AI & Data Center Load Growth: On-Site Generation, Not Government Planning

By Mark Krebs and Tom Tanton -- April 17, 2024

“Wind and solar pose inherent problems; especially to the ultra-high electric energy ‘purity’ requirements of AI/data centers. Data centers and AI generally require nine-nines reliability and quality metrics such as voltage, frequency, harmonics, etc.”

Several recent articles have highlighted that artificial intelligence (AI) and data centers are increasing electricity usage, creating concern about adequate supply and its effect on local communities. These articles include:

• Amid explosive demand, America is running out of power

The nation’s 2,700 data centers sapped more than 4 percent of the country’s total electricity in 2022, according to the International Energy Agency. Its projections show that by 2026, they will consume 6 percent.

• Newmark: US data center power consumption to double by 2030

While the hyperscalers typically need 10-14kW per rack in existing data centers, this is likely to rise to 40-60kW for AI-ready racks equipped with resource-hungry GPUs. This means that overall consumption of data centers across the US is likely to reach 35GW by 2030, up from 17GW in 2022.

Fundamentally, supporting accelerating AI/ML adoption requires more power and cooling than much of the existing data center inventory can accommodate,” the report said. “Not all existing data centers lend themselves to retrofitting, catalyzing demand for new product in both existing and emerging markets.

Meanwhile, the Biden Administration, largely through the perversely titled “Inflation Reduction Act” (IRA), is providing massive and unsustainable economic incentives to move the electric generation market towards virtually exclusive reliance upon renewable energies (wind and solar in particular) plus batteries.  However, such forms of electric energy pose inherent problems; especially to the ultra-high electric energy “purity” requirements of AI/data centers. Data centers and AI generally require nine-nines reliability and quality metrics such as voltage, frequency, harmonics, etc.

A recent WSJ article cogently articulated such problems as follows:

• The Coming Electricity Crisis

Meantime, the Inflation Reduction Act’s huge renewable subsidies make it harder for fossil-fuel and nuclear plants to compete in wholesale power markets. The cost of producing power from solar and wind is roughly the same as from natural gas. But IRA tax credits can offset up to 50% of the cost of renewable operators.

Many (if not most) electric utilities take these perverse subsidies as long as they can get them blessed by their (Federal, State and/or local regulators). Electric utilities probably self-justify such actions as serving their fiduciary duties.  Typically, however, they receive massive bonuses too. In so doing, they ignore both the physics denying reality of all renewables all the time for everything and the long-term economic devastation thereof. 

Other free market advocates are also weighing-in on these issues. They include:

• Travis Fisher: The Inflation Reduction Act’s Energy Subsidies Are More Expensive Than You Think
• Robert Bryce: Wind/Solar/Alt-Energy Subsidies To Cost Federal Taxpayers $425 Billion Between Now And 2033

Fisher’s article included the following graphic cogently illustrates IRA long-term costs.

Source: https://www.woodmac.com/news/opinion/IRA-tax-credits-for-renewables/

Fisher also presented testimony on these matters before the Subcommittee on Economic Growth, Energy Policy, and Regulatory Affairs within the House Committee on Oversight and Accountability for a hearing titled The Power Struggle: Examining the Reliability and Security of America’s Electrical Grid

Cogen Solution in Waiting

The purpose of this article is to call for increased attention to be paid to market-based technology approaches for addressing these problems. That focuses upon (but is not limited to) the rapidly growing needs of AI/data centers.  An approach, previously known as cogeneration and combined heat and power (CHP) gained popularity beginning in 1978 with the passage of PURPA. Great technology, bad policy.[1]

CHP technologies under the Biden Administration have lost favor given the myopic fixation upon establishing an energy monoculture via renewables, with all electricity and the electric grids necessary to deliver this fantasy. Customer-sited supply is taking a backseat.  Recent estimates, just for the U.S., and just for storage batteries places the total costs of this transition in the range of multiples of the world GDP.  These costs could double if the unlikely objective to “electrify everything” takes hold. [For more information see Climate Change Conundrum (and comments) and  “State-by-State “Electrification” Costs Report”.]

CHP provides a case-in-point showing how electric and natural gas infrastructures can and should work together for the betterment of the economy, the environment and the public-at-large focusing on data centers burgeoning electrical demand. Certainly, there are many more energy end-use examples that could likewise benefit from diversity.  Unfortunately, at present, diversity has become a priority of the Biden Administration for everything except energy.

Doing the Math

According to the U.S. Energy Information Administration: “In 2022, total U.S. utility-scale electricity generation was about 4.24 trillion kWh.

4.24 trillion kWh = 4.24 × 10^{12 (}1,000,000,000,000) = 4,240,000,000,000 kWh

4% of 4,240,000,000,000 kWh = 169,600,000,000 kWh/year (2022 electric consumption of data centers)

169,600,000,000 kWh/year ÷ 2,700 data centers = 62,814,815 kWh/year per data center. Assuming 8,760 hours per year of base-loaded operation, this equates to 7,170.64 kW (or 7.17 MW ) of demand.

This demand puts average data centers in the technology range of aero-derivative gas combustion turbines (a.k.a. CT’s; e.g., Solar Turbines) and/or an array of VERY large reciprocating internal combustion engines (RICE; e.g., Jenbacher or Caterpillar). The latter (RICE) alternative may provide more economical redundancy by having at least one more “genset” than needed to meet peak demand in times of scheduled (maintenance) or unscheduled outages.

At present, it is common practice to install diesel-fueled generators for emergency backup.  Therefore, it is a manageable incremental cost to upgrade to continuous duty-rated natural gas engines and use them as such for CHP.

