Imagine if one company decided to build a second Samsung — from scratch — in Texas, in just 6 years.
That is exactly what Elon Musk just told his suppliers.
According to Nikkei on Nov 9, 2026, Musk’s Terafab told Japanese equipment makers it wants to make 1 million silicon wafers per month by 2032. Of that, 700,000 wafers per month will be just for memory.
Why is that insane? Because Samsung — the biggest memory maker on Earth, after 50 years of building fabs — makes 650,000 to 700,000 wafers per month today.
So Musk is saying: “We will match Samsung’s entire memory empire, plus build another 300,000 wafers for logic chips, by 2032. And we have never built a chip fab before.”
Quick Answer: Yes, Terafab is aiming for 700,000 memory wafers/month. No, it is not built yet. The first test chip is May-June 2027. Mass production is targeted for late 2028. And the whole thing will cost up to $119 billion. This is either the boldest industrial bet in history, or the most expensive science project ever.
Wait, What Is A Wafer? (In 20 Seconds)
Forget the tech jargon. Think pizza.
- A wafer = a pizza base. A super-flat, super-pure silicon disc, 300mm wide (about a dinner plate).
- Chips = slices. You print hundreds of chips onto one wafer, then cut them out. One wafer = 500-800 AI chips, or 1,000+ memory chips.
- Wafers per month = how many pizzas your kitchen can make. Samsung’s kitchen makes ~700,000 pizzas a month. TSMC makes ~3 million pizzas a month (all types).
- Terafab wants to make 1 million pizzas a month. Just for itself. No customers. Just for Tesla cars, Optimus robots, and SpaceX satellites.
Now you get why this is crazy. You don’t just build a bigger kitchen. You need ovens that cost $200 million each (EUV machines from ASML), air 1,000x cleaner than a hospital, and engineers who have spent 20 years learning how not to ruin the pizza.
The Numbers – Terafab vs The World
| Who | Memory Wafers / Month | Logic Wafers / Month | How Long They’ve Been Doing This |
|---|---|---|---|
| Terafab (Target 2032) | 700,000 | 300,000+ | 0 years – hasn’t made 1 chip yet |
| Samsung (Today) | 650,000 – 700,000 | ~300,000 | 50 years |
| SK Hynix (Today) | ~565,000 – 660,000 | ~0 (memory only) | 40 years |
| Micron (Today) | ~200,000 – 250,000 | ~0 | 45 years |
| Rapidus, Japan (Target) | 0 | 25,000 | New, backed by Japan govt |
| TSMC (Today, all chips) | 0 (TSMC doesn’t make memory) | ~3,000,000 | 38 years |
Put it simply: Terafab wants to be Samsung + a mini-TSMC, combined, in one campus in Texas. Samsung has 5 giant campuses in Korea to do what Terafab wants to do in 1 campus in Grimes County, Texas.
Why Does Musk Even Want This? The Real Reason Is Not About Money
WCCFTech says “vertical control.” That sounds boring. Here is the real story:
1. He Can’t Buy Enough Chips
Musk said it himself in March 2026: “We would like Samsung and TSMC to expand as quickly as they can and we will buy all of their chips. But there’s a maximum rate they’re comfortable expanding, and that rate is much less than we would like. So we either build Terafab, or we don’t have chips.”
He needs 100 to 200 BILLION chips per year for:
- Optimus robots: Every robot needs 1 AI brain + memory. If he makes 10 million robots, that’s 10 million chips.
- Cybercab / Robotaxis: Every car needs AI5 / AI6 chip + memory.
- SpaceX DataSats: This is the wild one. SpaceX wants to put AI data centers IN SPACE, powered by the sun. Those satellites need radiation-proof chips that don’t exist yet.
All the fabs on Earth combined can only make ~2% of what he says he will eventually need. So he has to build his own.
2. Memory Is The New Oil
You have felt this. RAM prices in 2026 are insane. DDR5 for your PC costs 3x what it did in 2024. Why? Because AI ate all the memory.
Nvidia’s Rubin GPUs need HBM4 memory — stacked memory that takes 3x more wafers to make than normal RAM. Samsung and SK Hynix now have less than 10 days of finished memory inventory. They are prioritizing AI customers over you.
Musk doesn’t want to beg Samsung for memory. He wants to make his own. 700,000 wafers of it.
