There’s a crisis brewing in Washington; a foolhardy rush to build energy-devouring data centers to feed the least-useful forms of artificial intelligence.
The narrative peddled by OpenAI and Anthropic is that cutting-edge AI is “the future,” and requires a near-infinite burden on public power grids, server farms that displace real farms and rivers sacred to Washington’s tribes cooling computers instead of salmon.
As an AI engineer and company founder based here, I’d like to point out an awkward truth: Data center expansion is entirely unnecessary for advanced AI.
What most people think of as AI is a Large Language Model, or LLM — a chatbot. To respond convincingly to inquiries, chatbots digest tens of trillions of words: more or less the entire internet.
Those “frontier models” compress the trillions of words into numbers that represent those words, ending up with a massive file that requires racks upon racks of expensive and power-hungry computer chips to answer even a single “yes-or-no” question.
It’s brute-force computing, the equivalent of driving a 7-liter Ford diesel truck that gets 8 mpg to the store, just next door. In their haste to “win” the AI race, tech giants prematurely scaled up this wasteful approach — and now they’re saddled with astronomical infrastructure debt.
They’re also pushing state and federal lawmakers to back their land and power grabs, and are hinting at socializing their losses with a bailout, more massive contracts or partial government acquisition, should there be a downturn.
We have many other forms of this idea called “AI” that actually have proven their usefulness, and are nowhere near as ravenous in their appetite for resources. Outside of Silicon Valley’s bubble, this intuition leads the rest of the world to take a vastly different path: efficient AI.
Here in Renton, at my company, Lowdown Labs, we also build specialized, efficient AI models.
We don’t do chatbots, though we do participate in research to help make them more efficient. We build AI that solves real problems:
● A model that reconstructs MRI images from magnet data using four times fewer data samples. This AI model could help researchers speed up scans for claustrophobic patients, help them cut energy use of the MRI machine and make it easier to avoid sending private medical data to an Amazon data center for processing;
● Real-time weather forecasting models that can run on solar panel microprocessors to maximize power generation;
● Image models that organize family photos without sharing private data anywhere;
● And speech transcription services that can help build affordable assistive technology, among many others.
These models can be over 100,000 times smaller than ChatGPT. Some require no more resources to train or serve than any other website on the internet would use to display a webpage.
History reminds us that constraints accelerate innovation. In the fuel crises of the 1970s and early 2000s, car manufacturers didn’t survive by selling bigger gas tanks. They invested in technologies like the turbocharger and hybrid engine to stretch every bit of energy further. If I can do the equivalent for AI, so can tech giants like OpenAI and Anthropic.
AI has reached its “fuel crisis” moment. The global scientific community, and especially China, is rapidly shifting toward these kinds of hyperefficient, open-access architectures, democratizing AI science. Heavy-handed export controls are a constraint that means international laboratories and researchers are directly incentivized by U.S. policy to innovate around compute limits rather than throw more energy at the problem.
In just a few years, “scale at all costs” will be seen as an embarrassing relic of excess — one that many prominent voices have already called out. My field moves very quickly. We don’t need to burn our climate commitments or pave over fields and ranches to realize the benefits of AI.
If Washington state approves data center expansions, sacrificing communities and expanding fossil-fueled infrastructure to accommodate Silicon Valley’s half-baked build out, then we will be left holding a big ugly, dirty bag.
We’ll be stuck with stranded industrial assets, higher utility rates, the debasement of our natural environment — and to add insult to injury, the economic impacts of Big Tech’s Big Debt. Washington doesn’t have to go down with Big Tech.
