SMR Summary

The articles discuss the prospects and challenges of small modular nuclear reactors (SMRs). The key points are:

An IEEFA report finds SMRs are still too expensive, slow to build, and risky to play a significant role in transitioning from fossil fuels in the next 10-15 years. The report argues investing in SMRs diverts resources from cheaper, available renewables that can drive the transition forward. It expects 375,000 MW of new renewable capacity in the U.S. in 7 years, while SMRs are unlikely to be operational in that time.

However, the U.S. Department of Energy highlights potential benefits of SMRs, including lower upfront costs, scalability, safety enhancements, and suitability for locations that can't accommodate large reactors. SMRs could help drive economic growth and high-paying jobs if deployment reaches sufficient levels.

SMRs are gaining attention globally as they can provide reliable, low-carbon power and heat to help meet climate goals. Governments like the U.S., U.K., France and Canada are providing funding and policy support. Companies like EDF, Rolls-Royce and NuScale are developing SMR designs.

However, challenges remain, especially in regulation, licensing, and public perception. The proliferation of designs risks spreading expertise too thin. SMRs will face scrutiny and potential opposition.

Overall, SMRs offer a compelling low-carbon option, but proponents will need to work with stakeholders to overcome challenges and unlock their full potential. Near-term prospects seem limited compared to renewables.

Small modular nuclear reactors get a reality check in new report

newatlas.com

By Michael Franco

Report Summary

The report by the Institute for Energy Economics and Financial Analysis (IEEFA) analyzes the current state and prospects of small modular reactors (SMRs). The key findings are:

1. SMRs are still too expensive, slow to build, and risky to play a significant role in transitioning from fossil fuels in the next 10-15 years.

2. Investing in SMRs will divert resources away from cheaper, available renewable technologies that can drive the transition forward significantly in the next decade.

3. Experience with operating and proposed SMRs shows the reactors will continue to cost far more and take much longer to build than promised.

4. Regulators, utilities, investors and governments should recognize that renewables, not SMRs, are the near-term solution for the energy transition.

The report examines cost overruns and construction delays of SMRs built or under construction in Russia, China and Argentina. Projected costs for U.S. SMR projects by NuScale, X-energy and GE-Hitachi have also risen significantly.

The authors argue SMRs are a poor fit to complement renewable energy, as developers would have financial incentives to operate them continuously rather than flexibly. They also highlight the "Boeing problem" - the risk that a design flaw in a standardized SMR could affect an entire fleet.

In contrast, the report expects 375,000 MW of new renewable capacity to come online in the U.S. in the next 7 years, while it's unlikely any SMRs will be operational in that timeframe. The authors conclude regulators and decision-makers should embrace renewables as the available near-term energy transition solution.

---

A new report has assessed the feasibility of deploying small modular nuclear reactors to meet increasing energy demands around the world. The findings don't look so good for this particular form of energy production.

Small modular nuclear reactors (SMR) are generally defined as nuclear plants that have capacity that tops out at about 300 megawatts, enough to run about 30,000 US homes. According to the Institute for Energy Economics and Financial Analysis (IEEFA), which prepared the report, there are about 80 SMR concepts currently in various stages of development around the world.

While such reactors were once thought to be a solution to the complexity, security risks, and costs of large-scale reactors, the report asks if continuing to pursue these smaller nuclear power plants is a worthwhile endeavor in terms of meeting the demand for more and more energy around the globe.

The answer to this question is pretty much found in the report's title: "Small Modular Reactors: Still Too Expensive, Too Slow, and Too Risky."

If that's not clear enough though, the report's executive summary certainly gets to the heart of their findings.

"The rhetoric from small modular reactor (SMR) advocates is loud and persistent: This time will be different because the cost overruns and schedule delays that have plagued large reactor construction projects will not be repeated with the new designs," says the report. "But the few SMRs that have been built (or have been started) paint a different picture – one that looks startlingly similar to the past. Significant construction delays are still the norm and costs have continued to climb."

Too Expensive

The cost of SMRs is at the forefront of the report's argument against the deployment of the reactors. According to some of the data it provides, all three SMRs currently operating (plus one now being completed in Argentina) went way over budget, as this graph shows.

SMR costs have a habit of blowing past initial estimates

IEEFA

The report authors also point out that a project in Idaho called NuScale had to be scrapped because during its development between 2015 and 2023, costs soared from $9,964 per kilowatt to $21,561 per kilowatt. Additionally, the costs for three other small plants in the US have all skyrocketed dramatically from their initial cost assessments.

More out-of-control costs for SMR projects in the United States

Not only are the excessive costs of building SMRs problematic in and of themselves, says the IEEFA, but the money being poured into the projects is money that is not being spent on developing other sources of energy that are cleaner, quicker to deploy, and safer.

"It is vital that this debate consider the opportunity costs associated with the SMR push," write the authors. "The dollars invested in SMRs will not be available for use in building out a wind, solar and battery storage resource base. These carbon-free and lower-cost technologies are available today and can push the transition from fossil fuels forward significantly in the coming 10 years – years when SMRs will still be looking for licensing approval and construction funding."

Too Slow

That last bit gets to another of the report's findings: that building SMRs simply takes too much time. The Shidao Bay project in China, for example, was supposed to take four years to build, but actually took 12; the Russian Ship Borne project had an estimated completion time of three years, but took 13; and the ongoing CAREM project in Argentina was supposed to be done in four years, but it's now in its 13th year of development.

It takes a lot longer than the plans indicate to actually build SMRs

IEEFA

The report also points out that the MPower PWR project, which was one of the first planned SMRs in the US, had its plug pulled in 2017 after it was clear it wouldn't meet its 2022 deployment date – a decision that effectively wasted the $500 million that had already been spent on the effort.

"Despite this real-world experience, Westinghouse, X-Energy and NuScale, among others, continue to claim they will be able to construct their SMRs in 36 to 48 months, perhaps quickly enough to have them online by 2030," write the authors. "GE-Hitachi even claims it ultimately will be able to construct its 300MW facility in as little as 24 months.

"Admittedly, there is a not-zero chance this is possible, but it flies in the face of nuclear industry experience, both in terms of past SMR development and construction efforts and the larger universe of full-size reactors, all of which have taken significantly longer than projected to begin commercial operation."

Despite breakthroughs in SMR manufacturing, such as the welding advance that allows workers to put together an SMR reactor vessel in 24 hours instead of 12 months, the time it takes to get these facilities into the field will likely continue to be a major barrier to their adoption.

Too Risky

Both the unpredictable costs and the extraordinary building delays makes SMR development just too big of a risk, says the IEEFA. But that's not the only potential peril. Because the technology for this small-scale nuclear facility is fairly new and untested, risks could exist in terms of functionality and safety as well. For example, the authors question if the new SMRs will actually be able to output the kind of power they claim. Based on cost and development estimates going so widely afield, the sense in the report is that power output claims could also be off.

In terms of safety, the report quotes a 2023 study for the US Air Force that said: "Since SMR technology is still developing and is not deployed in the US, information is scarce concerning the various costs for [operations & maintenance], decommissioning and end-of-life dissolution, property restoration and site clean-up and waste management."

The authors also point out that because many SMRs are being built using identical technologies, if a component of that tech fails, it could easily affect reactors around the world.

For example, they bring up the fact that steam generators have needed to be replaced at more than 110 pressurized water reactors (PWRs), with half of those operating in the US, because of the denting and wall thinning of tubes made from a material called "heat-treated Alloy 600."

"We’re not arguing that new SMRs will have these same issues," they write. "We expect that the design and material decisions made for SMRs will reflect remedial measures taken at existing reactors. Our concern is broader in that a problem at one SMR might have serious repercussions at many other SMRs with the same standardized design."

Conclusion

So: too expensive, too slow, and too risky. And not at all where we should be focussing our, um – energy – these days, as the study authors make clear in their conclusion.

"At least 375,000 MW of new renewable energy generating capacity is likely to be added to the US grid in the next seven years," they say. "By contrast, IEEFA believes it is highly unlikely any SMRs will be brought online in that same time frame. The comparison couldn’t be clearer. Regulators, utilities, investors and government officials should acknowledge this and embrace the available reality: Renewables are the near-term solution."

You can read the full report in PDF format online.

Source: IEEFA

---

Benefits of Small Modular Reactors (SMRs)

energy.gov

Small modular reactors offer a lower initial capital investment, greater scalability, and siting flexibility for locations unable to accommodate more traditional larger reactors.  They also have the potential for enhanced safety and security compared to earlier designs. Deployment of advanced SMRs can help drive economic growth.

