There’s a Blinking Warning Sign for the Data Centers in Space Industry

Key number: Elon Musk proposes 1,000,000 space-based data-center satellites; Reuters-cited estimates put barebones cost at ≥$1 trillion (Eric Berger) and 'trillions' per Moffett Nathanson. Moffett Nathanson calculates SpaceX would need ~3,000 Starship launches/year (~8/day) versus 167 total launches last year, while Starship is years behind schedule with 11 test flights failing to reach orbit. Experts warn modular, non-upgradable units and weak client demand (echoing Microsoft's Project Natick) make large-scale orbital data centers economically unviable, limiting them to niche military or space-station use rather than replacing ground facilities.

Analysis

The core investment implication is that upgradeability and time-to-market will trump exotic deployment venues for compute when hardware evolves on 12–36 month cadences. That dynamic increases the marginal value of onshore footprint flexibility — colo space, modular rack-to-rack swaps, and rapid PPA-backed power — and compresses returns on any asset that is “locked‑for‑life.” Expect a multi-year premium for assets that minimize hardware obsolescence risk and for vendors that sell fast refresh cycles.

A second‑order supply‑chain effect: capital and skilled production capacity devoted to high-profile orbital ambitions crowds out components and launch manifest slots, transferring short‑term pricing power to incumbents in transformers, precision cryogenics, high‑end PCBs and satellite avionics. In the next 6–24 months that can mean 10–20% cost inflation on critical datacenter build items and longer delivery tails, accelerating hyperscalers’ preference to expand existing ground campuses rather than chase novelty.

Strategically, the only plausible durable niche for off‑earth compute is mission‑critical, low-latency space/defense applications where upgrade cycles are controlled by a single prime — not commercial AI training. That supports defense primes and integrated suppliers over speculative small-cap satellite OEMs. The binary catalysts to reprice the entire debate are (1) a major launch vehicle achieving sustained, cheap cadence and (2) an architectural break in AI that makes current accelerators obsolete — the former would take years to normalize, the latter could arrive in 18–36 months and would strand capacity faster than capex can be redeployed.

Tactically, capital should favor providers of flexible on‑shore capacity and the supply chain that enables fast upgrades, while keeping option‑sized short exposure to equity stories that assume mass orbital deployment. Maintain sizing discipline: this is a regime shift played across multiple years, not a 90‑day trade.

Key Decisions for Investors

• Buy Digital Realty (DLR) — 12–24 month horizon. Rationale: direct play on onshore, upgradeable capacity; set initial position size 2–4% NAV, target +30% if AI-driven colo buildout persists, stop-loss at -12% to preserve capital. Reward skew ~2.5:1.
• Buy Equinix (EQIX) — 9–18 month horizon. Rationale: pricing power for fast‑turn colo and peering; position 1.5–3% NAV, target +20–25%, tactical take‑profit if REIT cap rates re‑compress. Lower beta than small‑cap infra names.
• Hedge MSFT exposure via a 12–18 month put spread (buy protective puts while selling lower strike puts to finance). Rationale: MSFT is exposed to concentrated capex cycles for AI infra and could underperform if on‑prem/colo winners reprice cloud economics. Max loss = net premium; payoff if MSFT falls >15–20% provides ~3:1 downside protection vs cost.
• Initiate a small, option‑sized short on AST SpaceMobile (ASTS) or similarly valued small‑cap satellite OEMs — 6–18 month horizon. Rationale: valuation assumes large commercial TAM for orbital compute that is unlikely; cap size <=1% NAV due to binary risk, tighten stops at 50% loss given volatility.