Year One of Space Compute: Five Hot Spots and an Investment Map for 2026 LEO
In March 2026, Jensen Huang said something on the GTC stage:
"Space computing, the ultimate frontier, has arrived."
This was not a metaphor. In November 2025, US startup Starcloud launched an NVIDIA H100 into orbit (Starcloud-1) and completed the first large language model training session in space. Two months later, on January 11, 2026, Axiom Space launched two orbital data center nodes (ODC Node 1 & 2), connected them to Kepler Communications' optical relay network, and began offering cloud compute services to external customers.
Earth orbit has officially become a new location class for data centers.
Five Ignition Points in 2026
1. Orbital Data Centers (ODC) Enter Operational Phase
The reason this is happening now, not ten years from now, is that three trends converged simultaneously:
Ground power consumption has hit its ceiling. Global data center power usage was 415 TWh in 2024; 2026 is projected to exceed 650 TWh. Taiwan's power grid is a microcosm — AI compute is exploding, and where the power comes from is every IDC operator's top question.
Space has free energy. In orbit there are no clouds, no rain, no day-night cycles — 7×24 hours of solar power, at far higher generation efficiency than ground-based panels. Thermal management? Simpler still: directly radiate to the cosmic background, no compressors or cooling towers required. Jeff Bezos's judgment is that within decades, ODC construction costs will be lower than ground-based data centers.
Commercial GPUs in orbit validated as feasible. This is the most critical step. Starcloud-1 didn't use any specially hardened space chip — it used the same NVIDIA H100 you'd find in a ground server rack. COTS hardware operating stably under orbital radiation conditions means the entire ground-side GPU ecosystem can be directly reused.
The combined conclusion: ODC is no longer science fiction. 2026–2030 is the first capex wave.
2. Spectrum and Orbital Slots Entering Scarcity Crisis
LEO's "real estate" is not unlimited. Per ITU spectrum allocation memos from late 2025 to early 2026, the 500–1,200 km commercially valuable orbital inclinations and the Ku/Ka/V bands are approaching exhaustion.
This means:
- SpaceX (Starlink), OneWeb, and Amazon Kuiper have already filed first and locked in their positions; latecomers have very limited opportunity in traditional comms satellite themes.
- ITU rules are "first to file, first to claim" — this resource-grab was effectively settled in 2020–2023.
- Latecomers only have differentiated space left via "novel satellite types" like ODC — because ODC's business model depends on compute supply scarcity, not spectrum scarcity.
3. China's Three-Body Computing Constellation — ADA Space
In May 2025, China's ADA Space launched its first batch of 12 satellites. Each carries:
- 100 Gbps optical links (inter-satellite communications)
- 744 TOPS on-orbit accelerator (AI inference)
Target: a 2,800-satellite distributed orbital supercomputing network.
This is not a startup venture — it is national-level strategic scale-up. AI, sensing, and communications unified — orbital compute has become a new strategic resource, on par with spectrum and rare earth minerals.
By comparison, the US side has Starcloud (commercial innovation) + Axiom Space (ODC nodes) + Google Suncatcher (TPUs in orbit, connected via optical comms) — taking a market-driven approach. Both sides are racing to see whose on-orbit compute reaches scale first.
4. Amazon Kuiper Commercial Launch Countdown
Amazon's Project Kuiper has launched more than 150 satellites toward its 3,236-satellite constellation goal. Compared to Starlink, Kuiper is a "late mover," but Amazon's capital firepower and AWS ground cloud integration advantage should not be underestimated.
Taiwan's activity is worth watching: Far EasTone Telecommunications plans to bring Kuiper services into the Taiwan market through a partnership with Amazon. This is the first clear signal of a non-Starlink LEO service entering the Taiwan ecosystem.
5. Starlink V3 Upgrade × Surging Power Requirements
Starlink's second-generation satellites are being upgraded to V3 specs:
- Significantly higher proportion of laser inter-satellite links (ISL) — no longer fully dependent on ground station relay; satellites communicate at light speed directly
- Per-satellite power demand surging from 2 kW to 10 kW or more
The supply chain implications: technical thresholds for RF components and thermal management keep rising, low-spec suppliers get weeded out, and the moat for high-spec vendors deepens further.
Full Upstream-Midstream-Downstream Breakdown
Understanding the LEO industry requires understanding that it is not a single pipeline but five structurally distinct segments.
Upstream: Core Components (Taiwan's Home Turf)
| Category | Representative Companies | Taiwan Position |
|---|---|---|
| RF power amplifiers (GaAs PA) | WIN Semiconductors (3105) | ★★★★★ Global leader |
| Filters / waveguides | Auden Technology (3491) | ★★★★★ >50% gross margin, in Starlink & Kuiper supply chain |
| LEO-specific PCB | Tripod Technology | ★★★★★ ~80% global share |
| High-efficiency solar cells | (Multi-player competition, no Taiwanese leader) | — |
| Radiation-tolerant compute chips | (No Taiwanese leader) | — |
Taiwan's upstream position is genuinely world-class. WIN Semiconductors' GaAs PA is essential for LEO high-frequency communications; Auden Technology has broken into both major global LEO constellation supply chains; Tripod's LEO-specific PCB market share approaches monopoly level (Q4 2025 LEO board revenue exceeded NT$15.5 billion).
