xAI came to Memphis claiming one thing, then in a sneak move, increased their project size by 10 times. This will essentially require a new power plant, and puts billionaires’ benefits above local Memphians.

In June of this year, Elon Musk claimed his artificial intelligence xAI “Colossus” facility would have the world’s largest supercomputer capacity, and he began building it right here in Memphis, Tennessee. The problem that no one seems to grasp now is that Musk has recently expanded the xAI facility plans by 10x, and the power of an AI facility of this size will likely require amounts to 1/3 of the current peak demand of the City of Memphis.

This power load would create extraordinary strain on Memphis Light, Gas, and Water (MLGW), an already strained system in a city that has routinely experienced one of the highest energy burdens of any metro area in the United States. Another way to visualize this massive energy demand is that it is equivalent to roughly a full nuclear or fossil gas power plant’s worth of energy.

The Bait…

During initial plans for, as well as the announced expansion of, the “Colossus” xAI supercomputer facility, Musk’s company involved no public meetings, nor communicated directly with the media. City officials signed nondisclosure agreements in order to engage xAI in negotiations to bring the plant to Memphis. Original plans announced in June, 2024, called for a capacity of about 100,000 Graphics Processing Units (GPUs – the preferred chips for AI technology) and would consume a maximum of 150 megawatts (MW). At this size, the proposed unit would require its own substation from Memphis Gas, Light and Water (MLGW), as well as Tennessee Valley Authority (TVA) Board approval since the project is over 100 MW. The TVA Board approved the 150 MW project on November 7 and MLGW agreed to take on xAI as a customer, instead of having TVA serve them directly, which is a more common option for a project this size.

In Musk’s rush to reach his perception of grandeur, he brought in portable fossil gas turbines. Under a waiver from the Tennessee Department of Environment and Conservation (TDEC), Musk can run these, largely uncontrolled and unchecked for pollution and environmental impacts, while he waits for MLGW and TVA  to build the substation required, which could take months or up to a year. There remain unanswered questions on water usage for the necessary cooling load.

And that’s just the technology; the physical impact of the facility has also caused concern. It’s being built in South Memphis, known for both historically Black neighborhoods and poor air quality. It will place significant demands on the city’s electricity and water systems, and the impacts on both could harm residents.

…And The Switch

On December 4, at the Greater Memphis Chamber Annual Chairman’s Luncheon, a representative for xAI revealed plans for an expansion of its aptly named “Colossus” supercomputer facility. According to the Chamber’s press release, the expansion is already underway, and will incorporate a minimum of one million GPUs, marking the largest capital investment in the region’s history. Additionally, Fortune 500 tech giants Nvidia, Dell, and Supermicro Computer (SMC) will be establishing operations in Memphis.

What wasn’t mentioned at that meeting was the new level of power that will be required for this size of facility: presumably, this jump in capacity would increase power demand by roughly 10x as well (from 150 MW to over 1,000 MWs). This previously unannounced capacity makes this a “hyperscale” project.

Unless xAI can say definitively otherwise, we are operating under the logical assumption that by scaling the project 10x, the required energy will scale proportionately to over 1 gigawatt (GW, or 1,000 MW). 

Without evidence to the contrary, this raises critical concerns

Even more concerning is the fact that we are far from convinced that xAI didn’t know when they first came to Memphis that they would expand the plans 10x. This is a classic bait and switch: xAI proposed one thing, and grew support from stakeholders like the Greater Memphis Chamber, the Memphis Shelby County Economic Development Growth Engine (EDGE), TVA, and governing authorities. Musk and crew moved forward on the path to operating the facility, and only then announced a change in magnitude that creates a fundamentally different situation for public health in Memphis, economic stability for ratepayers, and environmental security.

Essentially, TVA will now have to contemplate building a power plant just to serve this one facility  – –  the unanswered question is who pays for the new generation?

The Math is Simple – and Alarming

100,000 – the number of GPUs previously proposed by xAI

150 MW – the electric energy that would be needed to operate this facility

100,000 x 10 = 1,000,000 – the GPU capacity of the expanded proposal

150 x 10 = 1,000 to 1,500 MW (1.5 GW) – the electric energy presumably needed to operate a facility at this scale. This depends on GPUs, configuration, cooling etc.

Again — this is 1/3 of the current peak demand of the city of Memphis, further stressing an already stressed MLGW system in a city that has routinely experienced the highest energy burden of any metro area in the United States.

Short of any independent facts stated otherwise, this raises a valid concern. 

Another way to visualize this massive energy demand is that it is equivalent to roughly a full nuclear or fossil gas power plant’s worth of energy, (a typical combined cycle gas plant is around 1 GW, between 800-1100 MW.) In turn, this would require another (or multiple more) substation approval from TVA. Significant cooling water usage issues remain.

Our original hope with the original 150 MW level facility was that Musk would use his engineering expertise to bring Tesla magapacks with solar and storage capability to make this facility a model of clean, renewable energy. It would be good to hear if there is an expansion to the facility that there’s a scalable commitment to renewable energy and storage. Again, more information is needed.

And while it may be possible to gain some level of efficiency, to power 1 million GPUs, even if efficiencies improve the system by 25% or so, will still approach at 1 GW – or a power plant’s worth – of electricity. Chip manufacturer Nvidia has two future chips in development that could have some efficiency gains over the ‘Hopper’ chip currently being used, but those gains are unlikely to be significant enough to offset the increase, and the next “Blackwell” chip is behind schedule. Furthermore, if efficiency gains or software modifications are going to adjust this formula, then xAI is not being forthcoming with that information. Why not?

So our question to xAI is: How much power will you use? 

Without more transparency and some clear information from xAI to the contrary, we are convinced that the risks are not worth the rewards in the current scenario. Furthermore, who will be accountable for the cost of building this infrastructure, especially if xAI packs up and leaves or closes the project after the power plant is built?

TVA Should Proceed with Caution

TVA needs to stop and decide how to handle the risk of this situation. Without answers to fundamental questions, it is absolutely wreckless to rush headlong into committing to build AI capacity,

This Issue Extends Beyond Memphis

Speculation like this is growing around the country, and the load growth numbers being touted are unprecedented. According to recent analysis by Grid Strategies, ”In just two years, the forecast of cumulative electricity growth over the next five years appears to have increased by a factor of five, from 2.8% to 15.8%,” driven by several factors including AI and data centers.

Right here in our region, cryptocurrency and AI speculation are driving load growth speculation to unheard of levels. For instance, last month Georgia Power released its expected load growth for the next decade, and it has tripled from 12,000 MW to 36,000 MW, primarily coming directly from large load (data centers), most of which is power load growth for AI and bitcoin speculation.

Do we really want to have a speculative bubble for nonessential services that puts ratepayers and our environment at risk? TVA needs to use this opportunity to set a precedent for smart growth instead of rushing headlong into a messy situation.

AI may indeed have benefits for society, but we should first do no harm… We need to slow down, have transparent and accurate information, and make smart choices for a clean, safe, and healthy future.