The term “modular” in the context of SMRs refers to the ability to fabricate major components of the nuclear steam supply system in a factory environment and ship to the point of use.  Even though current large nuclear power plants incorporate factory-fabricated components (or modules) into their designs, a substantial amount of field work is still required to assemble components into an operational power plant. SMRs are envisioned to require limited on-site preparation and substantially reduce the lengthy construction times that are typical of the larger units. SMRs provide simplicity of design, enhanced safety features, the economics and quality afforded by factory production, and more flexibility (financing, siting, sizing, and end-use applications) compared to larger nuclear power plants. Additional modules can be added incrementally as demand for energy increases.  

SMRs can reduce a nuclear plant owner’s capital investment due to the lower plant capital cost.  Modular components and factory fabrication can reduce construction costs and duration.

SMRs can provide power for applications where large plants are not needed or sites lack the infrastructure to support a large unit. This would include smaller electrical markets, isolated areas, smaller grids, sites with limited water and acreage, or unique industrial applications. SMRs are expected to be attractive options for the replacement or repowering of aging/retiring fossil plants, or to provide an option for complementing existing industrial processes or power plants with an energy source that does not emit greenhouse gases.  

SMRs can be coupled with other energy sources, including renewables and fossil energy, to leverage resources and produce higher efficiencies and multiple energy end-products while increasing grid stability and security.  Some advanced SMR designs can produce a higher temperature process heat for either electricity generation or industrial applications.

SMR designs have the distinct advantage of factoring in current safeguards and security requirements. Facility protection systems, including barriers that can withstand design basis aircraft crash scenarios and other specific threats, are part of the engineering process being applied to new SMR design.  SMRs also provide safety and potential nonproliferation benefits to the United States and the wider international community.  Most SMRs will be built below grade for safety and security enhancements, addressing vulnerabilities to both sabotage and natural phenomena hazard scenarios. Some SMRs will be designed to operate for extended periods without refueling. These SMRs could be fabricated and fueled in a factory, sealed and transported to sites for power generation or process heat, and then returned to the factory for defueling at the end of the life cycle. This approach could help to minimize the transportation and handling of nuclear material.  Light water-based SMRs are expected to be fueled with low enriched uranium, i.e., approximately 5 percent U-235, similar to existing large nuclear power plants. The “security by design” concepts being applied to these technologies are expected to increase SMR resistance to theft and diversion of nuclear material. Also, reactor cores for these light water SMRs can be designed to burn plutonium as a mixed oxide (MOX) fuel. Further, SMRs based on non-light water reactor coolants could be more effective at dispositioning plutonium while minimizing the wastes requiring disposal.

The case for SMR economic competitiveness is rooted in the concept that mass manufacture of modular parts and components will reduce the cost per kilowatt of electricity on par with current generating sources. There is both a domestic and international market for SMRs, and U.S. industry is well positioned to compete for these markets. DOE hopes that the development of standardized SMR designs will also result in an increased presence of U.S. companies in the global energy market. If a sufficient number of SMR units were ordered, it would provide the necessary incentive to develop the appropriate factory capacity to further grow domestic and international sales of SMR power plants. 

SMR deployment to replace retiring electricity generation assets and meet growing generating needs would result in significant growth in domestic manufacturing, tax base, and high-paying factory, construction and operating jobs. A 2010[1] study  on economic and employment impacts of SMR deployment estimated that a prototypical 100 MWe SMR costing $500 million to manufacture and install would create nearly 7,000 jobs and generate $1.3 billion in sales, $404 million in earnings (payroll), and $35 million in indirect business taxes. The report examines these impacts for multiple SMR deployment rates, i.e., low (1-2 units/year), moderate (30 units/year), high (40 units/year), and disruptive (85 units/year). The study indicates significant economic impact would be realized by developing an SMR manufacturing enterprise at even moderate deployment levels.

_{[1] Economic and Employment Impacts of Small Modular Reactors, June 2010, Energy Policy Institute of the Center for Advanced Energy Studies}

---

Why small modular reactors will shape the future of nuclear debate | White & Case LLP

whitecase.com

---

The authors would like to thank Shannon Quinn, Vice-President of Science, Technology and Commercial Oversight at Atomic Energy of Canada Limited and Seth Grae, president and CEO of Lightbridge for their contributions to this article.

SMRs are now gaining the attention of governments and power providers across the world, and even though they are a relatively nascent concept, judging by the backers lining up behind it, that won’t be the case for much longer

The vessel Akademik Lomonosov—named after an 18th-century Russian scientist, and anchored off Russia's Arctic coast—is a surprising place to find the future of nuclear energy. Yet for now, the barge is home to the world's only operating small modular reactor (SMR), a nuclear plant that produces enough electricity to power a city of about 100,000 people.

SMRs are a relatively nascent concept, but judging by the backers lining up behind it, that won't be the case for much longer. Newly elected US President Joe Biden has already signaled that they have a key role to play in the world's biggest economy's US$2 trillion investment in clean energy, while UK Prime Minister Boris Johnson has also said he'll pour money into the concept. And it's not just governments; companies including EDF—the world's biggest operator of atomic plants— and Rolls -Royce are championing the future role of the SMR.

SMRs are now gaining the attention of governments and power providers across the world because of the optionality they offer. They can provide reliable energy in the form of both electricity and heat. The heat can help lower the emissions from carbon-intensive industries such as steel and cement making, while the power offers baseload energy that can help underpin more intermittent supply from renewables such as wind and solar. The SMR offers a step change from the existing world of nuclear power.

In 1954, the Soviet Union started the world's first nuclear power plant, sending nuclear-generated electricity to a power grid, creating a whole new industry that promised to deliver a never-ending supply of reliable power. Yet that dream has faced challenges to deliver on its potential. Despite nuclear being a safe, carbon-free source of energy, the cost and scale of projects, along with major incidents at Three Mile Island, Chernobyl and Fukushima, have threatened to undermine the industry. However, the arrival of the SMR offers a solution to many of these issues.

400 nuclear power reactors are in operation around the globe

The case for nuclear

Nuclear reactors generate virtually emission-free power, which means they can play a crucial role in global efforts to lower emissions and reach the Paris climate goals. The global power sector accounts for approximately one-third of global emissions, with the burning of fossil fuels still the dominant source. The role of renewable energy generation—amid dramatic falling costs—has risen exponentially in recent years, but the inability of solar and wind to deliver reliable baseload generation has meant that many countries still rely on fossil fuels to fulfill that role.

Yet the pressure is on governments around the world to replace fossil-fuel power generation. The provision of affordable and clean energy is one of 17 United Nations Sustainable Development Goals, and at least 80 percent of the world's electricity must be low carbon by 2050—by which point the world's energy consumption is expected to have more than doubled—to have a realistic chance of keeping warming within 2°C of pre-industrial levels. Nuclear plants have a small environmental footprint and keep the air clean. They require only a small amount of fuel compared to gas or coal, and take up a fraction of the space required for wind and solar farms.

Currently, the roughly 400 nuclear power reactors that operate in some 30 countries around the globe provide about 10 percent of the world's power supply. The majority of the installed capacity is in Western countries, but a significant shift to developing nations—predominantly in Asia—is forecast over the next two decades. There is currently uncertainty about plans for replacement or life extensions in dominant producers France, Japan and the US.

80% of the world's electricity must be low carbon by 2050

What exactly is an SMR?

SMRs are broadly defined as nuclear reactors that produce less than 300 MWe—compared to the more than 1,600 MWe that modern nuclear power plants can produce.

The attraction of building smaller plants, based on a set design framework, is clear. Plants can be built quickly and to a proven standard. Additional plants can be added as more power is required, and economies of scale can be achieved as more numbers are produced. The capital outlay also becomes manageable for smaller utilities, whereas at the moment only the biggest players can shoulder the initial capital outlay burden and risks associated with some developments.

The smaller size and diversity of reactors can also mean they can be built in locations not traditionally suitable for nuclear power plants, and—importantly—close to power-intensive industries or remote communities. This allows them to cover areas of the world that rival power solutions can't reach, with the potential to replace highly inefficient and polluting power sources such as diesel generators.

SMRs can also be deployed on the sites of retiring coal plants, making the new small nuclear plants more locally acceptable by providing employment. Taking advantage of existing infrastructure, including electrical switchyards and coal plant turbines, could reduce SMR construction costs and avoid the need of adding new transmission lines from the sites.

Political support is widespread

Governments have been quick to see the potential. After years of setbacks in the UK in developing the next generation of large nuclear power plants, Prime Minister Johnson has promised funding and political support to develop the next generation of small and advanced reactors as part of his government's 10 Point Plan for a green industrial revolution.

