Why electrification needs a stronger grid—and more flexible data centers

Aaron Rutkoff asks Josh Parker about recent news concerning Nvidia CEO Jensen Huang and whether leadership uncertainty could affect the company’s strategy. Parker says the subject is outside his expertise and defers to others.

Asked how Huang’s management affects sustainability, Parker says it is embedded in Nvidia’s priorities. Huang has described performance per watt as the company’s measure of success, he says, making energy efficiency central to its platform. Parker also points to a new Earthshot Prize partnership intended to help organizations apply AI to sustainability challenges.

Rutkoff asks about employee morale following the news; Parker, who works remotely, says it is too recent for him to comment. Rutkoff then raises US political opposition to data centers and concern about the cost of the AI buildout, citing his understanding that around 40 data-center moratoriums passed in the midterm elections.

Parker says data centers can be community assets if they are built well, with attention to energy and water. He describes Nvidia’s S45 cooling technology as a closed loop that runs at 45°C rather than the 30–35°C of a traditional system, which he says improves energy efficiency and essentially eliminates ongoing water consumption for cooling.

Nvidia generally supplies tools rather than building data centers itself, Parker says. He points to European partner EcoDataCenter, which he says uses reinforced timber and the cooling approach, and puts waste heat from GPUs to work in district heating or drying pellets for heating.

Rutkoff turns to the COP31 presidency’s “35 by 35” pledge: raising electricity from about 20% of final energy use to 35% by 2035. He asks what would have to change to meet the target when industrial policy, the energy transition and AI are competing for constrained grid capacity.

Luke Nelson says a functioning grid is the main enabler, supporting energy security, affordability, reliability and sustainability. He sees a need for permitting reform in the US and implementation in Europe. He cites a European Commission estimate of €1.3 trillion in grid investment by 2040 and an IEA estimate of $1 trillion a year globally until 2030; legislation, workers and conditions for private capital all matter, he says.

Parker identifies supply chains as the immediate constraint. AI factories are a source of new demand, he says, but cites an IEA report in which electric vehicles, air conditioning and industry rank above data centers as drivers of net load growth. He argues that AI can also help make grids smarter.

Parker says large technology companies’ climate commitments are stimulating clean-energy purchases and investment in technologies including geothermal, nuclear, wind, solar and fusion. He estimates hyperscalers accounted for at least 75% of US corporate clean-energy purchases the previous year and around half globally. He expects supply-chain constraints on clean power to ease.

Rutkoff questions whether an electrification target should explicitly require clean generation: data-center expansion is driving renewable demand, he says, but also new gas plants, and more than half of global electricity still comes from coal and gas. Electrification could therefore grow alongside fossil-fuel power.

Nelson says renewable generation is already the cheapest new generation in much of the world, and expects economics and energy security to increase its share. Homegrown renewables, nuclear and geothermal can reduce exposure to imported energy in many places, he says. Parker agrees that the trend is toward clean power even without an explicit condition in the target.

Asked why grids are not financed and built faster, Nelson says grid projects can be unfamiliar investments. Institutional investors need suitable vehicles, and JPMorganChase is working with the Green Finance Institute in Europe on attracting private capital. Distribution projects may also need measures that improve their creditworthiness.

Nelson says cross-border grids require governments, operators and financiers to cooperate, potentially using dedicated investment vehicles. Because new lines take years, he also wants technologies that improve existing grids scaled up. Public capital could take some early risk, drawing on lessons from moving climate technologies beyond their first installations.

Nelson does not see a shortage of capital: investors are interested, he says, but enabling conditions must allow money to reach projects. From conversations at COP, he hears a strong focus on resilience to shocks and diversifying energy supply, alongside a pragmatic interest in electrification and energy security.

Rutkoff asks whether the rapid growth of AI data centers helps electrification or competes for scarce power. Parker says they can be part of the solution, both by driving investment in clean technology and by using AI to make grids smarter and better manage variable supply and demand.

Parker says data centers can provide demand response: some non-urgent computing can be paused or shifted to another location, and on-site batteries can supply power during grid stress. Utilities normally plan for extreme peaks, such as the hottest summer day; flexible demand could provide a buffer and make more use of existing infrastructure.

As an illustration, Parker describes a data center able to cut its load by 30% quickly. He says partners in the UK, including National Grid, have demonstrated data-center flexibility, and Nvidia is bringing a facility of its own online in Virginia that year to test the approach. He argues that wider use could make electricity more affordable by using infrastructure more fully.

Rutkoff asks whether increasingly powerful Nvidia chips mean an ever-growing need for electricity. Parker distinguishes total chip power from energy per task: each newer GPU can draw more power because it performs more work, while the electricity used for the same workload falls. He cites an IEA estimate of roughly tenfold annual improvement in AI energy efficiency.

Parker expects further efficiency gains from designing cooling, networking, chips and grid integration together. He points to silicon photonics, which uses light in networking, and smarter management of computing workloads as areas of work.

Asked whether continued AI growth will make the world’s 2035 climate goals easier or harder, Nelson says it is a major opportunity, but the result depends on the collective response. He points to examples of AI making electricity demand more responsive.

Parker says he expects AI to produce net emissions reductions because efficiency can save money as well as energy and materials. He cites forecasts of potentially significant reductions from the IEA and World Economic Forum, conditional on deploying the technology broadly and taking advantage of it.

Asked to imagine the 35% target achieved in 2035, Nelson says success would require turning the international goal into national and local policy that gives investors and companies clear signals. Parker says the question is whether new, increasingly clean energy infrastructure can be built quickly enough, which depends on supply chains and regulatory frameworks.