to consume it,” Layden says.

A modular approach

Layden emphasizes that Rune is a product company, not a construction company. RELIC modules are manufactured at Rune’s facility in Mountain View, California, and are designed for ultra-fast and simple deployment once they reach a solar farm. “You literally take two wires, plug it into the DC link, and you’re done,” says Layden, adding that Rune recently fulfilled a request to bring a data center online within a week. “We were able to do that,” Layden says. “We are the fastest time to compute.”

The modular approach also sidesteps some of the supply-chain bottlenecks facing conventional data centers. A traditional facility can require large transformers, switchgear, cooling systems, and other specialized electrical equipment, some of which can take months or even years to procure.

One Relic module can contain anywhere from 128 Nvidia B300 GPUs to over 2,000, and multiple modules can then be interconnected to form a large network of GPUs. “The only difference [between Rune and a conventional data center] is that it’s not in a centralized building connected to transformers and switchgear,” Palivela says. “We sell this to inference customers and those who are doing small training runs or fine-tuning training runs on state-of-the-art models.”

Rune’s approach is part of a broader push by data-center operators to go behind the meter, connecting directly to power generation to avoid relying on the grid. The strategy has gained attention as data centers struggle to secure new grid connections, with companies exploring direct connections to nuclear plants, natural-gas generators, and other sources of power.

When the sun doesn’t shine

The big question regarding Rune’s technology is the intermittency of solar power; what happens when it’s cloudy for several days in a row? “We have batteries integrated at the Relic level to make sure we always have compute 24/7,” says Layden. Sizing the batteries correctly is key, Crozier says, particularly because Rune modules don’t have a backup grid connection. “They need to size the batteries to store enough energy for nighttime and cloudy day operations, and generate enough solar to fill these batteries on ‘good solar’ days,” she says. This also means that some of the available power at any given time is directed toward battery charging rather than directly powering compute.

The company does extensive modeling to figure out the optimal data center and battery sizing, using historical data to understand how much power is likely to be available under different conditions at each site where it operates. Current sites are in California, Texas, and Massachusetts.

“Far more significant flexibility can be obtained by co-locating batteries with a data center, especially if they’re paired with a generation source,” Crozier says. “This way the data center can remove part of the load from the grid without changing its operation.”

Rune is planning to scale its operations to 100 megawatts of compute over the next 12 months.