“Amazon Buys Globalstar, Delta to Add Leo, The Apple Angle” — @BenThompson (Apr 15 2026)
Why this is in the vault
Satellite/telco business analysis, which on its face is not RDCO-adjacent. But the load-bearing idea in the piece is a general strategic pattern: leading companies will pay a premium to avoid being locked into a dominant supplier’s terms. Apple doesn’t want to depend on SpaceX/Starlink; Delta doesn’t want Starlink’s whole-fleet-only terms; Verizon is doubling down on AST SpaceMobile instead of taking Starlink. Thompson reads Amazon’s $11.6B Globalstar acquisition as Apple-engineered: Apple couldn’t accept SpaceX terms, Globalstar was failing, so Amazon was enlisted to buy Globalstar and play ball on Apple’s terms. That “paying a premium to maintain leverage over your supplier” dynamic is the same pattern the vault tracks in the AI stack: who controls the state, who controls the harness, who has the veto.
The core argument
- Framing Amazon’s Globalstar acquisition as Amazon vs SpaceX misses the real story.
- Globalstar’s assets are middling: aging 24-satellite constellation, ~25 MHz of spectrum (vs SpaceX’s 65 MHz), bent-pipe architecture.
- Globalstar’s only significant customer was Apple (Apple had 20% equity + rights to 85% of network capacity).
- Thompson’s thesis: Apple engineered this deal. Apple wouldn’t partner with SpaceX because of who-is-in-charge conflicts; Globalstar was failing; Amazon acquiring Globalstar gives Apple a supplier who will take Apple’s terms. Amazon gets a cellular side business; Apple gets to avoid being at SpaceX’s mercy.
- Same dynamic shows up across the industry: Delta chose Leo over Starlink because Starlink demanded whole-fleet + free-to-customer terms; United (#2 airline trying to catch Delta) accepted those terms. Verizon is investing in AST SpaceMobile for the same reason.
Quote (≤15 words): “ensuring that the iPhone maker isn’t answering to Elon Musk.”
Mapping against Ray Data Co
This is a strategic-positioning piece, not a technical one, but the pattern maps directly onto the AI stack question that drives RDCO’s thesis:
- SpaceX is in the Anthropic/OpenAI position. Best service, steep terms, demands integration on their terms. Smart enterprises will pay a premium to maintain leverage.
- Apple’s move maps to the state-ownership thesis. Apple would rather pay Amazon (AWS spend + partnership premium) than surrender control over the iPhone’s satellite UX to SpaceX. That’s the same logic as ../04-tooling/rdco-state-ownership-architecture — the client owns the state/vault/skills so the model is commoditized; enterprises don’t want to be at the mercy of whichever AI provider dominates.
- “Who is in charge” is the core unresolved question in vendor partnerships. Thompson explicitly names this as why Apple+SpaceX didn’t happen. Same question is the unresolved friction in enterprise AI adoption: does the AI provider define the integration, or does the enterprise? The agent-deployer role (2026-04-14-levie-agent-deployer-role-jd) exists precisely to make that answer be “the enterprise.”
- Delta-over-Starlink pattern applies to RDCO’s consulting posture. Premium consulting shops will want to work with a framework-and-skills-provider who takes their terms, not the other way around. That’s an argument for why RDCO’s open-source-skills + client-owned-state positioning is more compelling than a “we’ll run everything for you” managed-service pitch.
The piece doesn’t mention AI at all. But the strategic pattern (leaders paying to preserve optionality) is the cleanest articulation I’ve seen of why the state-as-moat / thin-harness thesis will get traction at the enterprise level — not because of technical superiority, but because decision-makers have seen this movie before and know how it ends when you surrender control to the dominant supplier.
Related
- ../04-tooling/rdco-state-ownership-architecture — same pattern applied to AI state
- 2026-04-13-jaya-gupta-ai-lock-in-state-moat — state as the lock-in layer
- 2026-04-14-levie-agent-deployer-role-jd — the role that ensures the enterprise, not the vendor, is in charge
- 2026-04-13-moura-entangled-software-agent-harnesses-dead — Moura’s dissent on harness thesis
- synthesis-harness-thesis-dissent-2026-04-12 — the dissent cross-check
- 2026-04-14-stratechery-openai-memos-anthropic — yesterday’s Stratechery on OpenAI/Anthropic competitive dynamics