3. He Hates Waiting In Line
Right now, if Tesla wants a chip, it has to wait behind Apple, Nvidia, AMD, Qualcomm at TSMC. If SpaceX wants a chip, same line. Terafab is Musk cutting the line by building his own line.
What Is Terafab Actually Building? (It’s Not One Fab)
This is the part most articles get wrong. Terafab is not a factory. It’s a city.
| Part | What It Does | Where |
|---|---|---|
| Austin Research Fab | Small test fab, learns how to make chips | North campus of Giga Texas, Austin |
| Gibbons Creek Main Fab | The big one – 100 million sq ft, 4 phases | Grimes County, Texas |
| Power Plants | Own natural gas plants – Terafab needs as much power as a small country | On-site |
| Water System | Own water purification – fabs use millions of gallons of ultra-pure water per day | On-site |
| Packaging + Test | Normally done in Taiwan/Malaysia. Terafab does it all in Texas. | Same campus |
100 million square feet. That is 1,700 football fields. Or 2x the size of Tesla’s Giga Texas. Or roughly the size of Manhattan’s Central Park, but filled with cleanrooms.
And the craziest part: Everything under one roof. Design, lithography, logic, memory, packaging, testing. No company on Earth does this. TSMC does logic, Samsung does memory + logic but in different campuses, Micron does memory only. Terafab wants to do ALL of it, in ONE place.
The Timeline – From Zero To Samsung In 6 Years?
| When | What Musk Says Will Happen | Reality Check |
|---|---|---|
| March 2026 | Musk announces Terafab at Seaholm Power Plant, Austin | ✅ Happened – $20-25B announced |
| April 2026 | Intel joins as tech partner with 14A process | ✅ Happened – Intel needs a win, Musk needs tech |
| Aug 2026 | $16.8B committed for Phase 1, 3,000 jobs | ✅ Happened – Tax filings in Texas confirm |
| Oct 2026 | Musk admits on X: “I have no idea how to make chips” | ✅ Happened – At least he’s honest |
| Oct 2026 | TSMC in talks to join Terafab (Musk confirms) | 🟡 Talks only – No deal yet |
| Nov 2026 (NOW) | Terafab tells suppliers: “We want 1M wafers/month by 2032, delivery within months” | ✅ Nikkei report Nov 9, 2026 – This article |
| May-June 2027 | First SRAM test chip at Austin research fab | 🟡 Target – SRAM is a baby test, not a real AI chip |
| H2 2028 | Mass production starts at Gibbons Creek | 🔴 Very aggressive – Musk himself says “large error bar” |
| 2030 | First fab at 100,000 wafers/month | 🟡 Plausible if 2028 happens |
| 2032+ | 1,000,000 wafers/month (700k memory + 300k logic) | 🔴 Moonshot – No one has ever done this |
For context: TSMC took 10 years to go from 0 to 100,000 wafers/month. Samsung took 15 years to reach 700,000 memory wafers. Musk wants to do both in 6 years.
Why This Could Actually Work (The Bull Case)
- 1. Musk Has Done Impossible Before: People laughed at SpaceX landing rockets. They laughed at Tesla making EVs. Both now dominate.
- 2. He Has The Money (Sort Of): $119B total is huge, but Tesla + SpaceX + xAI combined are worth $2+ trillion. And Texas is giving massive tax breaks.
- 3. He Has A Guaranteed Customer: Normal fabs need to find customers. Terafab’s customer is itself. Tesla + SpaceX will buy EVERYTHING it makes. No sales team needed.
- 4. He Is Hiring Everyone: Terafab is poaching engineers from TSMC, Samsung, Intel — and it’s next door to Samsung’s Taylor, Texas fab, so talent is right there.
- 5. Intel 14A + TSMC Talks: He is not inventing the process from scratch. He is licensing Intel’s 14A (2nm class) and talking to TSMC. That’s smart.
Why This Could Fail Spectacularly (The Bear Case)
- 1. Making Chips Is NOT Like Making Cars: A car factory can tolerate dust. A chip fab cannot tolerate ONE dust particle. One speck = dead chip. It takes 5 years to learn yield (how many good chips per wafer). Musk has 0 years.
- 2. Memory Is Even Harder: Logic chips are hard. Memory chips are a different universe. Samsung has 50 years of memory secrets. Terafab has 0. Memory needs different machines, different chemistry, different everything.
