The grid is the ‘most complicated machine’ ever built; AI is stressing it out

August 28, 2024

Good morning. The Information reported that OpenAI is racing to integrate Strawberry — formerly known as Q*— into its products. Strawberry would represent a significant leap forward in reasoning capabilities; the company reportedly demoed the new tech to national security officials.

It will reportedly launch this fall.

We’ll see.

Anyway, Nvidia reports earnings today. We’ll catch you up on it tomorrow morning.

— Ian Krietzberg, Editor-in-Chief, The Deep View

In today’s newsletter:

🚒 AI for Good: Satellite-driven wildfire detection
🪨 Meta strikes deal for geothermal data center
🏛️ Regulation Roundup: Elon Musk, SB 1047 and AB 3211
🪨 The grid is the ‘most complicated machine’ ever built; AI is stressing it out

Source: NOAA

Turns out, there are a lot of different machine learning and AI-driven wildfire mitigation approaches. But unlike the approach we discussed yesterday — which relied on a network of on-the-ground cameras — NOAA recently tested a tool that links AI to satellite imagery.

The details: NOAA has a couple of satellites in geostationary orbit 22,000 miles above the Earth.

Its new tool — the Next Generation Fire System, or NGFS — leverages AI to “rapidly and autonomously” detect fires through observations made by those satellites.
The tool can detect a fire as small as one acre. After detection, the tool’s dashboard displays information regarding the fire’s location and intensity, which is then rapidly shared with the necessary responders.

Why it matters: Like the other fire-detection tools we’ve talked about, this enables firefighters to respond to a fire before it gets too big, which gives them a better shot at keeping fires small. It also allows for more advanced warnings and evacuations.

The toughest part about onboarding new employees is teaching them how to use the software they’ll need.

Guidde makes it easy.

How it works: Guidde’s genAI-powered platform enables you to quickly create high-quality how-to guides for any software or process. And it doesn’t require any prior design or video editing experience.

With Guidde, teams can quickly and easily create personalized internal (or external) training content at scale, efficiently sharing knowledge across organizations while saving tons of time for everyone involved.

Transform your organization today with Guidde.

Source: Rendering by Sage Geosystems

Facebook parent Meta has struck a deal with Sage Geosystems to supply geothermal power to support Meta’s ongoing data center expansion. The terms of the deal were not disclosed by either company.

The details: Meta aims to deliver up to 150 MW of power to support its data centers through this partnership. The first phase of the project will be online by 2027. (100 MW can power more than 16,000 average homes in the U.S.).

Sage is a four-year-old startup — that recently raised $17 million — that says it has developed technology that enables geothermal generation in a wider range of locations, since its tech harnesses geothermal pressure as well as heat.
Meta said that, to date, it has contracted 12,000 MW of renewable energy. According to Meta, its data centers are powered completely by renewables.

The context: This comes just a few weeks after President Joe Biden’s administration asked Big Tech companies to invest more in clean energy solutions to cover their surging energy demand. That surge has been fueled by their push into highly energy-intensive generative AI technologies.

It also follows a report that found that Big Tech emission numbers aren’t necessarily accurate, due to their purchase of unbundled renewable energy certificates.

The data center and AI expansion goes on, meanwhile, with Meta forecasting up to $40 billion of capital expenditure in 2024.

Join this virtual workshop to build a voice AI agent from the ground up and receive:

Hands-on coding and expert guidance
Best practices for building responsive voice interactions
Dev tools including $1K in Deepgram Voice AI API credits
Access to a community of experts in voice AI

📆 September 20 @ 9AM-12PM

Save your spot. Get 20% off with code: DV20

Physical AI startup Butlr has secured $38 million in Series B funding.
Legal AI startup Supio emerges from stealth with $25 million in Series A funding.
What the recent CrowdStrike outage misses is the aspect of human emotions related to deployments, specifically the fear of breaking production. Read more here.*
Chatsimple: Turn your anonymous website visitors into qualified leads on autopilot. Boost lead gen by 3X. Build an agent for free.*

OpenAI reportedly showed ‘Strawberry’ demo to national security officials (The Information).
GenAI startup Viggle admits on the record to training on content from YouTube (TechCrunch).
‘Being on camera is no longer sensible’: persecuted Venezuelan journalists turn to AI (The Guardian).
Mark Zuckerberg says White House ‘pressured’ Meta to ‘censor’ Covid-19 content (CNBC).
Meta to shut augmented reality studio used by third-party creators (Reuters).

