Musk showcases AI ambitions in chipmaking and autonomous fleets: Terafab to test 10 chip designs in a year, Cybercab targets Dallas after expanding eightfold monthly

According to Zhihu Finance APP, last weekend, Elon Musk, CEO of Tesla (TSLA.US) and SpaceX (SPCX.US), released two consecutive pieces of news pointing in the same direction—taking control of the most critical supply chain and operational links. On October 11, during the X Takeover event at the Texas Austin Gigafactory, he explained to more than 2,000 owners and investors the goal of the Terafab development facility: to produce and test about 10 chip designs per year. On the same day, Tesla’s AI software head Ashok Elluswamy provided another update on X—Austin’s Cybercab fleet has surpassed 300 vehicles, expanding eightfold within 30 days, with Dallas to follow soon.

The commonality between the two lines lies in being scrutinized for their "speed" and "certainty." The market response was relatively lukewarm: ahead of Monday, Tesla’s stock price climbed about 0.6% overnight, SpaceX (SPCX) rose about 1.2%, but retail sentiment for Tesla remained "bearish," and the stock is still down about 15% year-to-date.

10 Designs a Year, and That Building of “Non-Stop Iteration”

The "10 designs" Musk mentioned refers to faster testing and iteration, not putting 10 designs into mass production every year. “We’re aiming for more like 10 designs per year,” he said, noting that it typically takes two years to move a significant chip from design to mass production, which is already considered fast in the industry.

The planned Giga Texas development facility will integrate chip manufacturing, testing, packaging, and mask production all in the same building. Musk explained that the team can test a chip, create a new mask, and then build the next version without having to interrupt iterations. “You make a complete computer in one building, test it, create a new mask, and then make a new chip”—he termed this "closed loop" as a powerful tool.

He openly admitted his outsider status: “I have no idea how to make chips, but we’ll figure it out.” He did not provide a timeline: “It may take several years, but looking at the long term, I’m quite confident we can do it.” When touring competitor factories, he found little custom equipment, whereas Tesla and SpaceX specialize in custom machines; the path is to start with standard equipment, retrofit to boost throughput, and then develop machines in-house for bottleneck steps.

This R&D line is only the spearhead of a bigger plan. Terafab is essentially two plants, each making a chip design to supply Tesla vehicles, Optimus, and space-based AI data centers: the project was announced in March 2026, located in Grimes County, Texas, with phase one investments exceeding $16.8 billion and a reported full-phase cap at $119 billion, planning around 3,000 jobs and targeting over 1 terawatt of AI computing power per year; around 80% for space applications and 20% for ground. Mass production is expected between 2027 and 2029.

Why TSMC Was Involved, and Intel’s Situation

The variable on the supply side appeared the same month. On October 2, tech journalist Tim Culpan reported in his newsletter Culpium that TSMC is exploring how to participate in Terafab’s Texas operations, most likely through owning and operating a new factory itself, with Terafab as the anchor customer (similar to TSMC’s JASM joint venture in Japan and ESMC in Germany); another model is for SpaceX to have majority ownership, TSMC to invest less but provide technology and operational expertise.

On October 3, Musk stated on X: “Just discussions, but maybe something will come of it.” On October 7, he drew the boundaries more clearly: Tesla and SpaceX will construct and operate Terafab themselves, “no question about it.” If TSMC wants to participate, it would at most rent part of the site and not have any operating role. At the same time, Intel CEO Pat Gelsinger declared, “We will continue to participate.”

There are two reasons TSMC was brought to the table. First, its ambitions in Texas are broader—after pledging to invest up to $265 billion in Arizona, it is evaluating new locations for as many as six wafer fabs in Texas, with Dallas and North Texas being candidates. Second, Terafab needs a second technical pipeline: when Intel joined in April, it was designated to provide the 14A process (about 1.4nm), but industry insiders revealed actual progress is "quite limited," with licensing for the 14A process design kit still unclear.

The market quickly read this as bad news for Intel. On Monday, TSMC’s Taiwan shares closed up 3%, ADRs in U.S. markets closed up about 2.8% at $485.80, hitting a record high; Intel closed down about 2.6%, falling for three consecutive days, dropping another 3.18% on Tuesday, only stabilizing after Musk and Gelsinger reassured markets. The response was far greater than the news merits—Intel has rallied over 200% in the past year, but of its $5.8 billion foundry revenue for Q2, only $293 million came from external clients. Terafab is its only “flagship client” in this segment, and sharing the flagship means revisiting valuation assumptions. Intel is set to release its 14A design kit in October, another checkpoint; the company itself admits in its annual report that it may suspend this process if 14A doesn’t secure key external clients.

The sell-side split into two camps. Mizuho’s Jordan Klein believes the Texas center explored by TSMC is “very likely part of this project, and Musk will be the anchor customer”; Wedbush’s Matt Bryson reiterated that TSMC “outperforms the market”; D.A. Davidson’s Gil Luria thinks Intel investors are “overly optimistic”; famed semiconductor analyst Lu Xingzhi raises a more fundamental question—If TSMC funds and provides the process, does this factory still count as Musk’s Terafab?

But the physical clock for chip-making bends for no one’s attitude. TSMC Chairman and CEO C.C. Wei stated at the April earnings call: “There’s no shortcut”—it takes two to three years to build a new fab and one to two more to ramp up capacity.