In the US, President Biden has set goals of achieving 100 percent carbon-free electricity production by 2035 and reducing net CO₂ emissions to zero by no later than 2050. Biden's energy platform specifically cites advanced nuclear as part of "critical clean energy technologies," and his administration also plans to create an Advanced Research Projects Agency for climate that will have a specific focus on modular reactors. During April's US-led climate summit, Biden pledged to cut the country's emissions to half of 2005 levels by 2030, stressing that the world is in a "decisive decade" for tackling climate change. 

The nuclear industry has not been shy to showcase its potential role in the fight. Coinciding with Biden's two-day summit, six nuclear industry trade associations released a statement importuning that nuclear power be included in the fight against emissions, arguing that the size and urgency of the challenge demands that a "realistic, science-based approach that addresses all sectors" is required. In order to deliver the quantity of clean energy required to meet carbon neutrality, "this will require that we use every low-carbon technology at our disposal. Nuclear power must be one of those technologies," the associations said.

70  There are currently almost 70 different SMR technologies under development

Source:  International Atomic Energy Agency

The private sector is also keen

Some of the biggest industrial companies and utilities have also gotten behind the technology, with reactor designs being developed by companies from Rolls-Royce to NuScale Power and TerraPower, which has drawn investment from Bill Gates. According to the International Atomic Energy Agency, there are currently almost 70 different SMR technologies under development, a significant jump from just a couple of years ago.

EDF said earlier this year that it sees a huge global market developing for small reactors over the next decade to replace fossil-fuel generators. The company plans to complete the basic design of a 170 MW reactor by the end of 2022, and will try to convince the French government to build a pilot project by about 2030. The signs are positive. In April, French ministers signed an updated strategic plan for the French nuclear industry to account for the challenges the pandemic has brought. Almost €500 million is to be made available for the industry, with €70 million of this directed towards special projects that include the construction of a prototype SMR.

Canada released its own SMR Action Plan in December, laying out a roadmap for the development and deployment of the technology, described by the Minster of Natural Resources as "the next great opportunity," as the country seeks to phase out coal and electrify carbon-intensive industries. Shannon Quinn, Vice-President of Science, Technology and Commercial Oversight at Atomic Energy of Canada Limited, a government agency which owns the country's largest nuclear science and technology laboratory, spells it out more clearly: "SMRs are a real option in the fight against climate change. In order to meet greenhouse gas reduction targets, there is a need to bring to bear all technology options—renewables, nuclear, hydrogen and many more. We need all of them to be successful." The Baltic state of Estonia is also developing plans to place SMRs at the center of its low-carbon future. In March, Rolls-Royce and Fermi Energia—a company founded by Estonian energy and nuclear energy professionals to bring SMRs to their homeland—signed a Memorandum of Understanding to study the potential deployment of SMRs across the country.

It's not all about state-level decision-making though; there are also opportunities for smaller, more innovative companies that have been locked out of the more costly and technologically advanced traditional plants. The emergence of both private investors and smaller companies in the space offers a step change from the traditional route of government-led and funded research, development and implementation, and offers an opportunity for new methods and approaches.

One such smaller, more innovative company is Lightbridge, which has developed new metallic fuel rods that it believes are significantly more economical and safer than traditional fuel. "We have less time than the length of a typical mortgage to change the world‘s energy mix," says Seth Grae, president and CEO of Lightbridge. "Renewables most likely will generate only a small portion of the clean energy the world will need by 2050."

For Grae, there are arguably too many SMR technologies in development, which risks spreading expertise too thin: "After all, Apollo didn't happen because of a hundred private sector players pursuing different options," he adds. The world needs to get behind the most reliable and cost-effective methods. This would better enable what to scale up more quickly and provide the consistent baseload that helps offset the intermittent nature of renewables.

It’s not all about state-level decision-making when it comes to SMR technologies; there are opportunities for smaller, more innovative companies

5% of the world's total energy consumption comes from nuclear energy  

Challenges remain, namely regulation 

Regulation has always been a challenge in nuclear plant development, and that's likely to remain the case for SMRs. Issues include factors such as certification of design, construction license and operation license costs, which are hurdles that are likely here to stay. 

Licensing risk has long been a difficult and controversial issue in nuclear power, and has attracted significant attention from policy makers, the public and environmentalists. The proliferation of nuclear plants envisaged by the widespread rollout of SMRs will be unlikely to avoid this scrutiny. Given the required oversight at all levels of development, the prospect of delays, cost overruns and disputes remains, particularly in the initial rollout of new technologies.

The industry must also overcome many of the perceptions from previous generations of nuclear power; that the projects are too costly and complex, especially when compared to both traditional fossil-fuel power generation and the breakneck rollout of renewables. The issue of waste is also pertinent to the conversation. NuScale says that because SMRs contain less radioactive material and can be located below ground, their risks are lower; however, this has received criticism from some experts.

As of yet, there is no guarantee that SMR producers will not face the same obstacles that have plagued developers of traditional nuclear power. There also exists the challenge of convincing industrial users—a customer that seems well suited for the SMR offering—that it can compete with low-cost natural gas or proven renewables such as wind and solar. And the nature of SMRs mean they will have to be built close to the communities they serve, raising new challenges for public engagement.

The rollout of a new generation of nuclear power will inevitably lead to legal and commercial disputes. The intense scrutiny, from policy-makers and the public—given the safety and security angle combined with a nascent technology—will likely cause delays and conflicts. So will jockeying for position in a potentially more competitive nuclear industry marketplace. With first-of-a-kind designs, and potentially new and less experienced players in the market, comes increased risk of regulatory issues, delays, cost overruns and disputes. Understanding the nuclear regulatory framework and the nature of risks that commonly arise concerning nuclear projects will be key to navigating these issues and minimizing their impact.

Unlocking the SMR's full potential

The SMR has widespread political support. The technology's offering is compelling: carbon-free power that's reliable, safe, more affordable and can be built and deployed without the significant costs and complexity of traditional nuclear power. 

In a world where almost every investment decision will now be measured against its climate impact and whether it is compatible with the Paris climate goals, the SMR offers a solution without many of the drawbacks that have hobbled its larger predecessors. 

Yet, as with all new technologies, there are challenges to overcome and, as is always the case with nuclear power, these can be costly and protracted. The champions of the SMR will have to work with all stakeholders, from governments and investors to the wider public, to ensure its potential can be fully unlocked.

Click here to download 'Why small modular reactors will shape the future of nuclear debate' (PDF)

_{White & Case means the international legal practice comprising White & Case LLP, a New York State registered limited liability partnership, White & Case LLP, a limited liability partnership incorporated under English law and all other affiliated partnerships, companies and entities.}

_{This article is prepared for the general information of interested persons. It is not, and does not attempt to be, comprehensive in nature. Due to the general nature of its content, it should not be regarded as legal advice.}

_{© 2021 White & Case LLP}

 

USAF is rushing NGAD's drone fighter jets into service - YouTube

USAF is rushing NGAD's drone fighter jets into service

 Summary

The U.S. Air Force has selected Anduril and General Atomics to build and test production-representative Collaborative Combat Aircraft (CCA) as part of the Next Generation Air Dominance (NGAD) program. CCAs are autonomous unmanned aircraft designed to cooperate with crewed fifth- and sixth-generation fighters in a "loyal wingman" role.

Key points:

1. The Air Force plans to procure 200 NGAD fighters and 1,000 CCAs, with each CCA costing about a quarter to a third of an F-35.

2. CCAs will have a modular design, with some configured for weaponry and others for additional systems.

3. The program aims to deliver CCAs at a speed, cost, and scale to counter the pacing threat.

4. Anduril and General Atomics will build production-representative CCA vehicles and conduct test flights, while Boeing, Lockheed Martin, and Northrop Grumman will remain part of the broader industry partner vendor pool for future efforts.

5. The Air Force is on track to make a competitive production decision for the first increment of CCA in fiscal year 2026 and field a fully operational capability before the end of the decade.

6. Planning for CCA Increment 2 development is ongoing, with initial activities starting later this year, and all current and potential future industry partners competing for this follow-on effort.

7. The Air Force is exploring international partnerships as part of the CCA program to provide affordable mass at scale and drive interoperability across international partnerships.

Transcript

The US is knee-deep in work on its sixth-generation fighter, the NGAD, or Next Generation Air Dominance project. While we wait for more info on the development of said plane, this video will cover perhaps an even more important aspect of the whole sixth-generation fighter jet idea: the loyal wingman fighter jet drones, or as the Pentagon likes to call them, collaborative combat aircraft (CCA). We'll analyze the Air Force's future drones and when they might start being fielded. We'll explore the recent US Air Force decision to down-select to two companies for the first generation of its drones, with neither company being Northrop, Boeing, nor Lockheed. Air combat drones are rushing ahead, coming into service sooner than you think. Unmanned wingman jets are poised to drastically change future air warfare.

