The Friday File: Amazon Globalstar; Anthropic; Orbital

Mobile World Live brings you our top three picks of the week as Amazon made an $11.6 billion play for Globalstar, US officials sought access to Anthropic's latest restricted AI model and startup Orbital set its sights on space-based AI data centres.

Amazon agrees $11.6B deal to acquire Globalstar

What happened: Amazon agreed an $11.6 billion deal to acquire Globalstar, giving it control of the satellite operator's fleet, ground infrastructure and MSS spectrum licences.

Why it matters: The proposed buyout equips Amazon's satellite business Amazon Leo with the assets needed to add direct-to-device (D2D) services. Beginning in 2028, the company plans to launch its own D2D offering touting "substantially higher spectrum use and efficiency" than legacy platforms. The tech giant said the acquisition will support consumer, enterprise and government connectivity services, while bolstering its long-term target of deploying 3,200 satellites by 2029. In parallel, the tech giant inked a deal with long-term Globalstar partner Apple to power satellite services on current and future iPhone and Apple Watch devices. CCS Insight analyst Luke Pearce noted that "despite its delayed start...Amazon is emerging as Starlink's most significant challenger", arguing the company is now "armed with its own MSS spectrum and an exclusive Apple relationship". Meanwhile, Tim Hatt, head of research and consulting at GSMA Intelligence, noted the hefty price tag on the purchase "speaks to Amazon's desire to move to market fast", adding the deal indicates "the centre of gravity is clearly shifting toward larger constellation providers with infrastructure scale".

US agency pursues access to new Anthropic AI model

What happened: The US Treasury Department is reportedly seeking access to Anthropic's powerful Mythos AI model, which has been restricted under its Project Glasswing programme over fears it could be weaponised by hackers.

Why it matters: The White House is preparing to make a version of Anthropic's Mythos model available to major federal agencies to test government software for vulnerabilities, with safeguards under development to manage risks. The move marks a sharp pivot by the US government, which in March designated Anthropic a supply chain risk after it declined to grant the Pentagon unfettered deployment rights, citing surveillance and autonomous weapons concerns. According to Anthropic, Claude Mythos, currently limited to a small group of companies including Amazon Web Services, Apple, Nvidia and Google, "is capable of autonomously identifying and exploiting software vulnerabilities at a level which surpasses most human experts", warning the capability could extend "potentially beyond actors who are committed to deploying them safely". During testing, the model uncovered thousands of zero-day flaws across every major operating system and browsers. The company has since launched Claude Opus 4.7, a less capable model with enhanced safeguards to block high risk cybersecurity use, now broadly available across its platforms and cloud partners.

Orbital starts countdown to space data centre test

What happened: California-based startup Orbital set out plans to launch its first in-orbit test mission in 2027 as a foundation to eventually put AI data centre infrastructure in low Earth orbit (LEO).

Why it matters: The company's first satellite, which is scheduled to launch on a SpaceX Falcon 9 in April next year, is designed to validate sustained GPU operations in space and support early commercial AI workloads. The startup is also preparing an application with the US Federal Communications Commission (FCC) to deploy a broader AI compute constellation. CEO Euwyn Poon said AI progress is "being constrained by the grid", adding data centre "economics are dominated by electricity and cooling, and both are getting harder". Elon Musk's SpaceX has also made noises about orbital data centres, filing a plan with the FCC in February to deploy up to 1 million solar powered satellites to "accommodate the explosive growth of data demands driven by AI". Orbital similarly positioned space as a route to remove the sector's "energy ceiling". Yet practical constraints remain around satellite power capability, spectrum coordination and hardware resilience in orbit.