The past 2 weeks have been a whirlwind of activity in high tech sustainability. We’ve moved from using nuclear power for data centers to only some nuclear power for data centers. A change of administration where support for renewables and sustainability will not be as strong as under the previous administration, and COP29 is taking place in Baku, Azerbaijan. Since my last article covered nuclear power for data centers let’s start there.
As quickly as the search for nuclear power started, it also unraveled. Two of the three arrangements will need to go back to the drawing board. The Federal Energy Regulatory Commission. (FERC) unwound a portion of Amazon’s request for power from Talen’s Susquehanna nuclear plant. FERC agreed to proceed with 300 megawatts, but at this time an additional 480 megawatts will not be allowed.
The ruling stated that Amazon didn’t meet the high burden to justify the change. It was argued by parties opposing the deal that Amazon would benefit from the grid’s transmission without paying for it. The deal is not completely dead. FERC is considering how to create an environment where transmission services are paid for in these behind-the-meter deals, and Amazon appears to be appealing the ruling.
Meta’s plans are temporarily delayed due to a rare species of bees that inhabit the site where they were exploring nuclear power generation. Meta is continuing to explore other options for a nuclear power source.
Microsoft is using a slightly different approach in its search for power. The start-up of the Three Mile Island reactor is what is called a front-of-the-meter deal. In this circumstance, Microsoft pays for the power from the nuclear reactor but doesn’t get the electricity directly from that plant. Instead, the power is delivered directly to the grid and Microsoft offsets the power it draws from the grid for its data center with the nuclear energy. This type of agreement appears to have better support from FERC to help support GRID investment, which is sorely needed as the energy demands rise in the USA
If you are keeping score, it appears the company closest to powering data centers directly with nuclear power would be Amazon if the 300-megawatt part of the deal continues to move forward.
The frantic search for clean power comes from the capital spending that is taking place for data centers. According to the Wall Street Journal, over $239 billion will be spent on capital expenditures by Amazon, Amazon, Meta, and Alphabet most of this is for data centers to support AI. New data centers are estimated to use 100 megawatts of energy each, and the total demand in 2023 is estimated to be 400 Terra watthours. The current demand is insatiable, as a result, this demand is expected to prolong the need for coal and oil-sourced power generation.
Google, Meta, AWS, and Microsoft, once upon a time, had committed to being net zero in 2030, but in their words, that was before AI reared its head. Now the companies have backed off of net zero or will need to back off of net zero due to the power demands of AI.
Eric Schmidt the former CEO of Google was quoted a few weeks ago stating: ‘We’re not going to hit the climate goals. I’d rather bet on AI solving the problem.’ While other CEOs have not made similar statements, they have gone all in on AI.
With COP29 in mid-session, while this is being written there is a great deal of angst regarding climbing CO2 levels. The commentary being published in Bloomberg Green and the WSJ climate and energy reports claim the world is going to miss emission targets, and temperature levels will rise to unbearable levels.
In researching this, the question now becomes, do we bet on AI to solve the problem? Ramp up the renewable low-carbon energy supply chain to provide the power? Look for solutions along the lines of building AI models that use less power by creating more energy-efficient chips? All of the above? Or something out of the blue?
IEEE recently reported on learning tests using new AI chips. While there is at least 2x or more improvement across the board using these chips, when will we see it from an energy consumption perspective? The Nvidia chips use ~2x the energy than the older generation so is there actually an improvement? The good news is the industry is proactively looking at how to improve energy consumption and model performance. The bad news is that in the short term, we are burning the wrong type of energy.