Stealth was the defining feature of the fifth-generation fighter jets. For sixth-generation fighters, the next buzzword requirement seems to be integration of loyal wingman drones to fighter jets. Right now, the US Air Force fighter jet program, the NGAD, is shaping up to be very expensive. In the words of Air Force Secretary Frank Kendall, it will cost multiple hundreds of millions of dollars per plane. Kendall also said, "The Air Force has a nominal quantity of collaborative combat aircraft for planning purposes. The planning assumption is 1,000 CCAs. This figure was derived from an assumed two CCAs for each of the 200 NGAD platforms and an additional two for each of 300 F-35s, for a total of 1,000." That was said back in March 2023, so the sixth-generation fighter is seemingly not going to be that plentiful. For comparison, only 187 F-22s were ever made. The F-35 is going to be more plentiful; the US inventory is probably around 700 planes right now, but the ultimate goal of 2,400 to be procured is still looking possible. Also, the figure of 1,000 CCAs is preliminary. Kendall himself said the actual final figure could be twice as many, but such decisions won't be made for years and may keep changing with the changes in geopolitics.

Before we go on to answer an important question: What is a CCA? A CCA is an unmanned fighter jet. It has a jet engine, wings, and it carries fuel. Some designs may have some stealth features and may carry some weapons inside to maintain stealth. It also has sensors which, like the weapons, can be swapped out between missions. Basically, several CCAs can fly alongside manned fighter jets, but not all would be the same. Some would work as sensor nodes, others would carry weapons, and some may act as radar jammers. That description is fairly vague, but it draws upon existing quotes from Pentagon officials. It's very likely the route the Air Force will follow. Also, much is being invested in autonomy, the AI brain of the drone, but that can only go so far. It seems no one is confident enough in current and near-future AI capabilities to simply cut out the NGAD manned fighter. The AI inside the drones may be good, but it seems it will mostly deal with quick reactions and operational interpretations of tactical decisions coming from the controlling planes. Basically, the AI is expected to be good enough to try to evade enemy missiles on its own, good enough to fly from point A to point B while evading collisions and minimizing its exposure to the enemy, good enough to prioritize its sensors, and maybe even to fire weapons in self-defense, but not good enough to go far out into enemy territory to search for targets on its own, to identify targets, and to decide whether to engage them, how to engage them, and so on. All that may change in the coming decades, but for now, the Air Force seems to be playing it safe with incremental development of its CCAs.

So that's why control is important, but why do it from nearby fighters and not from far away? Drones like the Reaper and such are usually controlled by a satellite, which means the input signal travels quite a bit around the Earth then back to get feedback. That takes roughly a second, which may not seem like a lot, but in combat, that can be an eternity. An additional issue is that such communication involves wide beams, which means it's fairly detectable and can be more easily jammed. A US DOD explored narrower beam satellite comms, but those too still resulted in 600-meter radius spots. Against a peer opponent, such wide spots can still be exploited. So what the Air Force needs is a way to communicate using very narrow beams, which are inherently harder to detect and thus harder to interfere with, and it needs the latency to be almost non-existent, meaning the actual control input happens fairly close to the drone. While communication signals coming from above, from satellites, can be safer from jamming than ones coming from roughly the same altitude as the CCA is at, proximity to the CCA can also help protect the signal, as signal wavelength over very short distances can be such that it's harder to pick up by the enemy. All that points to a need to fly the plane issuing commands to the CCA right at the combat zone. So not 300 miles away in some sort of AWACS-like command center, but more like 50 miles away, which means such control aircraft are bound to get shot at as well. It can't be a big AWACS-like plane; it needs to be a proper combat plane to survive. Hence the NGAD fighter being the platform which will issue commands.

Now, how often will those CCAs need input? Well, said info isn't available, but it's curious that even F-35s were mentioned as being able to control CCAs. F-35s famously do not come in two-seater versions, so a single pilot will both have to pilot their own plane, manage their own systems, and manage two or more CCAs. This April, Kendall said 2 to 5 CCAs may be acquired for every fighter. That doesn't necessarily mean so many would be flown in missions, but it does point towards future plans and needs. Kendall also said the Air Force doesn't plan to have CCAs for every single fighter, at least not initially, which also points towards future plans. Kendall did at times mention that more CCAs could be coupled with every manned fighter. Major General Scott Jobe, Director of Force Design Integration and Wargaming, said they tested former pilots in F-22 simulators, and they were comfortably managing up to six CCAs.

The Collaborative Combat Aircraft competition is quite new. Officially, five contender companies were awarded development contracts early this year, but the competition was announced in 2022, back when the CCA program was a black program shrouded in secrecy. Who knows how many years it was developed before that? Certainly, loyal wingman concepts were at times mentioned even two decades earlier. That being said, official contracting is done at a breakneck pace, signaling the Air Force's urgency. The really interesting bit is this: In January 2024, the US Air Force awarded contracts to Boeing, General Atomics, Lockheed Martin, Anduril, and Northrop Grumman. Three out of five are big names, big players who make all the US combat jets. One, General Atomics, is a known drone maker which makes the Reaper drone, for example, but it doesn't make fighter jets. And one, Anduril, is a very new company founded in 2017, with little experience. Then in April, mere months later, a down-select happened, and the US Air Force eliminated all but two companies. The companies still in the game: General Atomics and Anduril. Not a single one of the big players remains in the competition. That's really unprecedented in a way, but it also suggests where the Air Force's priorities lie, and it may help us understand what sort of CCA the US is going for.

It also has to be said that CCA is a continuous development effort for the Air Force. The contracts that General Atomics and Anduril got are for Increment 1 effort. The goal of Increment 1 is to test the whole idea, not just if the plane works, but if many planes can be built quickly. The idea behind many recent programs, not just CCAs but missile production, various other drones, is to quickly ramp up production and have hundreds of aircraft made per year if needed. The incremental development idea is also not very new. At first, even the whole NGAD plane was supposed to be done via incremental design improvements. That seems to have not worked out; a big sixth-generation plane is not a design that can be redone every several years. But with missiles and CCAs, that's still being attempted. So Increment 1 is not going to be very demanding. It's going to have some autonomy, but not too much. Air Force Acquisition Executive Andrew Hunter said autonomy is the hardest part of CCAs, and that Increment 1 will have a useful degree of autonomy, though not as much as originally thought. Hunter added that later increments will have greater autonomy, but that the focus of Increment 1 is achieving the quickest route to mass production. That is why Anduril and General Atomics beat the others.

Increment 1 is also not going to be big. It may not have a lot of range or carry lots of weapons, and it seems it will be subsonic. The winning design decision is planned for 2026, but Air Force Secretary Kendall said the Air Force might decide to build two separate designs from both contenders, if it can afford to do so. Kendall added that increments are planned to be done 2 years apart. Increment 1 is meant to go into production quickly, within a few years, and that first deliveries will happen within five or so years. Specifically, the in-service date for Increment 1 is planned for 2028. CCAs will have a common airframe and modular mission equipment, all optimized against an expected threat set. Kendall further explained that mission payloads will be swappable depending on mission requirements. He also said that fielding CCAs will not come at the expense of manned fighters, but that CCAs will enable them to be more capable. CCAs will be independent platforms carrying weapons and sensors. What that means is that the Air Force will deploy more missiles and more sensors in the same airspace at the same time, compared to today's capabilities.

While previously $40 million per CCA was the expected unit cost, lately different figures were mentioned. Kendall said CCA cost will top out at $25 to $30 million. That's the upper range. Earlier, Kendall mentioned that the lower range could be just 1/4 of the F-35's cost, meaning $20 million. Some 100 CCAs are currently planned to be built by 2029, ramping up to a thousand should happen at a greater pace, possibly by the mid-2030s. Interestingly, Air Force officials said that the Air Force started off with expecting an exquisite, high-end, stealthy platform. But then wargame simulations were done, and the service went back to the drawing board. The results of the simulations showed that in the Pacific, a large number of cheaper CCAs were more valuable than a small number of high-end drones. Indeed, at one time, the Air Force had suggested CCAs might cost half the F-35's price, $40 million, but those cost figures have since dropped, possibly after those wargames.