Since every kWh of data center CPU/GPU electricity consumption use produces 3412 Btu’s of heat energy, that heat must be removed to safeguard equipment and operations.  Consequently, a significant portion of the electricity used in AI and data centers is for large tonnage electric motor driven centrifugal chillers to deal with internal heat gain from electric operated computing devices (etc.). In this analysis we are taking a SWAG that 25% of the total 7.17 MW load is for cooling equipment to deal with such internal  heat gains.

Such cooling loads can be at least partially offset through engine heat recovery absorption chillers (e.g., Broad USA, Hitachi, Carrier, etc.).  Conversely, heat from engines in this size range can power steam boilers, that can be coupled to electric generators to make even more electricity.  Even still, there is ample excess heat to power absorption chillers. In any case, installing heat recovery absorption chillers instead of electric motor-drive centrifugal chillers substantially lowers the added investment for absorption alternatives.

 Simply assuming a cooling load of 7,171kW times 3,412 Btu’s per kilowatt hour then divided by 12,000 Btu per ton-hour equates to 2,039 tons of refrigeration for a typical or average data center. But that’s not just for the CPU/GPU load.  There is also some internal heat gain from lighting fans, people, etc., but that’s probably not much relative to CPU/GPU load from pushing electrons through silicone.  Let building designers thoroughly calculate cooling requirements. According to Broad USA, their heat recovery absorption chillers operate at an average of 140 % efficiency (i.e., a C.O.P. of 1.4) as long as the heat recovered meets their products’ requirements as far as flow rate, temperature, etc., etc.

At this point, analyses become site/application specific. To begin to conduct a proper feasibility study, a full 12 months of electric utility bills for an actual facility are needed and cooling load percentages factored in.  With utility billing records in hand, a “’before & after” spreadsheet can be created that looks at utility billing impacts for a given facility operations with and without on-site generation and heat recovery absorption cooling systems. 

To be meaningful, the spreadsheet should be based upon actual electric rate tariff sheets that the facility is billed on. If the analyst can replicate monthly utility bills based upon monthly demand and consumption, they are off to a good start to estimate savings potential. In addition to pecuniary savings, improvements in power quality and reliability from having generation on site should be evaluated.

One significant benefit of such an approach is the savings that accrue to other customers. As perfect power quality customers like data centers increase in number and demand, should the grid be designed and operated to meet their requirements and allocate those cost to all? Or should the power quality and reliability of the masses be met, and those with more stringent demands self-serve their nine-nines requirements.

Of course, permitting such systems has become increasingly complicated.  Therefore, it’s important to use the services of knowledgeable experienced consulting engineers to develop plans and get them through permitting processes.

Summary & Conclusions

Combined heat and power using waste heat recovery is a natural for AI/data centers deserves more consideration.  We hope that we have lit that spark. Whether that spark grows or not depends upon how permanently the Biden Administration has stifled market-based alternatives to renewables plus batteries via the “Inflation Reduction Act”  that strongly moves the market to higher and unsustainable levels of “clean energy” electrification.

The last of the additional references below indicates that there may be certain incentives for CHP through this Act, but we maintain that all such incentives are hidden taxes that are fueled by massive and ruinous deficit spending. Worse, there is no nexus between the subsidy levels and the value of CHP in specific cases. It is highly likely that some projects would see an excess level or insufficient level. We would be far better off as a society if everyone “just said no” and the Federal government got out of the way of innovation instead of trying to manage it.

Additional References

AI/data centers

• January 2024: Numark 2023 U.S. Data Center Market Overview & Market Clusters Report
• International Energy Agency: Electricity 2024 Analysis and forecast to 2026

On-site power generation

• The Line Blurs Between Small Cogen Plants and Large Combined Cycles
• HRSGs for small combined-cycle and cogen plants
• CHP AND WHP IN THE INFLATION REDUCTION ACT FREQUENTLY ASKED QUESTIONS

——————————————–

Mark Krebs, a mechanical engineer and energy policy consultant, has been involved with energy efficiency design and program evaluation for over thirty years. Mark has served as an expert witness in dozens of State energy efficiency proceedings, has been an advisor to DOE and has submitted scores of Federal energy-efficiency filings. His many MasterResource posts on natural gas vs. electricity and “Deep Decarbonization” federal policy can be found here. Mark’s first article was in Public Utilities Fortnightly, titled “It’s a War Out There: A Gas Man Questions Electric Efficiency” (December 1996). Recently retired from Spire Inc., Krebs has formed an energy policy consultancy (Gas Analytic & Advocacy Services) with other veteran energy analysts.

Tom Tanton is the Director of Science and Technology Assessment for E&E Legal. He is also president of T² & Associates, a firm providing services to the energy and technology industries. Tanton has 45 years of direct and responsible experience in energy technology and legislative interface, having been central to many of the critical legislative changes that enable technology choice. Until 2000, Tanton was the Principal Policy Advisor with the California Energy Commission (CEC) in Sacramento, California.  As General Manager at EPRI, from 2000 to 2003, Tanton was responsible for the overall management and direction of collaborative research and development programs in electric generation technologies, integrating technology, market infrastructure, and public policy. From 2003 through 2007, Tanton was Senior Fellow and Vice President of the Houston based Institute for Energy Research. He was also a Senior Fellow in Energy Studies with the Pacific Research Institute until 2010.

---

[1] The Public Utility Regulatory Policies Act (PURPA) was placed into law November 9, 1978 National Energy Act. It promoted energy conservation greater use of domestic energy and renewable energy (increase supply) and reflected the beginning of major increases in government intervention energy markets. One example  of this intervention is the mandate for utilities to buy the output, at government determined prices, from cogeneration facilities.

---