Mid-stream A: Satellite Platform and System Integration
Satellite bus production is dominated by constellation operators (SpaceX largely self-manufactures), with Taiwanese vendors' entry points being board-level and structural components.
Important sub-topic: laser inter-satellite links (ISL). Starlink V3 has increased its ISL fraction, allowing satellites to communicate directly at light speed without routing through ground stations each time. ISL requires high-precision Pointing, Acquisition, and Tracking (PAT) systems — a precision optics domain currently dominated by European and American companies like Mynaric and SA Photonics, with near-zero Taiwanese participation.
Mid-stream B: Launch Services and In-Orbit Lifecycle
Global LEO-related investment exceeded 25 billion in 2024). Much of this flows toward launches. SpaceX Falcon 9 and Starship's continued cost reduction is the key leverage for the entire lifecycle commercial viability — without cheap launch, Bezos's ODC math simply doesn't work.
Mid-stream C: Space Data Centers (Hottest New Theme, Taiwan's Biggest Gap)
This is the most important new segment across the entire LEO industry in 2026, and also the area of Taiwan's most glaring absence.
Mid-stream C requires integrating:
- Compute hardware (COTS GPU vs. space-hardened engineering trade-offs)
- Thermal management (convection-free environment; only radiative cooling + heatspreaders)
- Power management (10 kW+ solar arrays → compute modules)
- Optical communications (high-speed downlink to ground and inter-node)
- System integration capability (bundling launcher, ground station, and optical comms together)
Taiwanese companies are at "can build it, but not to space grade" for the first three; the fourth and fifth have almost no accumulated expertise.
Downstream A: Ground Infrastructure
Ground stations (gateway stations), user terminals (antennas, LNBs, routers), hybrid ground-space architecture.
Ground-space hybrid cloud is the most realistic first landing point for ODC commercialization: ODC nodes handle inference / edge compute, ground DCs handle training / storage, connected via high-speed downlink. Enterprises don't need to fully migrate from ground; they can adopt incrementally.
Taiwan has decent participation in the ground terminal segment (antennas, LNBs, power supplies) (★★★★).
Downstream B: End Services
| Scenario | Notes |
|---|---|
| Direct-to-device (D2D/NTN) | T-Mobile × SpaceX; AT&T/Verizon × AST SpaceMobile |
| Earth observation AI | On-orbit real-time satellite imagery processing; wildfire detection, agricultural monitoring, vessel identification |
| Defense / sovereign cloud | Geography-independent, physically indestructible, global coverage; earliest paying use case |
| Aviation / maritime | Starlink Aviation in-flight Wi-Fi; already a mature market |
Taiwan's Opportunities and Risks
Taiwan's LEO supply chain position can be summarized in one sentence: upstream king, mid-stream C absent.
| Segment | Presence |
|---|---|
| Upstream RF / PCB | ★★★★★ |
| Mid-stream A (board integration) | ★★★ |
| Mid-stream A (ISL optics) | ★★ |
| Mid-stream C (ODC hardware integration) | ★ |
| Downstream ground segment | ★★★★ |
The upstream high margins are real, but they rest on a "selling components" business model. If ODC matures and the value-add in compute services concentrates in mid-stream C and downstream service layers, upstream component vendors could gradually be compressed into pure OEM/ODM mode — a replay of Taiwan's PC contract manufacturing history.
2026–2028 is the first commercial wave window for mid-stream C.
Three possible paths:
- Path A: Upstream vendors like WIN Semiconductors and Auden Technology extend into mid-stream C. Technically closest, but culturally the hardest transformation (from "selling components" to "selling subsystems").
- Path B: PCB makers like Tripod and Unimicron raise their board-level integration stack, extending into ODC compute board assemblies. Most likely to produce a dark horse, because PCB makers are already accustomed to system-level integration.
- Path C: Defense / sovereign cloud procurement as catalyst. If Taiwan includes ODC in defense procurement, it will force a mid-stream C vendor to emerge — the most certain catalyst, with a timeline depending on policy will.
Timeline
| Date | Event |
|---|---|
| 2025-05 | ADA Space launches first batch of 12 satellites (744 TOPS/unit) |
| 2025-11 | Starcloud-1 launch, H100 in orbit, first successful space LLM training |
| Late 2025 – Early 2026 | ITU spectrum allocation memos: 500–1200 km + Ku/Ka/V bands reaching exhaustion |
| 2026-01-11 | Axiom Space ODC Node 1 & 2 launch; Kepler optical relay connected |
| 2026 GTC | Jensen Huang: "Space computing, the ultimate frontier, has arrived" |
| 2026–2028 | First commercial wave for mid-stream C |
| 2030 estimate | McKinsey: agentic commerce (including space compute) intermediates $3–5 trillion |
Conclusion
The LEO industry in 2026 ignited on five fronts simultaneously: ODC validated as feasible, spectrum scarcity becoming strategic, China's Three-Body constellation scaling up, Amazon Kuiper going commercial, and Starlink V3 upgrading. This is not "a story for the next five years" — it is capex deployment happening right now.
Taiwan's upstream RF and PCB vendors are among the most certain beneficiaries of this cycle. But the real long-game question is: can Taiwan produce a mid-stream C player within the 2026–2028 window — before the value-add of space compute fully migrates to the service layer?
Deep analysis: Taiwan LEO Structural Gap and Three Breakout Paths