Two designs have been chosen, as said. One is based on General Atomics' XQ-67A. That aircraft had its first test flight only this February. General Atomics President David Alexander said that aircraft is largely the same as the design submitted for the CCA Increment 1. It shares the same structure, though there has been a slight change to the wing slope. Alexander, perhaps bravely, said the following: "This is in our wheelhouse. We are the best in the world at it." He also said his company's design should be better than the Air Force's requirement, and that General Atomics can improve upon the Air Force's schedule requirements, delivering aircraft in perhaps half the time of the requirement. Now, the final design may change a little during development, but considering the aggressive development timetable, it's likely we will indeed see drones very similar to the proposed prototypes.

The XQ-67 looks to be fairly compact. Dimensions weren't disclosed, but sizing the plane to the runway at Palmdale Airfield where it was tested, roughly 44 ft looks plausible for its wingspan, 33 to 34 ft looks like a plausible length minus the nose pitot tube. Those dimensions may seem as if they're not that much smaller than an F-16 or F-35, but weight-wise, it's a whole different class of a plane. The F-35 doesn't look a lot bigger than the F-16, but it is; it is over 40% heavier. Comparing the XQ-67 with other small aircraft and interpolating its weight, one might get an empty weight of some 2 tons, give or take, meaning six times lighter than an F-35. Things like speed, range, payload, and such are much harder to estimate, but for supersonic speeds, the plane's wings would look different. Plus, the competitor Anduril's design was officially cited to be subsonic. It's thus extremely likely General Atomics' design is subsonic as well.

Again, one can use Kratos Valkyrie figures to get some semblance of performance. So besides a Mach 0.9 speed and decent 45,000 ft ceiling, which are within the average cruise speed and ceiling of fighter jets, one thing that does stand out is range. If the General Atomics design is anything close to the 3,000 nautical miles of the Kratos Valkyrie, it will be quite a long-legged drone. That's a little over 3,400 miles, and almost surely refers to ferry range. The F-16, for example, tops out at 2,400 miles. The F-35's ferry range is unknown, but it could plausibly be around 2,000 miles. Then again, range could be closer to Boeing's Ghost Bat jet drone, which is credited with 2,300 miles. Given the size, it's plausible General Atomics' offering might have more room than its competitor, perhaps enough for three or four AMRAAM-size missiles, or for two compact air-to-ground missiles, perhaps like the SiAW.

The other competitor, Anduril, is fascinating and has an interesting approach to the drone, which seems to have piqued the Air Force's interest. Anduril is a very young company, seen as a disruptor to the established military-industrial complex companies. Its founders are of SpaceX and Palantir heritage. SpaceX already disrupted the established space launch businesses, while Palantir is a new player in the new field of big data analytics, which can be crucial in AI development. In a span of just a few years, Anduril managed to secure contracts with the Pentagon for surveillance projects, software, and drones like the small Altius and Anvil drones, as well as a very novel Roadrunner drone-like SAM missile. It seems to specialize in very low-cost drones. Anduril also has an AI-related contract with the Pentagon, the Project Maven, which deals with battlefield target recognition. Anduril seems well-connected in the Pentagon to be able to do all that so quickly. But crucially, now it entered the CCA game out of nowhere.

In September 2023, Anduril bought Blue Force Technologies, a smaller company which had made a cheap adversary training drone, Fury. Such drones help simulate opponents in real-world training flights on the cheap. So Anduril, lacking expertise and time to make such a drone from scratch, bought Fury, retained the name, and retooled the project to serve as not only an adversary training drone but also as a CCA. Anduril executive Palmer Luckey said they're bringing Anduril's work on autonomy and communications to the Fury.

Given the tight deadlines, it's improbable large changes are being done to the shape and size of the Fury. Back when it was just an adversary aircraft, it was described as a 17 ft by 20 ft plane, a Mach 0.95 plane able to do 9G instantaneous turns. It's possible that the plan is not to have Anduril's Fury carry internal weapons at all, but to have it fire its AMRAAMs from rail launchers under its wings. The bottom line is, Fury seems to be a low-cost and low-capability counterpart to General Atomics' design. Its stealth shaping is minimal, which makes sense coupled with external missile carriage. It's a visibly smaller plane, and two AMRAAMs per drone seem plausible. Another drone design similar to Fury from Blue Force did feature four AMRAAMs and a different tail section. Who knows what design Anduril will ultimately choose? Anduril's CCA seems to be aiming for an even lower bar than General Atomics, but that may be exactly what the Air Force is also exploring.

Remember when we said the Air Force is pondering giving contracts to both competitors? It implies the Air Force wants to see how both concepts would work in the real world. Ideally, it already has wargaming simulations suggesting cheap and plentiful drones are the way to go. So it may come down to money available. Anduril's CCA may be more easily produced in large numbers quickly. General Atomics' drone may be a bit less numerous, but we're still talking about cheap aircraft for now. Stealth is not part of the CCA equation much, for future increment CCAs, who knows?

The Air Force said that any company, even the ones who lost the Increment 1 competition, can re-compete for Increment 2 on their own dime, which may be problematic.

Air Force Picks Anduril And General Atomics To Build And Test Collaborative Combat Aircraft

David Cenciotti

theaviationist.com

---

The renderings of the General Atomics CCA concept based on the XQ-67 family (upper left) and Anduril's Fury (bottom right) in a single image (Credit: The Aviationist using GA-ASI and Anduril artworks)

The U.S. Air Force has awarded the two companies the funding for detailed designs, manufacture, and testing of production representative test articles under the Collaborative Combat Aircraft (CCA) program.

The Air Force made the decision to continue funding Anduril and General Atomics for the next phase of the Collaborative Combat Aircraft program, the Department of the AF announced on Apr. 24, 2024.

The two companies will now build production-representative CCA vehicles, and test-flight them, whereas Boeing, Lockheed Martin and Northrop Grumman, that also competed in the program, but were not selected, will continue to be part of the broader industry partner vendor pool consisting of more than 20 companies to compete for future efforts, including future production contracts.

Here’s a public statement released by the Department of the Air Force:

“Just over two years ago, we announced our intent, as part of our Operational Imperatives, to pursue collaborative combat aircraft. Now, following the enactment of the fiscal year 2024 budget, we’re exercising option awards to two companies to construct production representative test articles. The progress we’ve made is a testament to the invaluable collaboration with industry, whose investment alongside the Air Force has propelled this initiative forward. It’s truly encouraging to witness the rapid execution of this program,” said Secretary of the Air Force Frank Kendall.

“We executed an acquisition and funding strategy for CCA with early operator, technologist, acquirer, and industry teaming to quickly iterate requirements given our fielding timelines. Continuous competition is a cornerstone at every stage of this program. The transparency and teamwork between industry and government really accelerated how quickly we could mature the CCA program,” Kendall said.

“As we navigate the next phase of CCA development, our collaboration with both current and potential industry partners remains pivotal. Their expertise, innovation, and resources are instrumental in driving this initiative forward, ensuring its success and impact on future operations,” said Assistant Secretary of the Air Force for Acquisition, Technology, and Logistics Andrew Hunter.

The DAF is on track to make a competitive production decision for the first increment of CCA in fiscal year 2026 and field a fully operational capability before the end of the decade. The DAF’s option exercise decision does not exclude any of the vendors from competing for the future Increment 1 production contract.

The DAF is exploring international partnerships, to include potential Foreign Military Sales, as part of the CCA program.  These partnerships will help provide further affordable mass at scale while driving horizontal integration and interoperability across our international partnerships.

Planning for CCA Increment 2 development is also ongoing, with initial activities starting later this year. All current and potential future industry partners from the CCA vendor pool will compete for this follow-on effort.

The CCA program aims to deliver at least 1,000 CCAs, prioritizing cost-effective scalability. With air superiority pivotal to America’s military dominance for more than 70 years, CCA offers expanded fighter capacity (affordable mass) at reduced costs and adaptable timelines.

The CCA program

The CCA program aims to develop autonomous unmanned aircraft that will cooperate in the “loyal wingman” role with fifth- and sixth-generation combat aircraft as part of manned-unmanned teaming concepts. The U.S. Air Force wants to procure 200 NGAD fighters along with 1,000 CCA that will cooperate in the “loyal wingman” role with the sixth-generation combat aircraft. While each NGAD aircraft is expected to hundreds of millions of dollars, the CCA are designed to cost “in the order of a quarter or a third” of the current unit cost of an F-35.

The anticipated CCA’s price is between $20.6 million and $27.5 million. This is less than the price of a MQ-9 Reaper, which costs roughly $32 million. Even if the cost is “small”, the CCA are not considered expendable or attritable, but rather “systems that you can accept losses of a fraction of them and not have a big operational impact”.

