SpaceX plans to launch its first NVIDIA-chip-powered AI satellites, named Starmind AI1, in the fourth quarter of 2027, with large-scale deployment following in 2028CITE:E1. Musk stated the schedule is at least one year earlier than the project's original planCITE:E1. On August 24, Musk said on social platform X that SpaceX and NVIDIA have jointly designed a Vera Rubin NVL72 system optimized for the space environment, which is scheduled to launch into orbit in the fourth quarter of next year and scale up to large-scale deployment in 2028CITE:E4CITE:E7.
Musk said the satellite-based AI data center will be far simpler, cheaper, denser, and lighter than traditional rack-based data centersCITE:E2. He described the design as being able to achieve lower cost and higher density while reducing weight compared with ground-based infrastructureCITE:E2. The Vera Rubin NVL72 system that SpaceX and NVIDIA co-designed for the space environment is set to enter orbit in the fourth quarter of next year and expand to large-scale deployment in 2028CITE:E7.
SpaceX has filed with the U.S. Federal Communications Commission (FCC) to deploy a network of up to 1 million satellites for running complex AI computationsCITE:E3. Research firm Wood Mackenzie estimates that a 1-gigawatt (GW) orbital data center would cost $170 billion, more than three times the cost of a comparable traditional data centerCITE:E8. Musk said SpaceX plans to raise its AI compute capacity to more than 2 GW by the end of 2026, and to reach a scale "closer to 10 GW than 5 GW" by the end of next yearCITE:E9.
| Metric | Value |
|---|---|
| FCC satellite network filing | Up to 1,000,000 satellites |
| Orbital data center cost (Wood Mackenzie estimate) | $170 billion per 1 GW |
| Cost multiple vs. traditional data center | More than 3x |
| SpaceX AI compute target, end of 2026 | Over 2 GW |
| SpaceX AI compute target, end of 2027 | Closer to 10 GW than 5 GW |
Unitech (華通) projects its satellite business will grow to 21% of revenue in 2026 and 24% in 2027CITE:E5. Sesoda (事欣科) said low-earth-orbit satellite-related business will see explosive growth, with this year's LEO satellite orders accounting for 18%-20% of its total order book, and growth next year expected to at least double, with a multi-fold increase not ruled outCITE:E6.
| Company | Metric | Value |
|---|---|---|
| Unitech (華通) | Satellite business share of revenue, 2026 | 21% |
| Unitech (華通) | Satellite business share of revenue, 2027 | 24% |
| Sesoda (事欣科) | LEO satellite share of orders, this year | 18%-20% |
| Sesoda (事欣科) | Expected order growth next year | At least doubling, multi-fold not ruled out |
SpaceX and Tesla plan to jointly invest at least $16.8 billion in the initial phase of the Terafab chip manufacturing project in Grimes County, TexasCITE:E10.
What this means: SpaceX's own compute targets — over 2 GW by the end of 2026 and near 10 GW by the end of 2027 — sit against Wood Mackenzie's estimate that each 1 GW of orbital data center capacity costs $170 billion, more than three times a terrestrial equivalentCITE:E8CITE:E9. The compressed Starmind AI1 timeline, moved up by at least a year to a fourth-quarter-2027 launch and 2028 deployment, runs in parallel with SpaceX and Tesla's $16.8 billion Terafab chip investment in Texas, and with rising satellite-related revenue and order shares reported by Unitech and Sesoda in TaiwanCITE:E1CITE:E10CITE:E5CITE:E6.
SpaceX plans to launch its first NVIDIA-chip-powered AI satellites, named Starmind AI1, in the fourth quarter of 2027, with large-scale deployment following in …
Musk said the satellite-based AI data center will be far simpler, cheaper, denser, and lighter than traditional rack-based data centersCITE:E2.
SpaceX has filed with the U.S. Federal Communications Commission (FCC) to deploy a network of up to 1 million satellites for running complex AI computationsCITE…
Unitech (華通) projects its satellite business will grow to 21% of revenue in 2026 and 24% in 2027CITE:E5.
The clearest figure in this story is the cost gap: Wood Mackenzie puts a 1-gigawatt orbital data center at $170 billion, more than three times a terrestrial equivalent, yet SpaceX is still targeting over 2 GW of AI compute by the end of 2026 and near 10 GW by the end of 2027. That gap is the real variable to watch — whether the Starmind AI1 timeline, already pulled forward by a year to a Q4 2027 launch and 2028 deployment, holds as a sign of engineering readiness or comes under pressure from that cost multiple. On the supply side, Unitech's projected climb to 24% satellite revenue share by 2027 and Sesoda's forecast of at least doubling LEO order share are concrete, quarter-trackable proxies for whether SpaceX's launch cadence is actually converting into hardware orders rather than staying on the roadmap.
NVIDIA announced the Jetson Orin Nano 2 robotics computer on August 25, 2026, packing 78 TOPS of AI compute, 8GB of memory, and an 8-core Arm CPU while doubling inference performance and cutting power draw 40% versus its predecessor. Modules and developer kits ship in the first half of 2027, targeting a robotics developer base NVIDIA says already tops 3 million.
NVIDIA confirmed on August 24, 2026 that its 256-chip Groq 3 LPX inference rack has entered full production, with Nebius named as the first cloud provider set to deploy it later this year alongside Vera CPU and Rubin GPU.
HanTest (漢測) reported NT$2.668 billion in cumulative revenue for January–July 2026, up 128.18% year-on-year and already exceeding its entire 2025 revenue of NT$2.425 billion, as the company prepares for a late-September 2026 Taipei Exchange listing built on expansion into memory testing, advanced packaging, and silicon photonics.