Starmind Suppliers: Who Builds the Hardware Behind SpaceX's 1 Million AI Satellites?

The Short Answer: Musk Wants to Own the Entire Stack

Starmind — SpaceX's plan for up to 1 million AI compute satellites, filed with the FCC on January 30, 2026 — is deliberately structured to depend on as few outside companies as possible. The pattern is the classic Musk playbook: buy or build every bottleneck.

The reason is stated plainly in SpaceX's own IPO filings: the company cannot currently source enough chips to build the constellation. Musk claims all existing fabs on Earth produce only about 2% of what his companies will eventually need. His answer, in his own words: “We either build the Terafab, or we don't have the chips.”

So the supplier question has two answers — one for 2026–2027 (external hardware, modular payloads) and one for 2028 and beyond (near-total vertical integration). This article maps both.

What Actually Goes Into an AI1 Satellite

The first Starmind hardware, unveiled June 8, 2026, is called AI1. It is derived from Starlink V3 components but is a fundamentally different machine — an orbiting server rack with a 70-meter deployed wingspan (wider than a Boeing 747-8) and a 20-meter height. Two AI1 prototypes are targeting an early 2027 launch.

The single most important design decision for suppliers: the compute payload is interchangeable and hardware-agnostic. The satellite bus handles power, cooling, and connectivity; the chips inside are whatever the customer wants. That keeps NVIDIA and Google in the game even as SpaceX builds its own silicon.

The In-House Stack: Three Musk-Owned Factories

1. Gigasat (Bastrop, Texas) — satellite assembly

SpaceX is standing up a dedicated Starmind manufacturing facility called Gigasat in Bastrop, Texas, aiming to be operational by year-end 2027. Volume production follows. The precedent is real: SpaceX already builds Starlink satellites faster than any spacecraft in history, and executives describe the AI1 bus as structurally simpler than current Starlink hardware.

2. Terafab (Texas) — the $119 billion chip play

Announced March 21, 2026, Terafab is a joint venture between Tesla, SpaceX, and xAI, with Intel joining on April 7 to contribute manufacturing expertise. The plan: consolidate chip design, lithography, fabrication, memory, packaging, and testing under one roof. County filings show $55 billion for phase one and up to $119 billion total for the full-scale site in Grimes County, with a prototype line at Giga Texas in Austin.

Two details matter for Starmind specifically. First, the fab will produce a dedicated space chip — the D3, custom-designed and radiation-hardened for orbital AI satellites. Second, Musk has said roughly 80% of Terafab's compute output is earmarked for space, not ground applications. Terafab is, functionally, Starmind's chip supplier being built in advance. The long-term target: 1 million wafer starts per month and 100–200 billion chips per year on Intel's 14A (1.4 nm) process.

Skepticism is warranted. Bernstein analysts estimate that reaching Musk's 1-terawatt annual compute goal would require up to $5 trillion and 358 individual fabs, and leading-edge fabs typically take 3–5 years from groundbreaking to first wafers. Nobody involved except Intel has ever fabricated a chip.

3. Mesh Optical Technologies — the optics acquisition

On June 25, 2026, the FTC granted early termination for Musk's acquisition of Mesh Optical Technologies, a high-speed optical transceiver startup founded by former Starlink laser-communications engineers. Optical transceivers determine chip-to-chip communication speed — the exact bottleneck in a distributed orbital data center. The deal lets Musk internalize what was previously bought from Broadcom and Coherent, closing the loop on the laser mesh that ties one million satellites into a single computer.

External Suppliers: Who's in the Chain Today

Until the in-house stack is running, Starmind depends on the broader semiconductor and space supply chain. Here is the 2026 map:

One more relationship shapes the whole chain: demand. SpaceX already leases AI compute to Google and Anthropic in deals worth billions annually — including a reported $920 million monthly compute agreement with Google. Those contracts are the revenue engine meant to fund the hardware buildout, with orbital deployments starting in 2028.

Winners, Losers, and the Ones to Watch

• Clear winners (2026–2028): NVIDIA and Google — the modular payload guarantees them volume during the interim window. ASML wins regardless: Terafab cannot exist without EUV lithography.
• Structural winner: Intel. Terafab is the first major external commitment to Intel's foundry business, and its stock had its best month ever in April 2026 on the news.
• At risk: Broadcom and Coherent lose the optical module socket to Mesh. Samsung and TSMC lose future Musk volume as Terafab ramps — though both are capacity-constrained anyway.
• Wildcard: STMicroelectronics. Its decade-long Starlink partnership (5+ million chips shipped per day at peak) makes it the most likely external supplier of space-qualified support silicon for AI1 — the unglamorous parts no one else makes at volume.

Supply Chain Timeline: What Happens When

• Early 2027: two AI1 prototype satellites launch with external (NVIDIA-class) compute payloads.
• Year-end 2027: Gigasat in Bastrop targeted operational; volume satellite production begins.
• 2028: first operational orbital AI deployments; Terafab prototype line in Austin produces first D3 test silicon.
• 2029–2030: Grimes County Terafab full-scale phase; the in-house D3 begins displacing external accelerators — if the fab hits timeline, which no first-time chipmaker ever has.
• Beyond: 1 million wafer starts/month target; external chip suppliers relegated to customer-specified payloads only.

FAQ

Bottom Line

Ask “who builds Starmind's hardware” in mid-2026 and the honest answer is: NVIDIA, Google, Intel, ST, and ASML build it today — SpaceX intends to build it all tomorrow. The constellation's economics depend on that transition succeeding: Starship must slash launch costs, Gigasat must out-produce every satellite line in history, and Terafab must do what no first-time fab operator has ever done, on a $119 billion budget.

The supplier map will shift with every quarter of 2026–2028. We track each contract, fab milestone, and FCC decision across the Musk industrial stack — check back as the AI1 prototypes head to the pad in early 2027.