AI infrastructure mega-deals: Qualcomm nears $4B Modular acquisition; SpaceX signs $6.3B compute deal with Reflection: Qualcomm is in late-stage talks to acquire AI chip startup Modular for ~$4B. Separately, SpaceX will provide open-source AI startup Reflection with Nvidia GB300 chips at its Colossus 2 data center for $150M/month starting July 1, totaling up to $6.3B through 2029. AI compute is no longer a cloud line item — it is becoming a strategic and geopolitical asset. Bloomberg · CNBC
China’s Z.ai releases GLM-5.2, an open-weight model that beats GPT-5.5 on coding benchmarks — on the same day the US banned Anthropic’s Fable 5: The 744B-parameter GLM-5.2 is released under an MIT license with no regional restrictions, self-hostable, and priced at 1/6 of GPT-5.5’s API cost. For teams building outside the US, the infrastructure calculus just changed fundamentally. StartupFortune
Five Eyes agencies issue unprecedented joint warning: AI models capable of toppling governments and businesses are “months away”: Cybersecurity agencies from the US, UK, Canada, Australia, and New Zealand issued a rare joint statement on AI threats, following the US block on Anthropic’s Fable 5. AI safety is now the front line of the G2 tech rivalry. The Guardian · CISA
Sakana AI ships Fugu, a multi-agent orchestration model matching Fable 5 and GPT-5.5: Tokyo-based Sakana AI released Fugu, an orchestration model that coordinates multiple agents and achieves frontier performance without relying on a single massive model. A signal that multi-agent architectures are becoming a credible path to state-of-the-art results. Nikkei Asia
Agent-native developer tools emerge: Oak (Git for agents), Block/Buzz (human+agent workspace), PMB (MCP-based memory): A wave of tools treating AI agents as first-class users is arriving — Oak reimagines version control for agents, Block/Buzz builds collaborative workspaces for humans and agents, and PMB provides local-first memory for coding agents over MCP. Oak · Block/Buzz · PMB
Oracle cuts ~21,000 jobs amid AI adoption; GM installs 50 robots after laying off 1,300 workers: Oracle reduced its workforce by roughly 21,000 citing AI-driven transformation. GM installed 50 FANUC robot arms at its Detroit Factory Zero plant while 1,300 workers remain on what was supposed to be a temporary layoff, drawing sharp pushback from the UAW union. The pattern of “temporary layoff → automation installation” is repeating across industries. Reuters · Ars Technica
Software engineers face an AI “identity crisis bordering on depression,” Menlo Ventures partner warns: A partner at VC firm Menlo Ventures describes a growing divide and burnout among software engineers as “tokenmaxxing” culture within companies creates new hierarchies and questions the role of human developers. Business Insider
Open-weight models as compute leverage: frontier coding performance at 1/6 the cost, self-hosted, no regional restrictions: Models like GLM-5.2 make it possible to run frontier-level AI outside the US API ecosystem. Teams evaluating AI adoption should consider open-weight models as their infrastructure default, not a fallback. StartupFortune
Agent-native infrastructure as a new category — time to experiment with Oak, Block/Buzz, and PMB as reference implementations: Git, workspaces, and memory are being redesigned with AI agents as first-class users. These early open-source projects provide a concrete starting point for teams to form hypotheses about what agent workflows actually need. Oak · Block/Buzz
The Anthropic Fable 5 export block and the trajectory of US AI controls — how far will they go, and will open models fill the gap?: The US blocked foreign access to Anthropic’s Fable 5, immediately followed by the Five Eyes warning. The key question: how far will frontier model export controls extend, and will Chinese and Japanese open models systematically fill the resulting gap? The Guardian
AI super PACs have spent $43.3M on US midterm races — is AI policy being decided by campaign cash?: Groups tied to the AI industry are flooding congressional races with money to shape future regulation. With bipartisan legislative efforts stalled, well-funded industry interests may effectively pre-determine who sits at the policy table. NPR