Nadella: AI’s New Limit is the Power Grid, GPU War Concludes

Key Takeaways

1. Microsoft faces challenges in AI not related to chip supply, but to the availability of fully developed data centers.
2. Current obstacles include limitations of local grids, planning delays, and power delivery bottlenecks.
3. Power shortages are impacting AI data centers, which consume electricity similar to small cities.
4. Major cloud companies are seeking long-term energy agreements and exploring on-site generation options.
5. The future of AI competition will depend on securing reliable and scalable power sources, not just GPUs.

Microsoft’s CEO, Satya Nadella, mentioned that the surge in AI has encountered a new challenge, and it isn’t related to GPUs.

In a recent episode of the BG2 podcast, where he spoke with OpenAI’s CEO, Sam Altman, Nadella pointed out that Microsoft is no longer facing constraints regarding chip supply. Instead, the key issue is the availability of adequately powered and fully developed data centers—these “warm shells” that are near grid capacity—needed to activate all those accelerators.

As he put it, one might have numerous chips stored in inventory that just can’t be connected.

A Shift in Industry Challenges

This marks a significant change from recent years when the focus was primarily on the shortages of Nvidia GPUs and various supply chain issues. According to Nadella, the main obstacles now are limitations of local grids, delays in planning and permitting, and bottlenecks in power delivery that can hinder or even completely stop AI projects long after the necessary hardware has been procured.

Power Supply Issues

These power shortages have further consequences. Data centers dedicated to AI are consuming electricity on par with small cities, leading major cloud companies to secure long-term energy agreements, consider on-site energy generation, and even investigate the use of small modular nuclear reactors to ensure future projects can operate smoothly.

The takeaway for both investors and regulators is clear: the upcoming stage of the AI competition will not solely depend on who can acquire the most GPUs, but also on who can obtain the most dependable and scalable power sources to sustain them.

Source:
Link