Cipher News Cipher News

Investors need to think outside the box when it comes to addressing artificial intelligence’s energy problem.

The latest generative AI technologies, like OpenAI’s conversational tool ChatGPT, is creating a seismic shift in how most industries generate, process and disseminate information. From powering customer service solutions to streamlining content creation and enabling data interpretation, the impacts of generative AI are far-reaching.

As generative AI companies train larger and more sophisticated models, those models demand more powerful and more capable computational resources that run on more and more energy. The technology providers who rise to the occasion will be richly rewarded.

For example, OpenAI has taken a “more is more” approach to building artificial intelligence: More parameters, more data, more computation, more energy will win the day. And so far, they’ve stunned the world with the capabilities of their best-in-class models.

As demand for computational resources grows, so too does the strain on our electric grid. With the age of AI dawning, we must tackle the elephant in the room: Where on earth is all this energy going to come from?

Training a model like GPT-4 can emit as much carbon into the atmosphere as driving a gasoline car for 18 million miles.

New data centers expected to be built within the next decade could each require 1 billion watts (1 GW) of peak power to support growing AI demands. By contrast the human brain in all of its extraordinary talents consumes just 20 watts. Now rumors are swirling that Microsoft’s next generation of data centers will need 5 GW!

This cannot be the best we can do.

Our nation’s data centers already account for over 4% of the country’s electricity use and that figure is forecasted to rise by another 50% by 2026. But ubiquitous AI cannot require us to start terraforming the earth to build endless data centers that consume insatiable amounts of energy.

The road forward for AI hinges on solving these climate challenges, so it is imperative investors back the entrepreneurs hard at work building sustainable solutions.

I see investment opportunities meeting this challenge falling into three buckets:

1. Software that more efficiently trains and runs foundation models, which are trained on vast amounts of data to be applied to a broad range of use cases. For example, French AI startup Mistral’s open-source “Mixtral” models reduce computational overhead through an architecture that selectively activates only a subset of the model for complex predictions. Palo Alto-based AI startup Zyphra (which Bison Ventures has invested in) boosts efficiency with next-generation, inference-efficient hybrid architectures, reaching state of the art performance with only a fraction of the training data required for other models.
2. Hardware with architecture specifically tailored to AI algorithms. The approaches companies are exploring include placing processing units closer to memory, integrating them into memory modules directly and improving data transfer speeds between the two. Examples of companies making strides in hardware include d-Matrix, which delivers rapid AI inference (the process a trained machine learning model uses to draw conclusions from brand-new data) at scale, and Cerebras, which trains models at record speed.
3. Dependable energy sources with low greenhouse gas emissions and competitive pricing compared to fossil fuels, such as solar power integrated with long-duration energy storage or next generation nuclear energy.

Most climate investors focus primarily on number three while essentially ignoring software and hardware. Understanding clean energy is more straightforward than evaluating advances in algorithms and computational architecture. Technically savvy climate investors can wax poetic about the pros and cons of battery chemistry, but their eyes glaze over when AI researchers describe advances in their field.

This is a huge blind spot for the climate investing community — not just because of the substantial climate repercussions but also because of the financial opportunity. If you want to drive adoption and profit at scale, then it’s time we get to work unlocking this scalable need.

These non-obvious climate technologies are precisely the ones that have potential for massive long-term impact. By redefining our priorities, we can fuel the age of AI without compromising our planet — and make money doing it.