The Air Force is currently refining the specifications for the CCA program, particularly regarding the range and payload, to align with the operational needs of drones supporting crewed fighters. It’s established that CCAs will have a modular design, with some configured for weaponry and others for additional systems.

The potential involvement of international partners, like Australia and Japan, in the Increment 2 is still possible as confirmed by the DAF statement, although details remain limited.

Increment 2 will introduce a new design, potentially a more advanced asset, with requirements distinct from the initial phase. While the focus for Increment 1 is swift production, subsequent increments will emphasize increased autonomy and expanded capabilities, according to Air Force Service Acquisition Executive Andrew Hunter.

The reactions of the two selected companies

“There is no time to waste on business as usual. With the CCA program, Secretary Kendall and the Air Force have embraced a fast-moving, forward-looking approach to field autonomous systems at speed and scale,” said Brian Schimpf, Anduril’s CEO and Co-Founder in a statement released by the company. “We are honored to be selected for this unprecedented opportunity, which signals a demand for continued expansion of the defense industrial base. Anduril is proud to pave the way for other non-traditional defense companies to compete and deliver on large scale programs.”

“Anduril’s work on this program is just beginning,” said Jason Levin, Senior Vice President of Anduril’s Air Dominance & Strike Division. “U.S. and allied success in the future requires CCAs to be delivered at a speed, cost, and scale to beat the pacing threat. We look forward to continuing our partnership with the U.S. Air Force to deliver this critical capability to our Airmen as quickly as possible.”

The @usairforce has selected Anduril to design, manufacture, and test production-representative Collaborative Combat Aircraft (CCA).

In this era of great power competition, there is no time to waste on business as usual. We look forward to delivering CCAs on schedule, within… pic.twitter.com/EMz8b9LkcY

— Anduril Industries (@anduriltech) April 24, 2024

“The CCA program redefines the future of aviation and will shape the USAF acquisition model to deliver affordable combat mass to the warfighter at the speed of relevancy,” said Mike Atwood, Vice President of Advanced Programs for GA-ASI.

“Throughout our 30-year history, GA-ASI has been at the forefront of rapidly advancing unmanned aircraft systems that support our warfighters,” said GA-ASI President David R. Alexander following the announcement. “The USAF is moving forward with GA-ASI due to our focused commitment to unmanned air-to-air combat operations and unmatched UAS experience, ensuring the production of the CCA aircraft at scale to deliver affordable combat mass for the warfighter.”

GA-ASI has been selected to build production representative flight test articles of the Collaborative Combat Aircraft for the @usairforce Life Cycle Management Center’s Advanced Aircraft Division.

The CCA program aims to be a force multiplier, developing a low-cost, modular,… pic.twitter.com/0JGUIIuCRI

— GA-ASI (@GenAtomics_ASI) April 24, 2024

General Atomics said that in February 2024, GA-ASI achieved a significant milestone by conducting the inaugural flight of the XQ-67A CCA prototype aircraft. “This successful flight validated the innovative “genus/species” concept introduced by the Air Force Research Laboratory (AFRL) within the framework of the Low-Cost Attritable Aircraft Platform Sharing (LCAAPS) initiative. Under this program, multiple aircraft variants are developed using a common core chassis. Subsequent to its maiden flight, the CCA prototype has undergone two additional successful test flights, marking a promising start to the production and flight testing phase. GA-ASI’s design for CCA production is derived from the XQ-67A Off-Board Sensing Station, an earlier development by GA-ASI in collaboration with the AFRL.”

The company added: “In addition to the CCA contract, GA-ASI plans to carry out a sequence of autonomy and mission system evaluations on the MQ-20 Avenger UAS and XQ-67A. This initiative aims to expedite the deployment of operational autonomy. These live flight assessments will further showcase the capability of the complete mission system to facilitate the advancement of the U.S. Air Force Autonomous Collaborative Platforms (ACP).”

David Cenciotti is a journalist based in Rome, Italy. He is the Founder and Editor of “The Aviationist”, one of the world’s most famous and read military aviation blogs. Since 1996, he has written for major worldwide magazines, including Air Forces Monthly, Combat Aircraft, and many others, covering aviation, defense, war, industry, intelligence, crime and cyberwar. He has reported from the U.S., Europe, Australia and Syria, and flown several combat planes with different air forces. He is a former 2nd Lt. of the Italian Air Force, a private pilot and a graduate in Computer Engineering. He has written five books and contributed to many more ones.

---

---

Air Force exercises two Collaborative Combat Aircraft option awards

ARLINGTON, Va. (AFNS) --

af.mil

---

The Department of the Air Force made the decision to continue funding Anduril and General Atomics for detailed designs, manufacture, and testing of production representative test articles under the Collaborative Combat Aircraft program. 

The companies not selected to build these production representative CCA vehicles, and execute the flight test program, will continue to be part of the broader industry partner vendor pool consisting of more than 20 companies to compete for future efforts, including future production contracts.

“Just over two years ago, we announced our intent, as part of our Operational Imperatives, to pursue collaborative combat aircraft. Now, following the enactment of the fiscal year 2024 budget, we're exercising option awards to two companies to construct production representative test articles. The progress we've made is a testament to the invaluable collaboration with industry, whose investment alongside the Air Force has propelled this initiative forward. It's truly encouraging to witness the rapid execution of this program,” said Secretary of the Air Force Frank Kendall.

The CCA program is part of the Next Generation Air Dominance Family of Systems, which is a DAF effort to equip the force with crewed and uncrewed platforms that can meet the pacing challenge. 

“We executed an acquisition and funding strategy for CCA with early operator, technologist, acquirer, and industry teaming to quickly iterate requirements given our fielding timelines. Continuous competition is a cornerstone at every stage of this program. The transparency and teamwork between industry and government really accelerated how quickly we could mature the CCA program,” Kendall said.

“As we navigate the next phase of CCA development, our collaboration with both current and potential industry partners remains pivotal. Their expertise, innovation, and resources are instrumental in driving this initiative forward, ensuring its success and impact on future operations,” said Assistant Secretary of the Air Force for Acquisition, Technology, and Logistics Andrew Hunter.

The DAF is on track to make a competitive production decision for the first increment of CCA in fiscal year 2026 and field a fully operational capability before the end of the decade. The DAF’s option exercise decision does not exclude any of the vendors from competing for the future Increment 1 production contract.  

The DAF is exploring international partnerships, to include potential Foreign Military Sales, as part of the CCA program.  These partnerships will help provide further affordable mass at scale while driving horizontal integration and interoperability across our international partnerships.

Planning for CCA Increment 2 development is also ongoing, with initial activities starting later this year. All current and potential future industry partners from the CCA vendor pool will compete for this follow-on effort.

The CCA program aims to deliver at least 1,000 CCAs, prioritizing cost-effective scalability. With air superiority pivotal to America's military dominance for more than 70 years, CCA offers expanded fighter capacity (affordable mass) at reduced costs and adaptable timelines.

 

Power Consumption Trends from Increased AI and Data Center Utilization

US Data Center distribution aggregated together by general classifications of commercial, co-location, and hyperscaler. As of 2022, approximately 2,701 data centers were operational in the U.S.—with the geographic concentration following the sequential distribution of: California, Texas, Virginia, and New York. Illustrating the socio-economic determinants of population density and grid infrastructure on data center location

Electric Power Research Intitute (EPRI)

Power Consumption Trends from Increased AI and Data Center Utilization

Jordan Aljbour | Technology Innovation – Strategic Research August 2023

Summary

The document discusses the increasing power consumption trends driven by the growth of AI and data center utilization. Key points include:

1. AI models, particularly those used for process automation, predictive analytics, and natural language processing, require intensive resources and have a significant carbon footprint.

2. Data centers, supporting digital services like cloud computing and AI, account for a small but growing percentage of total electricity consumption in the US and worldwide.

3. Factors driving increased power consumption include the explosion of data, increasing complexity of AI models, and the 24/7 operation of data centers.

4. Strategies to manage efficiency, usage, and environmental impact include developing efficient AI algorithms, optimizing hardware, and utilizing renewable energy sources for data centers.

5. Future scenarios range from optimistic (50% reduction in energy use with successful adoption of efficient technologies) to pessimistic (doubling of energy use without significant advances and adoption of green practices).

6. The document emphasizes the need for ongoing research, collaboration between tech companies and policymakers, and the establishment of best practices and regulations to mitigate the environmental impact of AI and data centers while leveraging their benefits.

Powering Intelligence - Analyzing Artificial Intelligence and Data Center Energy Consumption 2024

KEY TAKEAWAYS

AI and data center systems, despite their transformative impact, pose substantial environmental challenges due to high energy consumption. This growing energy demand underscores a critical trade-off between digital innovation and environmental sustainability. However, advances in software and hardware efficiency present promising strategies to mitigate this energy footprint, paving the way towards a more sustainable future.

