Apple starts shipping US-made AI servers from Texas

Apple has begun shipping servers for AI applications from its Houston, Texas, the company said on October 23.

The servers, built with Apple’s own silicon, will support Apple Intelligence and Private Cloud Compute services.

Apple said this move is part of its previously announced commitment to spend US$600 billion in the US on manufacturing, suppliers, and related projects.

The company’s servers were previously made overseas.

Apple plans to expand the Houston facility and increase production next year.

The company said the factory is expected to create thousands of jobs.

Apple CEO Tim Cook has emphasized the company’s investments in US manufacturing, including partnerships with domestic semiconductor suppliers.

Apple also opened a manufacturing academy with Michigan State University in July.

🔗 Source: CNBC

🧠 Food for thought

Implications, context, and why it matters.

Apple’s chip plan favors packaging gains over quick node jumps

Apple may keep Taiwan Semiconductor Manufacturing Company (TSMC)’s 3-nanometer N3P process (a performance-enhanced 3 nm variant) for the A20 through late 2026, with 2 nm likely pushed to the A21 in 2027, as Wccftech ties the move to cost 1.
The A20 is slated to use TSMC Chip-on-Wafer-on-Substrate (CoWoS) packaging to pack parts closer for speed gains without a 2 nm jump 1.
This approach reaches the M5, which is expected to use TSMC System on Integrated Chips (SoIC) with 3D chip stacking (vertically connecting multiple chips) for future Macs and AI servers, targeting mass production in 2025–2026 2.
TSMC 2 nm trial output topped 60% yield in December 2024, yet advanced node costs still strain even high-volume buyers like Apple 1.

Texas AI power surge opens spots in Electric Reliability Council of Texas (ERCOT) growth zones

Texas had 279 data centers in September 2024, with over half in Dallas–Fort Worth 3. POWWR cites the Electric Reliability Council of Texas (ERCOT) and projects a 43 GW load increase by 2030 3. That equals about 30 new nuclear plants, driven by data centers and AI work 3.
ERCOT has 136 GW of grid connection requests from large users, up from 99 GW in March 2025 and 41 GW a year earlier 3. Some single data center plans ask for 1 GW each 3.
An AI query can use 10 to 30 times more electricity than a standard web search, which speeds up load growth 3.
Investors such as independent power producers, data center cooling vendors, and retail energy providers (electricity retailers that sell directly to businesses) can tie capital to ERCOT load forecasts to get in first where AI compute grows fastest 3.