If you want to get in front of an audience of 200,000+ developers, business leaders and tech enthusiasts, get in touch with us here.

Source: Created with AI by The Deep View

With a vote on SB 1047 scheduled for this week, Elon Musk — billionaire and CEO of AI startup xAI — put his support behind the legislation.

The details: “This is a tough call and will make some people upset, but, all things considered, I think California should probably pass the SB 1047 AI safety bill,” Musk said.

“For over 20 years, I have been an advocate for AI regulation, just as we regulate any product/technology that is a potential risk to the public.”
So that’s Musk and Anthropic in the ‘for’ column, and Google, Meta, Y Combinator, OpenAI and Andreessen Horowitz (plus many others) in the ‘against’ column. Both Musk and OpenAI chief Sam Altman have regularly and publicly called for AI regulation.

Context: Musk is well-known for hyping technology, whether he’s talking about Martian cities, brain implants or AI. He has talked often about the so-called ‘existential’ risk of AI technology, last year calling for a 6-month pause on advanced AI development.

The other one: At the same time as the controversy over SB 1047 continues, somehow, to grow, OpenAI has put its support behind a different California bill: AB 3211, a bill that would require companies to label AI-generated content as such.

It is also set for a vote by the full state Senate.

Source: Convergent Energy and Power

At the beginning of the year, the International Energy Agency (IEA) found that, in 2022, data centers consumed around 460 terawatt-hours (TWh) of energy, or around 3% of global electricity use. The IEA further predicted that this number will likely more than double to around 1,000 TWh by 2026 — roughly the equivalent of Japan — due, at least in part, to the energy demands of artificial intelligence.

Many of the massive investments being made around the world in AI have to do with building more data centers, even as existing data centers are consuming more energy than ever before. What this means for carbon emissions is complicated (since it depends on the cleanliness of the grid at each data center location), but today, we’re talking about the grid.

There have been plenty of reports in recent months regarding the ways in which this data center expansion is stressing our grid.

What’s going on here is that, driven by spiking data center demand, electricity demand is currently experiencing a prolonged surge. Mariko McDonagh Meier, the chief revenue officer for energy storage developer Convergent, told me that this is significant, as energy demand had been relatively flat for the past 20+ years.

This was due to increasingly energy-efficient technology, which compensated for increases in electrical usage.
“That’s where things are really shifting as data centers move toward AI, they just need so much more computing power it’s going to look very different,” she said.

The details: Grid operators, according to Meier, have the basic charge of ensuring the grid is reliable. This means managing electricity supply and demand in real-time. If a massive machine — or data center — comes online and its accompanying demand hasn’t been properly accounted for, it could “break a lot of stuff.”

The reality, according to Meier, is that “the grid is the most complicated machine that's ever been built, truly, because it is an interconnected machine.” The reason behind this interconnection has to do with built-in fail-safes; local problems can be solved by drawing on power from other places. This enhances the reliability of each interconnected grid (the U.S. and Canadian power grids, for example, are connected at 37 different points for this very reason).

The result of this interconnection — coupled with a lack of storage capabilities built into the grid — is that the grid has to be constantly balanced. This data center-driven increase then poses a significant challenge to grid operators, resulting additionally in delays for new data centers to come online.

In order to meet this surging demand, and in order to keep the grid reliable, the retirement dates of environmentally damaging coal plants are being pushed back in the U.S.

“You can’t replace the fossil plants fast enough to meet the demand,” Joe Craft, CEO of Alliance Resource Partners, one of the largest U.S. coal producers, told the FT. “In order to be a first mover on AI, we’re going to need to embrace maintaining what we have.”
Globally, 69.5 gigawatts of new coal capacity came online in 2023, compared to just 21.1 GW that were retired, according to Global Energy Monitor (this is largely due to China).

A big solution to this rather complex mess, according to Meier, involves solar plus storage tech (something that Convergent offers). Depending on the size of the solar array in question, the combination can be enough to power a data center, perhaps with a minimal connection to the grid — the result is cheaper, sustainable energy production that doesn’t impact grid reliability.