- 3. The Power Problem: 1 million wafers/month needs ~30 gigawatts of power. That’s like powering New York City. Terafab is building its own gas plants, but Texas grid already struggles in summer.
- 4. The Talent War: There are only ~20,000 world-class fab engineers on Earth. TSMC, Samsung, Intel, Micron, Rapidus, and Terafab all want them. You can’t just hire them like baristas.
- 5. The Money Problem: Tesla’s profit fell 46% in 2025. Free cash flow $6.2B. Terafab Phase 1 alone is $16.8B. Tesla’s 2026 capex is already $20B+. Where does the $119B come from if AI bubble pops?
- 6. Musk’s Track Record On Timelines: Full Self-Driving “next year” since 2016. Cybertruck 2 years late. Optimus “in factories” 2024 — still not. Terafab 2028? Add 2-3 years in Musk Time.
What Does This Mean For YOU?
You don’t buy wafers. So why should you care?
If You Are A Gamer / PC Builder
Short term (2026-2028): Nothing changes. RAM stays expensive. Terafab won’t help you for years.
Long term (2030+): If Terafab works, it could flood the market with memory and crash RAM prices. More supply = cheaper DDR5, cheaper GPUs. If it fails, Samsung + SK Hynix keep charging whatever they want.
If You Are An Investor
Watch 3 stocks:
- ASML: Sells EUV machines. If Terafab orders 100+ EUV tools, ASML wins big. Each tool = $200M.
- Intel: If 14A works for Terafab, Intel’s foundry business is saved. If it fails, Intel falls again. Intel stock already fell 4% when TSMC-Terafab talks leaked.
- Samsung / SK Hynix: If Terafab succeeds, they lose their biggest customer (Tesla) + face a new competitor. If it fails, they keep their monopoly and keep raising prices.
If You Live In Texas
3,000 jobs now, potentially 20,000+ later. Grimes County is about to become the next Austin. House prices near Gibbons Creek are already rising. But also: massive power and water use in a state that already has blackouts.
The One Sentence That Explains Everything
Musk said in October 2026: “Build Terafab, or have no chips.”
He is not building Terafab because he wants to beat Samsung. He is building it because he thinks if he doesn’t, Tesla and SpaceX will die from chip starvation.
That is either genius, or the most expensive panic attack in business history.
Is Terafab real or just an announcement?
Real, but early. Tax filings in Texas confirm $16.8B Phase 1, construction has started at Gibbons Creek, and suppliers in Japan and Korea confirm Terafab procurement teams visited them in Oct-Nov 2026. But no chip has been made yet.
How much will Terafab cost in total?
Texas filings say up to $119 billion across 4 phases. Phase 1 alone is $16.8B. For comparison, TSMC’s Arizona fabs cost $165B for 3 fabs.
When will Terafab make its first chip?
Target: First SRAM test chip May-June 2027 at Austin research fab. Mass production target: H2 2028 at Gibbons Creek. Most analysts expect 2029-2030 for real volume.
Will Terafab sell chips to others?
No. As of Nov 2026, Terafab is a captive fab — only for Tesla, SpaceX, xAI. No outside customers. That is what makes the economics so risky — no one else shares the $119B bill.
Is 700,000 wafers per month really equal to Samsung?
Yes, for DRAM only. Samsung makes 650,000-700,000 DRAM wafers/month in 2026. Terafab wants 700,000 memory wafers/month by 2032. But Samsung will also grow — by 2030, Samsung could be at 1.35M wafers/month. So Terafab is chasing a moving target.
What process will Terafab use?
Intel 14A (2nm class) for logic chips. Memory process is still unclear — likely licensed from Samsung or SK Hynix, or developed with Intel. This is the biggest unknown.
Final Take: Don’t bet against Musk, but don’t bet your house on Terafab either. If it works, it rewrites the chip industry. If it fails, it will be a $119B lesson that making chips is harder than landing rockets.
Sources: Nikkei Nov 9, 2026 via WCCFTech, The Elec, TrendForce, Grana Research, Texas Tax Filings (Grimes County), SpaceX Q3 Earnings Call Oct 20, 2026, Tesla Q4 2025 10-K, Samsung Q3 2026 Earnings.
