Author: Hu Lin Dance King

In the 1980s, Japan's semiconductor industry was at its peak, once accounting for over 50% of the global chip market.

Americans couldn't sit idly by—not because they couldn't buy chips, but because they realized that if a country can't control its own "silicon," it is equivalent to handing over its strategic lifeline to others. Thus, there was the "Semiconductor Agreement," the subsequent revival of Intel, and today’s "Chip Act."

Forty years later, the same anxiety has appeared in Elon Musk in a more personal way.

But this time, the one wanting to control "silicon" is not a country, but a person—and a group of companies under him.

On May 6, Bloomberg revealed a document: SpaceX officially proposed to invest $55 billion in Grimes County, Texas, to start the construction plan for the "Terafab" chip manufacturing facility. If all subsequent phases are pushed forward, the total capital expenditure could surge to $119 billion.

Let’s convert the numbers to feel the scale.

$55 billion is about more than twice TSMC's total revenue for 2023. $119 billion is close to NVIDIA’s entire revenue peak for the fiscal year 2024. This is not an "investment"; this is a gamble—or rather, a strategic declaration.

This project is a joint plan of SpaceX and Tesla. Musk's logic is clear: His group of companies—SpaceX, Tesla, xAI—consumes vast amounts of GPU computing power each year. Training Grok requires chips, the Starlink ground station network requires chips, Tesla's autonomous driving requires chips, and the future humanoid robot Optimus also requires chips.

Instead of sending money to NVIDIA every year, it's better to send it to himself.

From a strategic logic standpoint, this is flawless.

01 Musk's Vertical Integration Ambition

To understand Terafab, you first need to understand what Musk has been doing for the past two years.

In 2025, xAI acquired the social media platform X. Earlier this year, SpaceX swallowed xAI entirely in stock. Meanwhile, SpaceX's IPO plan is advancing, with the roadshow window reportedly set to open around June 8, and the S-1 registration document expected to be submitted later this month.

This is an increasingly dense web: rockets, satellite broadband, AI models, social platforms, electric vehicles, humanoid robots... and now, chip manufacturing must be incorporated into the layout.

DataCenter Knowledge's analysis hits the nail on the head: Terafab should not be understood as a "factory," but as a "complete stack AI infrastructure strategy"—attempting to unify computing production, energy procurement, and computing power deployment under one roof.

It's like Amazon not just wanting to sell books, but to build a freight network, build data centers, and build logistics satellites—only Musk wants to add one more thing on top of that: to build a "factory for computing production."

Apple's self-developing A-series chips was considered one of the most successful vertical integration decisions in tech history. But Apple only "designs" chips, while manufacturing is still entrusted to TSMC. What Musk wants is to take manufacturing as well.

This ambition is something even Apple hasn't dared to touch.

02 Behind the "15-Year Strategy"

However, between strategic vision and engineering reality, there is always a deep chasm.

Ben Bajarin, a chip analyst at Creative Strategies, used an interesting expression, saying Musk is pursuing a "15-year strategy"—this statement sounds like praise, but the underlying meaning is: don't expect to see returns in the short term.

Morgan Stanley's forecast is even more direct. They estimate that even under the most optimistic construction assumptions, Terafab's initial chip output will not come until mid-2028. From today, that's still over two years away. And by then, no one knows where AI chip technology will have progressed.

The harsh reality of semiconductor manufacturing is that it is the most unforgiving industry for "PPT commitments" in the world.

Building an advanced process wafer plant typically takes 3-5 years, requires extremely precise lithography machines (only ASML can produce high-end EUV), needs thousands of highly specialized engineers, requires stable ultra-pure water and power supply, and necessitates a complete system for managing clean rooms. Intel has spent hundreds of billions of dollars and years of time and is still struggling to catch up with TSMC in process technology.

Finance Monthly’s analysis directly points out this risk: chip projects are often underestimated in their execution difficulty, slow, expensive, prone to delays, and dependent on specialized machinery, skilled labor, and an already highly strained supply chain.

Interestingly, Intel CEO Pat Gelsinger gave a subtly worded response when asked about Terafab, stating, "Excitedly exploring innovative ways to reconstruct silicon process technology." This statement can be interpreted as a signal for collaboration, a validation of the market supply and demand tension, or even both.

03 Not Just Business

But if we only view Terafab as an investment return calculation of a business, we might have been wrong about the analytical framework from the start.

What’s really interesting about this matter is that it reflects the entire AI industry's increasingly deep obsession with "computing power autonomy."

In the past three years, the essence of the AI arms race has shifted from "whose model is smarter" to "who can access more computing power." NVIDIA's H100, H200, GB200 are all in high demand, and TSMC's advanced process capacity has been secured for years in advance. Microsoft, Google, Amazon, and Meta have invested billions of dollars in developing their own AI chips.

Musk's logic is actually the same as that of these big firms: in the AI age, computing power is the means of production; whoever controls the chips controls AI.

The emergence of Terafab, coupled with SpaceX's upcoming IPO, adds another layer of complexity to the meaning. A technology media editor commented that part of the design of this announcement was to "tie pressure-stricken Tesla with the soon-to-be-listed SpaceX and the AI supercomputing narrative together."

This judgment may not be completely fair, but it is not unfounded either. Just months ago, Musk admitted that "the construction of xAI was not right," and had to be integrated by SpaceX. Against this backdrop, Terafab's high-profile unveiling is both a strategic layout and a capital narrative—these two things are not contradictory.

Musk never just does one thing.

Now, the launch window for Starship Flight 12 is between May 12 and 18, and the Dragon cargo spacecraft will also head to the International Space Station on the same day. SpaceX's rocket business continues to operate vigorously.

And in Grimes County, Texas, a piece of land that could rewrite the chip supply chain landscape of the AI era is still just a coordinate on a piece of paper.

$119 billion, whether it can turn into a truly operational wafer factory, no one can guarantee. But one thing is already certain—when a company famous for building rockets decides to make chips, the boundaries of this industry have once again been redefined.