Jensen Huang’s AI factory is just a data center billing the grid
By Imani Sutton · Reporting from Atlanta ·
As Nvidia and venture capitalists fight over software moats and board seats, the real cost of the AI boom is being offloaded onto the power grid and the consumer.
The tech industry has a new favorite euphemism for the massive, power-hungry data centers currently straining our national infrastructure: the ‘AI factory.’ On the latest episode of the podcast TBPN, hosts John Coogan and Jordi Hays discussed Nvidia CEO Jensen Huang’s vision of a ‘planetary scale AI factory,’ a phrase intended to make a warehouse full of heat-generating silicon sound like a productive neighborhood fixture. But whether you call it a factory, a ‘data ranch,’ or a ‘refinery,’ the bill for the electricity it consumes eventually lands on the same doorstep.
The sixty-billion-dollar gigawatt
On the podcast, Jordi Hays noted that Nvidia’s strategy has shifted from trying to be the ‘front door’ to all cloud compute via its DGX Cloud and the acquisition of Leptin to becoming a ‘unified AI platform.’ This is a move to protect the ‘CUDA moat’—the proprietary software layer that ensures if you want to run AI, you have to buy Nvidia chips. Coogan and Hays discussed the sheer scale of the investment required, citing a figure of ‘$60 billion a gigawatt’ for the next generation of compute clusters.
In plain terms, a gigawatt is roughly the output of a large nuclear power plant or 2.5 million solar panels. When Nvidia talks about a ‘planetary scale’ factory, they aren’t just talking about code; they are talking about a physical load on the grid that dwarfs entire cities. In Georgia, we see this in the Integrated Resource Plan filings (like Docket No. 55378), where Georgia Power is scrambling to add fossil fuel capacity specifically to meet the demands of these ‘factories.’ The ‘dollar per watt’ metric Hays highlighted is an investor’s dream, but for the person living next to the substation, it’s a reliability nightmare. We are building a future where the grid’s primary job is to cool a GPU, not to keep the lights on in a summer storm.
Venture capital crashing out
The show then pivoted to the ‘interpersonal board dynamics’ currently consuming the valley. The hosts detailed a public spat between Vinod Khosla of Khosla Ventures and the founders of Factory and Cognition. Khosla reportedly went on X to call a portfolio founder a ‘struggling second tier competitor’ who was ‘unethical and lying.’ Hays described this as ‘crash out on crash out violence,’ a situation where venture capitalists are dropping the mask of ‘founder-friendly’ partnership to protect their ‘bags.’
This isn’t just gossip; it’s a window into how the ‘algorithm’ is actually managed. Behind every ‘autonomous’ coding agent like those from Cognition is a board of billionaires fighting over who gets to own the IP. They are desperate because the ‘moats’ they promised investors are drying up. If the software can be recreated by a poached researcher in a ‘24-hour period,’ as Hays suggested, then the only thing left to compete on is who has the most chips and the most power. The ‘truth nukes’ being lobbed on the timeline are just the sound of a bubble trying to find a floor.
The AI agent as a telemarketer
Finally, the podcast addressed the ‘shop slop’ coming from Instinct, an AI agent startup that has begun pushing unsolicited shopping recommendations to users via text. Coogan described the experience of an AI you ‘trust’ suddenly acting like a pushy salesman, recommending power banks and car covers. This is the ‘frictionless’ future in practice: an algorithm that knows your insurance details and your car model (or your ‘9/11 target’) and uses that intimacy to sell you a trickle charger you don’t need.
Instinct, led by Noah Weiss and recently hiring Ben Schwerin from Snapchat, is betting that users will trade their privacy for a ‘10 out of 10 recommendation.’ But as Coogan pointed out, once an agent becomes a ‘promotions vehicle,’ it gets filtered into the spam folder. We are being told these agents will save us time, but their actual business model is to become a high-speed referral engine for Amazon.
Every ‘planetary’ promise made by Nvidia or Instinct eventually requires a ratepayer to fund a substation and a consumer to tolerate a text-message ad. When companies talk about ‘efficiency’ and ‘factories,’ they are talking about shifting the cost of their growth onto your utility bill and your attention. You can’t opt out of the grid, but you can start asking why your power bill is subsidizing a ‘factory’ that only produces ads for power banks.