Austin, Texas, March 23, 2026, 03:57 CDT
Elon Musk is moving forward with plans to have SpaceX and Tesla construct two advanced chip plants in Austin, shifting his concerns about AI chip shortages into action. “We either build the Terafab or we don’t have the chips,” Musk said in a presentation. One facility is set to supply Tesla’s vehicles and Optimus robots, while the other will serve AI data centers in orbit. 1
That’s become crucial as the AI race shifts beyond splashy new models, straight into the guts of the business: chips, plants, electricity, cooling systems. Bridgewater Associates figures Alphabet, Amazon, Meta, and Microsoft are set to pour around $650 billion into AI infrastructure this year. “Compute demand continues to significantly outpace supply,” wrote co-chief investment officer Greg Jensen. 2
Musk’s latest move comes just days after Nvidia announced that Amazon Web Services will purchase 1 million Nvidia GPUs by 2027—a clear signal the major cloud players are tying up hardware supply far ahead of time. Ian Buck, Nvidia’s vice president, told Reuters that inference—the phase where AI systems spit out responses for users—is “wickedly hard.” AWS will tackle that using a lineup of seven different Nvidia chips. 3
OpenAI and SoftBank are putting their money into infrastructure, not just hardware. Back in January, they announced a $1 billion investment in SB Energy for Stargate—a $500 billion data-center project supported by Oracle—as tightening power supplies start to bottleneck AI expansion. Oracle, for its part, disclosed this month that it took on significant debt and aims to raise as much as $50 billion more to boost its buildout. 4
Meta’s locking in external AI computing power, grabbing what it can before options run out. According to Reuters, Meta signed a deal to purchase $12 billion worth of AI capacity from Nebius through 2027, and can tack on another $15 billion across five more years. The move underscores how GPUs and energy are now getting snapped up in advance, just as cloud storage was in earlier cycles. 5
Musk hasn’t committed to swapping out his current suppliers for Terafab just yet. Last week, he made it clear that both Tesla and SpaceX AI will stick with large-scale Nvidia chip orders. As for Tesla’s next-gen AI6 chip, Musk said the company might “tape out” the design in December—a milestone meaning the design is locked and ready to send to the factory. Samsung is expected to handle manufacturing, using its 2-nanometer process in the back half of 2027. 6
Manufacturing’s supply chain squeeze is getting tighter. Just last week, Samsung and AMD inked a deal aimed at boosting collaboration in AI memory and looking into foundry projects. That happened in the same stretch Nvidia’s Jensen Huang revealed a fresh foundry tie-up with Samsung, Reuters reported, highlighting a shift in competition: chip design still matters, but now it’s all about who can actually churn out enough chips. 7
Terafab lines up with Musk’s February move, when he merged xAI into SpaceX—valuing the combined entity at $1.25 trillion. Reuters reported that the deal brought together his AI and satellite operations. Now, the chip initiative looks set to extend that integration, handing the group a stronger manufacturing foothold. 8
But fabs are notoriously slow and costly undertakings, and Musk hasn’t committed to a timeline. This week, Morgan Stanley analysts led by Andrew Percoco estimated Tesla could be looking at $35 billion to $45 billion just to build and equip a plant. Even with a fast-track retrofit, they don’t see chips rolling out before mid-2028. The analysts described the project as a “Herculean task.” 9
The second fab, targeting orbital computing, could be the tougher nut to crack. Reuters noted back in February that AI data centers in space have their own headaches: radiation, cooling, latency, upkeep. “There’s some real challenges here,” said Armand Musey of Summit Ridge Group. Still, David Ariosto, who tracks the sector, insisted, “compute in space isn’t sci-fi anymore.” 10
So, Terafab looks more like a signpost than a locked-in roadmap. Just this month, Oracle and OpenAI scrapped plans to grow a Texas Stargate facility after financing negotiations stalled and OpenAI’s priorities shifted—spotlighting how even high-profile, flush AI infrastructure efforts can still bump up against hard stops on capital, schedules and actual need. 11