1. Recap of some MAG7 results.

Tesla missed estimates. Total revenue rose just 2% YoY, but automotive revenues sank 8% YoY to $19.8 billion, with $692 million coming from regulatory credits.

Musk promises to scale Optimus production 500% to 100K units/month at $20K per unit, with deliveries expected by 2H 2025.

Its robotaxis are planned for June in 1 city while Waymo is expanding its service to another 10 cities in the US.

2. Meta hit a new all-time high after its earnings per share and revenue both topped estimates.

Fourth-quarter revenues rose 21% YoY, while net income rose 49%. Daily active users rose to 3.35 billion.

Meta’s AI chatbot topped 700 million monthly active users. It expects that number to scale to more than a billion this year.

3. Microsoft moved lower after the bell.

Earnings per share and revenue topped estimates, but its Intelligent Cloud segment (which contains the Azure cloud), came in below estimates despite 19% YoY growth.

CEO Satya Nadella noted that Microsoft's artificial intelligence (AI) annualized revenue run rate now exceeds $13 billion.

4. The “Deepseek” selloff wasn’t just limited to technology stocks like Nvidia Corp.

Shares in energy companies key to the AI buildout were hammered too, leaving some on Wall Street wondering if their valuations are too high.

The magnitude of the drop was so great that it “implies any uplift provided by the promise of AI data center load growth has effectively been washed out,” Evercore ISI analyst Durgesh Chopra wrote in a note to clients.

He said that Vistra and NRG Energy were discounted to the point that their stock prices now imply earnings that are “on par with those of November 2023, a period prior to any data center sentiment accretion.”

Some on Wall Street see the dip in power producers representing a buying opportunity. Power is “the one area that will remain well needed,” regardless of competition in the AI race, said Peter Boockvar, chief investment officer at Bleakly Financial Group LLC.

“By owning energy stocks, one doesn’t care which model/tech company is going to win.”

5. In terms of energy consumption, it is important to differentiate between the training and deploying of an AI model (“Inference”).

While training is particularly energy intensive, deployment is likely to be the main driver of increased electricity demand.

Analists suggests that the training phase accounts for around 20 to 40% of energy use, with 60–70% for inference, and 10% for model development.

6. “The global ICT footprint could reach about 10% by the end of the decade and data centres (and specifically AI) would represent the largest increase in electricity demand.”

“The combination of rapid growth in the use of LLMs (themselves getting bigger) and computing demand are likely to outpace energy efficiency improvements (although significant) resulting in net growth in total AI-related energy use in the coming years.”

Not a subscriber yet?