Cybercab Expands Eightfold in a Month, Next Stop Dallas

Downstream, the pace is much faster. On October 10, Tesla’s official Robotaxi account announced that Austin’s Cybercab fleet “has expanded to over 300 vehicles, growing 8x in 30 days”; Elluswamy reposted, “Two years since the 10/10 Cybercab WeRobot press conference, these cars are everywhere in Austin, and Dallas is next.” Texas Motor Vehicle Registers show that on October 8, Tesla registered a one-time addition of 150 vehicles, boosting the total from 169 to 319—the biggest daily increase since the first batch of seven cars was recorded on August 31.

Currently, the acceleration is obvious: at the commercial debut earlier this month, Texas registrations were just 45, increasing to 169 by October 2, and 319 a week later; in the same period, Waymo was authorized for 1,154 commercial vehicles in Texas (including 359 of the new Ojai model), over 4,000 driverless vehicles nationwide, more than 500,000 paid rides per week, and a cumulative 270 million miles of fully autonomous driving across 15 U.S. markets. Amazon’s Zoox is also testing in Austin.

One clarification: registration count does not equal vehicles in operation. The Texas registry lists only VIN, model year, and type—not location, nor in-use numbers; according to community tracking site RobotaxiTracker, Austin had 275 driverless vehicles in operation last Saturday, with 105 being Cybercabs, noticeably below Tesla’s claim. Also, Tesla has not announced a Dallas launch date, and local rides are still handled by Model Y; California also has not issued it a permit for driverless road testing.

On the regulatory front, a deadline has been set. The U.S. National Highway Traffic Safety Administration (NHTSA) is auditing how Cybercab can meet federal safety standards without a steering wheel or pedals. A special order dated September 10 requires answers under oath to 21 questions, including how a vehicle with no foot controls meets brake standards that require pedal operation. Tesla requested to extend the deadline by eight weeks to November 20, but NHTSA only granted 30 days—new deadline is October 30, with answers requiring a sworn affidavit from a Tesla responsible officer. Prior context still stands: the audit covers approximately “1,000 vehicles (estimated),” more than 20 times the number operating on Austin roads at that time; incomplete responses could result in Department of Justice litigation and fines up to $27,874 per day.

“Gray Cats, Railway Crossings, and Assembly-Tent Lines”

After the first month, user experience gripes are accumulating. Paying riders complain about wait times (often over 45 minutes at launch), incorrect pick-up/drop-off points and butterfly doors, and minor trunk problems; a frequent Austin rider who took about 30 trips commented, “Drop-off and pick-up experience still needs polishing,” and believed Waymo is more mature. Austin fire chief Matt McElearney hopes regulators require such vehicles to have emergency takeover capability and “dynamic certification” throughout the lifecycle. A September survey showed 50% of Americans would not ride a robotaxi without a steering wheel.

Musk’s response is that “slow is fast.” He said expansion seems slow, but the month-over-month growth is high, “because we want to make sure there are no incidents.” “We do not want to cause casualties. If a Tesla robotaxi were to have such an incident, it would be global news.” He mentioned screening risky intersections before rolling into a new city, “especially being extremely careful with railway crossings,” and brought up the challenge of detecting small animals in low visibility—“A grey kitten, at night, on a grey asphalt seam—can we avoid it? By the way, we actually solved this problem.”

In that same interview, he unusually looked back at the bears. During the 2018 Model 3 production crisis, “We had to build 5,000 cars per week, or go bankrupt.” “The shorts were actually right,” Musk said, “they nearly saw our demise.” He recalled setting up a third assembly line under a tent at the Fremont plant, pushing cars off the line with scrap machinery, and “raising output by 30% within three weeks.”

Market and Valuation: Up 3% This Week, Still Down 15% YTD

The two storylines leave a split on stock prices. On October 9, Tesla closed up 2.05% at $382.70, up 3.27% for the week, the best since late August; but it remains down about 15% year-to-date, the worst performer among the “Magnificent Seven.” The company’s market cap is about $1.51 trillion, with a rolling P/E ratio around 355. SpaceX is on the other end—up about 1% year-to-date, up about 2% last week.

The fundamentals will hinge on Q3 deliveries: Tesla delivered 486,532 vehicles in Q3, exceeding expectations by about 25,000, but still down about 2% year-over-year. The delivery news lifted the stock by nearly 5% that day. Automotive operations under pressure and investors waiting for AI and Robotaxi monetization is the current basis for valuation—analysts’ one-year average target price is around $396.57, implying less than 5% upside. The company will release Q3 earnings after the close on October 21, with consensus EPS at $0.23.

Retail sentiment hasn’t recovered with the stock price. On Stocktwits, TSLA sentiment slipped from “neutral” the previous day to “bearish,” with message volume “normal,” while SPCX sentiment is “neutral” but with “high” message volume. Some praised execution pace: “Cybercab’s expansion strategy, logistics, and engineering all work...the team’s likely ahead of schedule”; others pulled no punches: “Fully autonomous driving still isn’t here.”

On the macro side, there’s extra pressure: Last Friday (October 9), all three major indices closed higher (Dow +0.83%, S&P 500 +0.59%, Nasdaq +0.64%), but oil remains above $100 a barrel—Brent December futures at $102.91, down nearly 2%, WTI at $90.03.

Back to the two narratives: one aims to compress a five-year factory build cycle into the pace of “10 designs per year,” while the other must deploy a scrutinized fleet across a second city before the federal October 30 deadline. Mizuho’s Klein believes TSMC and Intel will ultimately share and split roles, D.A. Davidson’s Luria worries Intel investors oversell the story, and Wedbush’s Bryson casts his lot with the definitive winner—the world’s largest foundry. As for Tesla, its valuation is still betting on that not-yet-at-scale fleet.