October's Big AI Story Is About Capacity, Not Just Hardware

Summary: TSMC just posted another record quarter and says AI demand is still climbing, while Nvidia and AMD used this week's Open Compute Project summit to push whole racks, not just chips. The real bottleneck this month is factory and packaging capacity, and everyone from AMD to Elon Musk is scrambling for more of it.

Checked on Oct 15, 2026.

TSMC: another record, and a Texas question

The world's biggest contract chipmaker reported third-quarter net profit of NT$763.1 billion, up 68.7% from a year earlier, its sixth straight record quarter, per CNBC. Operating margin hit 58.7%, above the top of its own guidance, and TSMC guided fourth-quarter revenue to $52.3 billion to $54.3 billion. It also said capital spending for 2026 through 2028 will be "significantly higher" than the prior three years.

The headline beyond the numbers: CEO C.C. Wei said on the earnings call that TSMC has sent teams to Texas to evaluate a site beyond its Arizona base, and that it is open to working with Musk's Terafab project, but "the fab got to be run by TSMC." No deal has been reached, and no site, node, or timeline has been announced. (More on Terafab in our SpaceX post.)

Nvidia: Vera Rubin is shipping, and Vera CPUs are going solo

Nvidia's next-generation Vera Rubin platform is no longer a slide. According to a summary of Nvidia's Q2 fiscal 2027 call, production shipments began in August, with racks running at partners including CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud and Nebius. CoreWeave says it has Vera Rubin NVL72 running for a named customer, Cognition, in limited availability.

At the OCP Global Summit in San Jose this week, Nvidia contributed Vera Rubin rack and compute-tray designs to the Open Compute Project through its MGX architecture, and published a spec for a standalone Vera CPU rack aimed at agentic AI work like tool calling and sandboxed code execution. An Nvidia product lead told reporters that dedicated Vera CPU racks are "shipping now". Lenovo, for its part, says it is shipping Vera Rubin NVL72 globally.

AMD: Helion joins Helios

AMD answered with its own rack. On Oct. 12 it introduced AMD Helion, an open blueprint for agentic AI racks: one CPU-only version with up to 32,768 cores of 6th Gen EPYC, and an all-in-one version mixing EPYC CPUs and Instinct MI350P GPUs. AMD says Helion configurations are planned for November 2026 and that it will contribute the spec to OCP.

Meanwhile its bigger Helios GPU rack is ramping. In her OCP keynote recap, AMD says initial Helios systems are installed and running at OpenAI, Meta, Microsoft and Oracle. That ties back to the 6-gigawatt AMD-OpenAI deal from October 2025, whose first 1-gigawatt MI450 deployment was slated for the second half of 2026.

The squeeze: packaging slots

All of this runs through the same few factories. CNBC reports Nvidia, Apple and Broadcom are in a "fierce" battle for TSMC capacity. A widely shared Morgan Stanley estimate puts Nvidia at roughly 60% of 2026 demand for CoWoS advanced packaging. Treat that as an analyst projection, not company data. AMD CEO Lisa Su, visiting Taiwan, said AMD plans to "significantly increase supply in 2027" and needs more advanced wafer capacity.

Intel: the 14A test

Intel said in May it planned to hand version 0.9 of its 14A process design kit, which CEO Lip-Bu Tan called "the Holy Grail," to outside customers in October, with risk production in 2028 and volume in 2029. We could not find an Intel announcement confirming the release has happened yet. Tan also reaffirmed Intel's role in Terafab in Tokyo on Oct. 7, as relayed in a post Musk reposted. Tesla has said it plans to use Intel 14A for Terafab chips, per a report on Yahoo Finance.

Apple: M6 Macs reportedly coming Oct. 28

Apple's M6, its first chip on a 2nm process, shipped first in the Mac mini after an Aug. 25 announcement. Bloomberg's Mark Gurman reports Apple is inviting media to an Oct. 28 launch experience for touchscreen MacBooks, a new iPad mini and M6 Macs. That is a report, not an Apple announcement.

Bottom line

Demand is not the question this month. Capacity is: who gets wafers, who gets packaging, and who gets to run the fabs.


Enjoyed this? Subscribe to The Weekly Three: https://theweeklythree.beehiiv.com/subscribe

Sources

Comments

Popular posts from this blog

SpaceX's Louisiana Tax Deal Lets It Contest Public Records Before They're Released

Who Gets to See the Terafab Deal? A Citizens Group Takes SpaceX to Court

Tesla Will Wind Down Cybertruck to Make Room for Cybercab, Musk Says