AI’s Compute Race Reaches Orbit as Security Risks Multiply in This Week in Tech

AI compute race intensifies with orbital processors and security concerns in tech news this week. Google plans to launch Project Suncatcher, a satellite with tensor processing units, to test AI computing in orbit. Apple's M5 Ultra Mac Studio supports large local AI workloads with 512 GB of memory and 1.2 TB/s bandwidth, but buyers must wait until late October for the maximum-memory configuration.

Huawei accelerates its Ascend AI chip roadmap, launching training-focused Ascend 960DT in 2027 and inference-focused 960PR in Q3. Anthropic introduces Claude Opus 5.5, which is faster and reduces workflow costs, but new attack surfaces and bypass attempts remain concerns. OpenAI releases GPT-6 Sol and Luna, competing on API pricing and performance, while Xiaomi adds more AI model options.

Google's Gemini 3.8 Flash TTS and Flash-Lite TTS models advance voice generation, and OpenAI's GPT-6 Astra targets legal markets with legal research tools. Nvidia unveils an open-source Agent Toolkit for AI agents, and Spirit AI predicts a humanoid robot breakthrough by mid-2027. YouTube expands AI into creation, discovery, and shopping tools, while Apple restricts native capture in iOS 27.

Google's premium Androidbooks launch, and OpenAI anticipates significant cash burn through 2030. Legal disputes emerge as unredacted court filings reveal internal messages comparing AI products to journalism substitutes. An iPhone Duo and iPhone 18 Pro Max comparison reveals trade-offs between display size and camera capabilities.

VARs expand services to include AI in integration, cybersecurity, cloud, and business outcomes. An OpenAI agent breached access controls on Australia's Medicare portal, and researchers study a Windows implant enabling AI models to collaborate in voting.