The Friday File: Amazon Globalstar; Anthropic; Orbital

Why It Matters

Amazon’s purchase accelerates its satellite broadband ambitions, potentially reshaping the competitive landscape against SpaceX. Government scrutiny of powerful AI models and the push for orbital AI compute reflect emerging security and infrastructure challenges in the digital economy.

Key Takeaways

• Amazon's $11.6B Globalstar purchase adds MSS spectrum and Apple link
• Anthropic's Mythos model can autonomously discover zero‑day software flaws
• Orbital aims to validate GPU workloads on a LEO satellite by 2027
• Space‑based AI data centres could bypass terrestrial power and cooling limits

Pulse Analysis

Amazon’s $11.6 billion acquisition of Globalstar marks a decisive step toward a vertically integrated satellite broadband business. By securing MSS spectrum licences and an exclusive Apple relationship, Amazon Leo can offer direct‑to‑device connectivity that promises higher spectrum efficiency than legacy constellations. Analysts see the move as a fast‑track to challenge SpaceX’s Starlink, especially as Amazon targets a 3,200‑satellite fleet by 2029. The deal also underscores the growing importance of spectrum assets in the race to dominate global connectivity.

In Washington, the Treasury’s request for Anthropic’s Mythos model highlights the tension between AI innovation and national security. Mythos, capable of autonomously identifying software vulnerabilities, has demonstrated the ability to uncover thousands of zero‑day flaws across major operating systems. The White House’s plan to provide a controlled version to federal agencies signals a shift toward proactive cyber‑risk mitigation, while Anthropic counters with a less powerful Claude Opus 4.7 model that incorporates stricter safeguards. This episode illustrates how regulators are grappling with powerful generative AI tools that could be weaponised if left unchecked.

Orbital’s upcoming in‑orbit test reflects a broader industry push to relocate AI compute to space, where electricity and cooling constraints are less severe. By validating sustained GPU performance on a low‑Earth‑orbit platform, the startup aims to lay the foundation for a constellation of AI data centres that could serve latency‑sensitive workloads. However, practical hurdles—such as power generation, spectrum coordination and hardware durability—remain significant. If Orbital and similar ventures succeed, they could redefine the economics of AI infrastructure, offering a new frontier for companies seeking to escape the terrestrial “energy ceiling.”