In a public filing with Grimes County, Texas, SpaceX revealed plans for a "multi-phase, next-generation, vertically integrated semiconductor manufacturing and advanced computing fabrication facility." The proposed Terafab would be located near the Gibbons Creek Reservoir, approximately 80 miles northeast of Houston [citation:1][citation:5].

The initial phase is estimated at $55 billion — roughly equivalent to 1.25 Twitters, as The Register colorfully noted. If all phases are completed, total investment could reach $119 billion, making it one of the largest private industrial projects in U.S. history [citation:5]. For context, Intel's leading-edge fab expansion in Chandler, Arizona, cost approximately $30 billion. Musk's ambition is an order of magnitude larger [citation:5].

Before diving into the memory supercycle, it's worth acknowledging the chasm between Musk's ambitions and his track record. "Musk has never built a wafer fab, but he wants to burn $119B on one anyway," The Register noted dryly [citation:5]. SpaceX knows launch vehicles and satellites. xAI knows model training. Twitter knows social media. None of Musk's companies have semiconductor manufacturing experience [citation:5].

But there are pieces in place:

• Intel partnership: Tesla plans to spend roughly $3 billion on a research chip facility in Texas, with Intel contributing chip design, fabrication, and packaging expertise [citation:3].
• Equipment discussions: Musk's lieutenants have already reached out to chip equipment makers including Applied Materials, Tokyo Electron, and Lam Research for pricing and delivery timelines [citation:3].
• 14A node: The fab would use Intel's yet-unfinished 14A process node [citation:5].

At minimum, Terafab represents a statement of intent: Musk believes that computing power is the single greatest bottleneck to AI advancement, and he is willing to spend nine figures to fix it.

While Musk tries to build chips, established memory manufacturers are printing money. May 6, 2026, was a historic day for semiconductor stocks [citation:2].

Price Rally Highlights (May 6, 2026):

• SK Hynix: +10.64% (all-time high)
• Samsung Electronics: +14.41% (all-time high)
• 江波龙 (Jiangbolong): +10% (all-time high)
• 兆易创新 (GigaDevice): +10% (all-time high)
• Micron Technology: +16.3% year-to-date [citation:6]

The rally was driven by extraordinary earnings reports across the sector, validating the "supercycle" thesis that began taking shape in late 2025 [citation:2].

The primary driver of the current memory crisis isn't a drop in total manufacturing capacity — it's a strategic reallocation of that capacity toward AI data centers [citation:4].

• 70% of global DRAM production is now consumed by data centers
• Each HBM (High Bandwidth Memory) unit displaces 2+ conventional DRAM wafers due to complexity [citation:4]
• 2028 DRAM capacity already booked
• SK Hynix: Entire 2026 DRAM, HBM, and NAND production sold out [citation:6]
• Samsung: HBM4 capacity fully sold, all advanced capacity reserved [citation:2]

Forecasts for 2026 are staggering:

• DRAM average price: Estimated +88% (花旗) [citation:2]
• NAND flash average price: Estimated +74% (花旗) [citation:2]
• Q2 2026 contract prices: DRAM +58-63%, NAND +70-75% (TrendForce) [citation:2]

Counterpoint Research expects DRAM prices to rise another 40% through Q2 2026 on top of 2025's increases [citation:6].

The memory squeeze has real-world consequences for consumers [citation:4]:

• PCs & Smartphones: Memory chips historically made up ~10% of materials cost. That proportion has now surged to 20-30% [citation:4].
• HP, Dell, Lenovo, Asus, Acer have already raised prices or are planning further hikes [citation:4].
• Smartphone output: TrendForce projects global smartphone unit output will drop by at least 10% year-over-year in 2026 [citation:4].
• 128GB phones disappearing: Brands are phasing out low-capacity models, making 256GB the new baseline [citation:4].

• Pricing power shift: Samsung has raised memory chip prices by as much as 60% since September, capitalizing on tight supply and panic ordering [citation:6].
• AI memory allocation: Even CEOs visiting suppliers in person are being told "there's simply no stock available" for certain components [citation:4].