Good morning. It's Monday, July 27, and we're covering the industry’s push for open AI models, why NVIDIA's valuation may be too pessimistic, and Anthropic’s new Claude Opus 5 performance leap.

Plus: this week's prompt for getting unfiltered AI feedback, and a new Original on the next frontier of humanoid robotics.

Claude Opus 5 approaches the performance of Claude Fable 5 while costing half as much, narrowing the gap to Anthropic's flagship model. Anthropic launches Opus 5 as its new default for Claude Max and top model for Claude Pro, targeting coding, automation, and knowledge work. It leads Frontier-Bench, CursorBench, and ARC-AGI 3, while matching Fable 5 on some tasks at roughly half the cost and remaining behind Mythos 5 in cybersecurity.

More than 25 tech companies warn that restricting open-weight AI models could weaken U.S. competitiveness rather than improve security. NVIDIA, Microsoft, Meta, Palantir, and others urge policymakers to target illegal AI distillation instead of imposing broad limits as Chinese models gain ground. OpenAI and Anthropic did not sign the letter, though Sam Altman later voiced support for both open and proprietary AI ecosystems.

OpenAI says AI demand is growing faster than new computing capacity can be built, forcing difficult choices about models and products. President Greg Brockman says the industry will remain constrained despite massive infrastructure spending by OpenAI, Anthropic, Microsoft, Amazon, and others. Google recently agreed to rent AI compute from SpaceX for $920 million per month, while Brockman argues customers are increasingly seeing measurable returns from AI investments.

Anthropic’s Claude Opus 5 outperforms Fable 5 on most benchmarks while costing half as much, signaling a major leap in AI performance and value.

_{^{Guest contributor: Nikita Rudin is Flexion's co-founder and CEO. Formerly at NVIDIA and ETH Zurich, he now leads Flexion in building AI systems for humanoid robots.}}

Last month, we gave a humanoid robot a single instruction in plain language: A parcel with snacks has been delivered for Flexion. Retrieve it using the stairs and come up using the elevator. Then unpack it and place the items into the empty drawer on the shelf in the snack area.

Then we stepped back. → Read the full article here.

A coalition of AI companies and technology organizations released a policy paper on July 24, 2026, arguing that open-weight AI models are essential to maintaining U.S. leadership in artificial intelligence. The signatories—including Meta, Microsoft, NVIDIA, Hugging Face, IBM, Mistral, Palantir, Replit, Perplexity, and Y Combinator—contend that openly available model weights promote competition, reduce costs, give organizations greater control over their AI systems, and accelerate adoption across industries.

While acknowledging security risks, the paper argues that transparency enables broader testing, red teaming, and vulnerability discovery, making open weight models an important component of AI safety. The group also urges policymakers to expand access to computing resources, invest in shared AI infrastructure, and avoid restrictions that could limit open-model development or push innovation overseas. → Read the full article here.

A new analysis argues that NVIDIA's stock is priced far more conservatively than its growth prospects suggest, despite analysts expecting revenue to increase 83% this year. Trading at roughly 17 times next year's earnings—well below its five-year average of 36 times—NVIDIA is valued as though growth will slow sharply after 2027, even as Wall Street forecasts sales to reach about $560 billion by fiscal 2028.

NVIDIA faces mounting competition from AMD, cloud providers, and startups developing custom AI chips, but it continues to dominate the AI accelerator market, particularly for inference. The analysis contends that NVIDIA's entrenched ecosystem, continued hyperscaler spending, and resilience in a potential AI downturn make the current valuation appear more pessimistic than its long-term fundamentals. → Read the full article here. (Paywall)

SpaceX Nears AI Break-Even Value (Paywall): Morgan Stanley says a $100 share price would value SpaceX's AI business at effectively zero.

ChatGPT Voice Comes to Desktop: OpenAI launched ChatGPT Voice for macOS and Windows, enabling voice control of multiple AI agents.

AI Opt-Out Classroom: "Avoiding AI" workshops at libraries are drawing 5x normal turnout as patrons ask librarians how to disable AI features.

Microsoft Bets on Smaller AI Models: Microsoft is replacing some frontier AI with cheaper in-house models, cutting GPU costs and expanding its strategy.

Cognition Buys Poke for Personality: Cognition acquired AI assistant Poke to give coding agent Devin a more conversational, collaborative personality.

Corgi Eyes $4B Valuation: The AI insurance startup reportedly raised another round just eight weeks after its last, fueled by rapid revenue growth.

Midjourney Acquires Co-Star: Midjourney bought astrology app Co-Star, adding its 4.3 million monthly users as it expands beyond AI image generation.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team