The future leans towards greener AI and data centers, but this direction hinges on relentless research and collaboration between tech companies and policymakers to establish best practices and regulations. In doing so, we can continue to leverage the transformative benefits of AI and databases while remaining cognizant of our environmental responsibilities.

Summary

The 2024 white paper "Powering Intelligence: Analyzing Artificial Intelligence and Data Center Energy Consumption" builds upon and expands the insights from the August 2023 report "Power Consumption Trends from Increased AI and Data Center Utilization." Key additions and changes include:

1. Updated projections of U.S. data center electricity consumption to 2030, with four scenarios (low, moderate, high, and higher growth) based on the latest available data and expert insights. The 2024 report projects data centers could consume 4.6% to 9.1% of total U.S. electricity by 2030, up from an estimated 4% in 2023.
2. Detailed analysis of the potential impact of generative AI models like ChatGPT on data center electricity demand. The 2024 report highlights how these AI applications are much more energy-intensive than traditional data center workloads.
3. State-level projections of data center electricity consumption, illustrating the geographic concentration of the industry. The 2024 report notes that 15 states accounted for 80% of national data center load in 2023.
4. Discussion of strategies to manage data center efficiency, usage, and environmental impact, including energy-efficient algorithms, hardware, and cooling technologies; scalable clean energy use; and monitoring and analytics.
5. A roadmap for supporting rapid data center expansion, emphasizing improved data center operational efficiency and flexibility, increased collaboration between data center developers and electric companies, and better anticipation of future point load growth through improved forecasting and modeling.

Overall, the 2024 white paper provides a more comprehensive and forward-looking analysis of the energy implications of AI and data center growth, reflecting the rapid evolution of these technologies and their increasing importance for the electricity sector.

State Projected Load 

The table below is sorted by projected 2030 Data Center Load fraction under moderate growth of 5%. It shows that by then Virginia is in real trouble, with 46% of its power capacity used. Surprisingly, the following states are also projected to have over 20% of their power consumed: North Dakota, Nebraska, Iowa, Oregon, and Nevada.

State	Low Growth (3.7%)	Moderate Growth (5%)	High Growth (10%)	Higher Growth (15%)	2030 Projected Total State Electricity Consumption (TWh)
New York	3.40%	3.69%	5.05%	6.75%	154.2
Pennsylvania	3.78%	4.11%	5.61%	7.49%	123.2
California	4.43%	4.81%	6.54%	8.70%	271.9
Georgia	5.08%	5.51%	7.48%	9.92%	157.4
Texas	5.47%	5.94%	8.04%	10.64%	514.4
New Jersey	6.46%	7.00%	9.44%	12.44%	74.6
Illinois	6.53%	7.08%	9.54%	12.56%	147.2
Washington	6.77%	7.34%	9.88%	13.00%	98.7
Arizona	8.81%	9.53%	12.73%	16.58%	91.6
Nevada	10.28%	11.10%	14.75%	19.07%	42.9
Oregon	13.39%	14.43%	18.93%	24.14%	61.8
Iowa	13.44%	14.48%	18.99%	24.21%	46.6
Nebraska	13.75%	14.81%	19.41%	24.71%	27.2
North Dakota	18.00%	19.31%	24.89%	31.11%	21.3
Virginia	29.28%	31.10%	38.47%	46.00%	149.5

 Based on the information provided in the 2024 white paper, here is a table showing the projected data center electricity consumption as a percentage of each state's total electricity consumption in 2030 under the four growth scenarios:

State
Low Growth (3.7%)
Moderate Growth (5%)
High Growth (10%)
Higher Growth (15%)
2030 Projected Total State Electricity Consumption (TWh)
Virginia
29.28%
31.10%
38.47%
46.00%
149.48
Texas
5.47%
5.94%
8.04%
10.64%
514.40
California
4.43%
4.81%
6.54%
8.70%
271.90
Illinois
6.53%
7.08%
9.54%
12.56%
147.20
Oregon
13.39%
14.43%
18.93%
24.14%
61.80
Arizona
8.81%
9.53%
12.73%
16.58%
91.60
Iowa
13.44%
14.48%
18.99%
24.21%
46.60
Georgia
5.08%
5.51%
7.48%
9.92%
157.40
Washington
6.77%
7.34%
9.88%
13.00%
98.70
Pennsylvania
3.78%
4.11%
5.61%
7.49%
123.20
New York
3.40%
3.69%
5.05%
6.75%
154.20
New Jersey
6.46%
7.00%
9.44%
12.44%
74.60
Nebraska
13.75%
14.81%
19.41%
24.71%
27.20
North Dakota
18.00%
19.31%
24.89%
31.11%
21.30
Nevada
10.28%
11.10%
14.75%
19.07%
42.90

Please note that the 2030 projected total state electricity consumption values were not provided in the white paper. These values would be necessary to calculate the absolute projected data center electricity consumption in each state.

LLM AI System Projections

The 2024 white paper highlights the potential impact of large language models (LLMs) and generative AI systems, such as ChatGPT, on data center electricity consumption. These AI applications are much more energy-intensive than traditional data center workloads, and their rapid growth could significantly influence future electricity demand.

Key points about the projected growth of LLM AI systems:

1. Demand surge: The release of ChatGPT in November 2022 triggered a surge in interest and demand for generative AI capabilities, with major tech companies like Microsoft, Alphabet, Meta, and Bing launching their own chatbots and LLMs.

2. Energy intensity: AI queries are estimated to require about 10 times more electricity than traditional searches. For example, a ChatGPT request uses about 2.9 watt-hours (Wh), compared to 0.3 Wh for a typical Google search.

3. Potential impact on search engines: If Google were to implement LLMs in every search, it could require an additional 80 gigawatt-hours (GWh) of electricity per day or 29.2 terawatt-hours (TWh) per year. Another analysis suggests that this could necessitate around 400,000 additional servers, consuming 62.4 GWh daily or 22.8 TWh yearly.

4. Emerging capabilities: New AI applications, such as generating original music, photos, and videos based on user prompts, could require even more power than current LLMs.

5. Widespread adoption: With 5.3 billion global internet users, the widespread adoption of LLMs and generative AI tools could potentially lead to a significant increase in data center power requirements.

The white paper emphasizes that the ultimate impact of LLMs on data center energy consumption will depend on factors such as the rate of adoption, the development of new applications, and the efficiency of the underlying hardware and algorithms. Nonetheless, the rapid growth and increasing complexity of these AI systems highlight the need for continued research and collaboration to ensure sustainable growth in the data center industry.

1-bit LLMs Could Solve AI’s Energy Demands - IEEE Spectrum

Matthew Hutson

Summary

1-bit LLMs Could Solve AI’s Energy Demands

Search:

Explore by topic

AerospaceArtificial IntelligenceBiomedicalClimate TechComputingConsumer ElectronicsEnergyHistory of TechnologyRoboticsSemiconductorsTelecommunicationsTransportation

Artificial IntelligenceNews

1-bit LLMs Could Solve AI’s Energy Demands

“Imprecise” language models are smaller, speedier—and nearly as accurate

 Summary

This IEEE Spectrum article discusses the potential of 1-bit large language models (LLMs) to address the increasing energy demands and computational power required by AI systems like ChatGPT. Researchers are exploring ways to compress these models by drastically reducing the precision of the parameters that store their memories, allowing them to run on smaller devices like cellphones.

Two main approaches are mentioned:

1. Post-training quantization (PTQ): Quantizing the parameters of a full-precision network after training.
2. Quantization-aware training (QAT): Training a network from scratch to have low-precision parameters.

Recent research has shown promising results:

1. BiLLM, a PTQ method, approximates most parameters using 1 bit and a few salient weights using 2 bits, significantly reducing memory requirements while maintaining performance.
2. BitNet, a QAT method developed by Microsoft Research Asia, creates LLMs that are roughly 10 times as energy-efficient as full-precision models.
3. BitNet 1.58b, an improvement on BitNet, performs just as well as a full-precision LLaMA model with the same number of parameters and training but is faster, uses less GPU memory, and consumes less energy.
4. OneBit, a method combining PTQ and QAT, achieves a good balance between performance and memory usage.

The article suggests that the development of custom hardware optimized for 1-bit LLMs could further enhance their efficiency and speed. However, it also notes that developing new hardware is a long process, and 1-bit models and processors should grow together.

