World model startup Odyssey reaches $1.45B valuation, backed by Amazon

Key Shifts

World model startup Odyssey reaches $1.45B valuation, backed by Amazon: Odyssey has raised a major round from Amazon, GV, NEA, and others, cementing itself as a leading player in world models — AI systems that simulate the physical world. This represents the next frontier beyond LLMs, with applications spanning robotics, manufacturing, autonomous driving, and gaming. For founders building in simulation-heavy industries, world models are becoming the new infrastructure layer. TechCrunch
Microsoft considers using DeepSeek V4 to power Copilot Cowork: Microsoft is exploring hosting Chinese lab DeepSeek’s V4 model as a cheaper option for its Copilot Cowork enterprise AI tool. The move creates a paradoxical landscape: while the US government restricts foreign access to Anthropic’s frontier models on national security grounds, Microsoft is simultaneously evaluating a Chinese model for its enterprise product. The geopolitical fragmentation of the AI supply chain is accelerating, and it is becoming harder to predict where regulatory lines will settle. The Information

Pramaana Labs raises $27M seed led by Khosla Ventures to bring formal verification to AI: The startup is developing mathematical verification tools that can prove AI systems behave as intended. As AI safety and reliability become core market and regulatory requirements, verification and auditing tools are emerging as a standalone category — a signal that the AI stack is maturing beyond raw model capabilities. TechCrunch
Google bets on Gemini to reinvent the smart home speaker: Google is integrating Gemini into its Nest smart home speakers, evolving them from simple voice-command devices into context-aware AI assistants capable of complex multi-step tasks. On-device AI inference is becoming a key differentiator in consumer hardware. TechCrunch
DeepL acquires Mixhalo to enter live-event audio translation: Text translation leader DeepL has acquired Mixhalo, a live-event audio streaming and translation company. AI translation is expanding from documents into real-time voice and events, opening new verticals in conferences, performances, and broadcasting. TechCrunch

AI coding agents can now autonomously direct physical robots to install GPUs and cut zip ties: Researchers demonstrated a fully autonomous loop where AI coding agents generate and deploy robot-training programs for complex physical tasks like GPU installation. The chain — “AI writes code that controls robots” — is now a working reality. This signals a fundamental acceleration in manufacturing and logistics automation, where the bottleneck shifts from human programming to physical hardware constraints. Ars Technica
Only 16% of Americans believe AI will have a positive impact on society: A new survey shows public perception of AI has continued to decline. For startups building AI products — especially in B2C — trust must be treated as a core product strategy, not an afterthought. Functional capability alone will not drive adoption if users distrust the technology. TechCrunch
World leaders want American AI — they just don’t want America to be able to turn it off: Governments and enterprises globally are eager to adopt US AI technology, but concerns are rising that Washington could remotely revoke access or impose controls. AI is transitioning from a technology product into geopolitical infrastructure, making AI sovereignty and local hosting central policy priorities worldwide. Startups building global B2B SaaS should factor this fragmentation into their architecture from day one. TechCrunch

Now is the time to start world-model PoCs in simulation-heavy industries: As companies like Odyssey mature, the ability to use digital twins of the physical world as training environments is becoming practical. Founders in manufacturing, logistics, construction, and autonomous systems should experiment with world-model APIs for robotics simulation, safety scenario testing, and process optimization. TechCrunch
AI formal verification is emerging as an enterprise trust differentiator: Pramaana Labs’ funding signals that verification and auditability are becoming independent product categories. Startups building enterprise AI products that embed verifiability and audit trails from the start will have a competitive advantage as regulations tighten. TechCrunch

Microsoft’s DeepSeek integration and the contradictory geopolitics of the AI supply chain: At the same time the US government is blocking foreign access to Anthropic’s frontier models on national security grounds, Microsoft is considering integrating DeepSeek’s Chinese model into its enterprise Copilot product. The lack of a coherent geopolitical framework makes regulatory outcomes unpredictable. Founders building global AI products need to design for jurisdictional fragmentation. The Information
Public trust in AI is eroding — is the trend reversible?: With US public positivity toward AI dropping to 16%, adoption headwinds may intensify, particularly for consumer-facing AI startups. The industry needs a coordinated response focused on transparency, accountability, and demonstrated reliability — not just capability benchmarks. TechCrunch