The next AI bottleneck isn't the chip factory - it's the power plant.
Naveen Rao, speaking on This Week in AI, tracks a seismic shift in the economics of compute. For the previous generation of Nvidia A100 chips, energy comprised about 10% of total data center ownership costs. For the latest hardware, like the Blackwell series, that figure has surged to nearly 40%. Rao projects energy will constitute over half of the total bill within the next three to four years.
"We are hitting an energy wall that manufacturing cannot solve alone. Simply building more chips won't work if the grid can't light them."
- Naveen Rao, This Week in AI
This trajectory threatens to bankrupt unoptimized models, especially as Rao estimates 20-30% of current AI compute is wasted on 'token maxing' - burning compute to climb leaderboards rather than solve problems. The industry is now forced to pursue new, more efficient architectures that merge memory and computing, mimicking the brain's energy frugality.
The soaring power demand is colliding with growing public skepticism. Rao blames companies like Anthropic for this, arguing their 'doomer' narratives have painted AI as an existential threat to the public. This sentiment manifests as local protests against new data centers over fears about water usage or health risks. Rao contrasts this with China, where public sentiment views AI as a competitive superpower worthy of celebration.
"If the public views AI as a playground for coastal elites that offers nothing to the heartland, they will vote to shut it down."
- Naveen Rao, This Week in AI
Part of the backlash stems from a perceived lack of shared benefit. Rao suggests companies building data centers should voluntarily invest in the local community - funding public transit or schools - to build tangible goodwill.
Meanwhile, the other major AI narrative of the moment - that it's causing massive layoffs - is being labeled a false flag. Co-host Alex Finn argues the current wave of job cuts traces back to reckless over-hiring during the 2020 zero-interest-rate period. AI, he says, is a convenient scapegoat for managerial mistakes. In reality, Rao notes his startup sees increased demand for engineers because AI tools make them exponentially more productive.
The fundamental challenge is no longer just building smarter models, but powering them in a world that may not want them.
