🔥 What Happened
Elon Musk's SpaceX has filed plans to invest $55 billion—potentially ballooning to $119 billion—in a sprawling semiconductor manufacturing complex in Grimes County, Texas. The project, called Terafab, isn't just another chip factory. It's an explicit declaration of independence from TSMC, Samsung, and the entire global chip supply chain that Musk's companies have been fighting for capacity at.
The filing, made public on Wednesday via a property tax abatement notice, caught the industry off guard with its sheer scale. We're talking about a single facility complex that could cost more than TSMC's entire Arizona investment portfolio combined. The first phase alone ($55 billion) dwarfs most semiconductor projects in history.
And here's the kicker: SpaceX isn't just building a fab. According to the S-1 registration document reviewed by Reuters, the company plans to manufacture its own GPUs. A rocket company wants to make graphics cards. Let that sink in.
🧠 Why This Matters
This is arguably the most aggressive vertical integration play in tech history. Musk is taking on the hardest hardware problem on Earth—advanced semiconductor fabrication—while simultaneously running Tesla, SpaceX, a social media platform, and just merged xAI into SpaceX at a $1.25 trillion valuation.
Why does he think he can pull this off? Because he has to.
During Tesla's January earnings call, Musk laid it out bluntly: key chip suppliers can't possibly produce enough hardware to satisfy his companies' needs. Tesla needs chips for self-driving systems. SpaceX needs them for Starlink satellites and Starship. xAI needs them for data centers. The humanoid robot (Optimus) needs them by the million.
The math doesn't work without controlling your own silicon. Nvidia's H100/B200 shortage is a preview of a much bigger problem. When you're planning to deploy millions of humanoid robots and a satellite constellation, you can't bid for capacity on the open market every quarter.
📊 Deep Dive
Let's get into the specifics, because the numbers here are genuinely absurd.
The Price Tag:
- Phase 1: $55 billion
- Full buildout: Up to $119 billion
- For comparison: TSMC's entire 2025 capital expenditure was ~$35 billion
- For more comparison: Intel's new Ohio fab complex is projected at $20 billion
The Location:
- Grimes County, Texas (about 90 minutes from Austin)
- A newly designated reinvestment zone for tax purposes
- Public hearing on June 3 to consider property tax abatement
- Local officials are clearly expecting this to reshape the local economy
The Partners:
- Intel is on board to use its 14A process for chip production at Terafab
- This is a huge deal for Intel's foundry business—their first major external commitment
- Intel's stock more than doubled in April, its best month ever, largely on Terafab momentum
- Tesla is building a $3 billion research fab in Austin as a proving ground
What They're Making:
- GPUs (manufactured in-house, per S-1 filing)
- Chips for Tesla self-driving systems
- Chips for Optimus humanoid robots
- AI data center compute hardware
- Logic, memory, and advanced packaging all under one roof
The Timeline:
- SpaceX is targeting a June IPO at a $1.75 trillion valuation
- xAI is being fully absorbed into SpaceX, rebranded as SpaceXAI
- Intel's 14A node is expected in 2027-2028 timeframe
- Musk has called this a 15-year strategy
⚠️ The Catch
Let's be honest about the risks here, because they are enormous.
Semiconductor manufacturing is brutally hard. TSMC has been perfecting its craft for four decades. Intel, despite decades of experience and hundreds of billions in investment, has struggled to keep pace. You can't throw money at a fab and expect it to work—ask Intel, whose 7nm/Intel 4 delays cost them their competitive edge.
The filing itself admits this. SpaceX's S-1 notes there is no assurance it will meet its Terafab objectives within expected timelines, or at all. That's not a boilerplate disclaimer—that's real risk recognition.
Analysts are skeptical. Ben Bajarin of Creative Strategies called it a 15-year strategy and noted you can't just wake up and decide to be a foundry. The semiconductor industry has constraints across equipment, materials, talent, and process engineering that don't dissolve with a big check.
$119 billion is a lot of money. Even for a company valued at $1.75 trillion, that's nearly 7% of your entire valuation bet on one factory. And that's before cost overruns, which are the norm in chip fabrication.
Geopolitical risk cuts both ways. Yes, domestic production insulates from Taiwan Strait tensions. But the US doesn't have the same equipment supply chain, chemical supply chain, or skilled workforce that Asia has built over generations.
🎯 What Happens Next
June 3, 2026: Grimes County holds its public hearing on the tax abatement. If approved, this opens the door for detailed planning and construction.
June 2026: SpaceX's IPO filing will likely contain more detail on Terafab's financial structure. Investors will be watching closely—a $55 billion capex commitment is material to any valuation model.
2027-2028: Intel's 14A process comes online. If Terafab's first phase aligns with this timeline, we could see initial wafer production by late 2028 or 2029.
The more immediate impact: Expect TSMC and Samsung to accelerate their US fab plans. Competitors don't sit still when a customer decides to become a competitor.
For Intel: This is existential. If Terafab succeeds, Intel becomes a critical partner in the Musk ecosystem. If it fails, Intel's foundry pivot loses its flagship external customer. Intel's best month ever was April 2026—they need to deliver.
🧩 Bigger Picture
Terafab represents something bigger than a single factory. It's the logical endpoint of a trend that's been building for years: the hyperscalers becoming hardware companies.
Amazon builds its own Graviton CPUs and Trainium AI chips. Google designs TPUs. Microsoft makes Maia chips. Meta builds MTIA. Apple designs everything in-house. The common thread is that they all still fab through TSMC.
Musk is the first to say: Fine, we'll build the factory too. That's a dramatic escalation of vertical integration.
If it works, it reshapes the semiconductor industry. Every hyperscaler will ask: should we be doing this too? If SpaceX—a rocket company—can build chips, why can't we?
If it fails, it's the most expensive lesson in humility the hardware world has ever seen.
Either way, the filing that dropped on Wednesday in rural Texas is one of the most important hardware documents of 2026. A rocket company filed to build a semiconductor fab that could cost more than some countries' GDPs, to make its own GPUs, because it couldn't buy enough chips anywhere else.
That's the state of AI hardware demand in 2026—and it's only going to get more intense.
