Author: Birch Forest Dance King
In the 1980s, Japan's semiconductor industry was at its peak, once accounting for over 50% of the global chip market share.
Americans became restless—not because they couldn't buy chips, but because they realized that a country that couldn't control its own "silicon" was effectively handing over its strategic lifeline to others. This led to the *Semiconductor Agreement*, the subsequent resurgence of Intel, and today's *CHIPS Act*.
Four decades later, a similar anxiety has manifested in a more personal way with Elon Musk.
Except this time, the entity seeking to control "silicon" is not a country, but an individual—and the cluster of companies under his control.
On May 6th, Bloomberg disclosed a document: SpaceX has formally proposed a $55 billion investment in Grimes County, Texas, to initiate the construction plan for the "Terafab" chip manufacturing facility. If all subsequent phases proceed, the total capital expenditure could surge to $119 billion.
First, let's translate these numbers to grasp the scale.
$55 billion is more than double TSMC's total revenue in 2023. $119 billion is close to NVIDIA's peak revenue for its entire 2024 fiscal year. This isn't an "investment"; it's a high-stakes gamble—or perhaps, a strategic declaration.
This project is a joint initiative by SpaceX and Tesla. Musk's logic is clear: His group of companies—SpaceX, Tesla, xAI—consumes a massive amount of GPU computing power annually. Training Grok requires chips, the Starlink ground station network requires chips, Tesla's Autopilot requires chips, and the future humanoid robot Optimus will also require chips.
Rather than paying NVIDIA billions every year, why not invest that money in himself?
From a strategic logic standpoint, this is impeccable.
01 Musk's Ambition for Vertical Integration
To understand Terafab, you first need to understand what Musk has been doing over the past two years.
In 2025, xAI acquired the social media platform X. Earlier this year, SpaceX absorbed xAI in an all-stock deal. Meanwhile, SpaceX's IPO plan is advancing, with the roadshow window reportedly opening around June 8th, 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 is to be added to the portfolio.
DataCenter Knowledge's analysis hit the nail on the head: Terafab should best not be understood as a "factory," but as *a "full-stack AI infrastructure strategy"* —an attempt to unify compute production, energy procurement, and compute deployment under the same roof.
This is like Amazon not just wanting to sell books, but also wanting to build its own shipping network, its own data centers, its own logistics satellites—except Musk is adding one more layer on top: building a "factory for producing computation."
Apple's move to design its own A-series chips was considered one of the most successful vertical integration decisions in tech history. But Apple only "designs" the chips; manufacturing is still outsourced to TSMC. What Musk wants to do is take over the manufacturing as well.
That's an ambition even Apple hasn't dared to touch.
02 Behind the "15-Year Strategy"
However, there's always a deep chasm between strategic vision and engineering reality.
Chip analyst Ben Bajarin from Creative Strategies used an interesting phrase, *saying Musk is pursuing a "15-year strategy"* —which sounds like praise, but the implication is: don't expect returns in the short term.
Morgan Stanley's forecast is more direct. They estimate that even under the most optimistic construction assumptions, *the initial chip output from Terafab wouldn't arrive until mid-2028 at the earliest*. That's over two years from now. And where will the technological generation of AI chips be by then? No one knows.
The brutality of semiconductor manufacturing lies in it being one of the world's least forgiving industries for "PPT promises."
Building an advanced-node wafer fab typically takes 3-5 years, requiring extremely precise lithography machines (only ASML can make high-end EUV machines globally), thousands of highly specialized engineers, a stable supply of ultra-pure water and electricity, and an entire suite of sophisticated cleanroom management systems. Intel has spent hundreds of billions of dollars and years of time, yet still struggles to catch up with TSMC in process technology.
Finance Monthly's analysis directly pointed out this risk: chip projects are prone to underestimating execution difficulties—slow, expensive, easily delayed, reliant on specialized machinery, skilled labor, and already highly strained supply chains.
Interestingly, when asked about Terafab, Intel CEO Pat Gelsinger offered a subtly worded response, saying he was *"excited to explore innovative ways to re-architect silicon technology."* This can be interpreted as a signal of willingness to cooperate, or as an indirect acknowledgment of the tight market supply and demand—or perhaps both.
03 More Than Just Business
But viewing Terafab solely through the lens of business investment returns might be applying the wrong analytical framework from the start.
The truly interesting aspect of this project is how it reflects the AI industry's deepening obsession with "compute autonomy."
Over the past three years, the nature of the AI arms race has evolved from "whose model is smarter" to "who can secure more computing power." NVIDIA's H100, H200, GB200 are in critically short supply, and TSMC's advanced node capacity is booked years in advance. Microsoft, Google, Amazon, and Meta have invested tens of billions of dollars in developing their own AI chips.
Musk's logic is fundamentally the same as that of these tech giants: *In the AI era, computing power is the means of production. Whoever controls the chips controls AI.*
And the emergence of Terafab, layered on top of SpaceX's upcoming IPO, adds another dimension of complexity. A tech media editor commented that part of the announcement's design was to *"tie Tesla, which is under pressure, to the narratives of SpaceX's impending IPO and AI supercomputing."*
This assessment may not be entirely fair, but it's not baseless. Just a few months ago, Musk himself admitted that *"xAI was not built correctly,"* forcing SpaceX to step in for the integration. Against this backdrop, Terafab's high-profile unveiling serves as both a strategic maneuver and a capital narrative—two things that are not mutually exclusive.
Musk never does just one thing.
Right now, the launch window for Starship Flight 12 is between May 12th and 18th, and the Dragon cargo spacecraft will fly to the International Space Station on the same day. SpaceX's rocket business continues to operate with great momentum.
And in Grimes County, Texas, a plot of land that could potentially reshape the AI-era chip supply chain remains just coordinates on a document.
No one can guarantee that $119 billion will turn into a fully operational wafer fab. But one thing is already certain—when a company famous for building rockets decides to build chips, the boundaries of this industry are once again being redefined.