Matthew Hutson

Getty Images

large language models artificial intelligence chatgpt meta llama

Large language models, the AI systems that power chatbots like ChatGPT, are getting better and better—but they’re also getting bigger and bigger, demanding more energy and computational power. For LLMs that are cheap, fast, and environmentally friendly, they’ll need to shrink, ideally small enough to run directly on devices like cellphones. Researchers are finding ways to do just that by drastically rounding off the many high-precision numbers that store their memories to equal just 1 or -1.

LLMs, like all neural networks, are trained by altering the strengths of connections between their artificial neurons. These strengths are stored as mathematical parameters. Researchers have long compressed networks by reducing the precision of these parameters—a process called quantization—so that instead of taking up 16 bits each, they might take up 8 or 4. Now researchers are pushing the envelope to a single bit.

How to Make a 1-bit LLM

There are two general approaches. One approach, called post-training quantization (PTQ) is to quantize the parameters of a full-precision network. The other approach, quantization-aware training (QAT), is to train a network from scratch to have low-precision parameters. So far, PTQ has been more popular with researchers.

In February, a team including Haotong Qin at ETH Zurich, Xianglong Liu at Beihang University, and Wei Huang at the University of Hong Kong introduced a PTQ method called BiLLM. It approximates most parameters in a network using 1 bit, but represents a few salient weights—those most influential to performance—using 2 bits. In one test, the team binarized a version of Meta’s LLaMa LLM that has 13 billion parameters.

“One-bit LLMs open new doors for designing custom hardware and systems specifically optimized for 1-bit LLMs.” —Furu Wei, Microsoft Research Asia

To score performance, the researchers used a metric calledperplexity, which is basically a measure of how surprised the trained model was by each ensuing piece of text. For one dataset, the original model had a perplexity of around 5, and the BiLLM version scored around 15, much better than the closest binarization competitor, which scored around 37 (for perplexity, lower numbers are better). That said, the BiLLM model required about a tenth of the memory capacity as the original.

PTQ has several advantages over QAT, says Wanxiang Che, a computer scientist at Harbin Institute of Technology, in China. It doesn’t require collecting training data, it doesn’t require training a model from scratch, and the training process is more stable. QAT, on the other hand, has the potential to make models more accurate, since quantization is built into the model from the beginning.

1-bit LLMs Find Success Against Their Larger Cousins

Last year, a team led by Furu Wei and Shuming Ma, at Microsoft Research Asia, in Beijing, created BitNet, the first 1-bit QAT method for LLMs. After fiddling with the rate at which the network adjusts its parameters, in order to stabilize training, they created LLMs that performed better than those created using PTQ methods. They were still not as good as full-precision networks, but roughly 10 times as energy efficient.

In February, Wei’s team announced BitNet 1.58b, in which parameters can equal -1, 0, or 1, which means they take up roughly 1.58 bits of memory per parameter. A BitNet model with 3 billion parameters performed just as well on various language tasks as a full-precision LLaMA model with the same number of parameters and amount of training—Wei called this an “aha moment”—but it was 2.71 times as fast, used 72 percent less GPU memory, and used 94 percent less GPU energy. Further, the researchers found that as they trained larger models, efficiency advantages improved.

A BitNet model with 3 billion parameters performed just as well on various language tasks as a full-precision LLaMA model.

This year, a team led by Che, of Harbin Institute of Technology, released a preprint on another LLM binarization method, called OneBit. OneBit combines elements of both PTQ and QAT. It uses a full-precision pretrained LLM to generate data for training a quantized version. The team’s 13-billion-parameter model achieved a perplexity score of around 9 on one dataset, versus 5 for a LLaMA model with 13 billion parameters. Meanwhile, OneBit occupied only 10 percent as much memory. On customized chips, it could presumably run much faster.

Wei, of Microsoft, says quantized models have multiple advantages. They can fit on smaller chips, they require less data transfer between memory and processors, and they allow for faster processing. Current hardware can’t take full advantage of these models, though. LLMs often run on GPUs like those made by Nvidia, which represent weights using higher precision and spend most of their energy multiplying them. New hardware could natively represent each parameter as a -1 or 1 (or 0), and then simply add and subtract values and avoid multiplication. “One-bit LLMs open new doors for designing custom hardware and systems specifically optimized for 1-bit LLMs,” Wei says.

“They should grow up together,” Huang, of the University of Hong Kong, says of 1-bit models and processors. “But it’s a long way to develop new hardware.”

Large language models, the AI systems that power chatbots like ChatGPT, are getting better and better—but they’re also getting bigger and bigger, demanding more energy and computational power. For LLMs that are cheap, fast, and environmentally friendly, they’ll need to shrink, ideally small enough to run directly on devices like cellphones. Researchers are finding ways to do just that by drastically rounding off the many high-precision numbers that store their memories to equal just 1 or -1.

LLMs, like all neural networks, are trained by altering the strengths of connections between their artificial neurons. These strengths are stored as mathematical parameters. Researchers have long compressed networks by reducing the precision of these parameters—a process called quantization—so that instead of taking up 16 bits each, they might take up 8 or 4. Now researchers are pushing the envelope to a single bit.

How to Make a 1-bit LLM

There are two general approaches. One approach, called post-training quantization (PTQ) is to quantize the parameters of a full-precision network. The other approach, quantization-aware training (QAT), is to train a network from scratch to have low-precision parameters. So far, PTQ has been more popular with researchers.

In February, a team including Haotong Qin at ETH Zurich, Xianglong Liu at Beihang University, and Wei Huang at the University of Hong Kong introduced a PTQ method called BiLLM. It approximates most parameters in a network using 1 bit, but represents a few salient weights—those most influential to performance—using 2 bits. In one test, the team binarized a version of Meta’s LLaMa LLM that has 13 billion parameters.

“One-bit LLMs open new doors for designing custom hardware and systems specifically optimized for 1-bit LLMs.” —Furu Wei, Microsoft Research Asia

To score performance, the researchers used a metric calledperplexity, which is basically a measure of how surprised the trained model was by each ensuing piece of text. For one dataset, the original model had a perplexity of around 5, and the BiLLM version scored around 15, much better than the closest binarization competitor, which scored around 37 (for perplexity, lower numbers are better). That said, the BiLLM model required about a tenth of the memory capacity as the original.

PTQ has several advantages over QAT, says Wanxiang Che, a computer scientist at Harbin Institute of Technology, in China. It doesn’t require collecting training data, it doesn’t require training a model from scratch, and the training process is more stable. QAT, on the other hand, has the potential to make models more accurate, since quantization is built into the model from the beginning.

1-bit LLMs Find Success Against Their Larger Cousins

Last year, a team led by Furu Wei and Shuming Ma, at Microsoft Research Asia, in Beijing, created BitNet, the first 1-bit QAT method for LLMs. After fiddling with the rate at which the network adjusts its parameters, in order to stabilize training, they created LLMs that performed better than those created using PTQ methods. They were still not as good as full-precision networks, but roughly 10 times as energy efficient.

In February, Wei’s team announced BitNet 1.58b, in which parameters can equal -1, 0, or 1, which means they take up roughly 1.58 bits of memory per parameter. A BitNet model with 3 billion parameters performed just as well on various language tasks as a full-precision LLaMA model with the same number of parameters and amount of training—Wei called this an “aha moment”—but it was 2.71 times as fast, used 72 percent less GPU memory, and used 94 percent less GPU energy. Further, the researchers found that as they trained larger models, efficiency advantages improved.

A BitNet model with 3 billion parameters performed just as well on various language tasks as a full-precision LLaMA model.

This year, a team led by Che, of Harbin Institute of Technology, released a preprint on another LLM binarization method, called OneBit. OneBit combines elements of both PTQ and QAT. It uses a full-precision pretrained LLM to generate data for training a quantized version. The team’s 13-billion-parameter model achieved a perplexity score of around 9 on one dataset, versus 5 for a LLaMA model with 13 billion parameters. Meanwhile, OneBit occupied only 10 percent as much memory. On customized chips, it could presumably run much faster.

Wei, of Microsoft, says quantized models have multiple advantages. They can fit on smaller chips, they require less data transfer between memory and processors, and they allow for faster processing. Current hardware can’t take full advantage of these models, though. LLMs often run on GPUs like those made by Nvidia, which represent weights using higher precision and spend most of their energy multiplying them. New hardware could natively represent each parameter as a -1 or 1 (or 0), and then simply add and subtract values and avoid multiplication. “One-bit LLMs open new doors for designing custom hardware and systems specifically optimized for 1-bit LLMs,” Wei says.

“They should grow up together,” Huang, of the University of Hong Kong, says of 1-bit models and processors. “But it’s a long way to develop new hardware.”