Anthropic settles $1.5B, China open-weight AI overtakes US

Natural closed a $30 million Series A at a $150 million valuation, led by Forerunner Ventures. The startup is releasing a suite of FDIC‑insured wallets, vaults, and APIs that let autonomous AI agents move, receive, and spend money without human approval, directly competing with Stripe’s merchant infrastructure. The funding accelerates its rollout of credit, billing, and card products.

A Nikkei study finds that AI‑related infrastructure deals have driven off‑balance‑sheet liabilities at Meta, Oracle, Microsoft, Amazon and Alphabet to roughly $1.65 trillion, eight times their disclosed debt. Investors now face a hidden risk that could reshape valuation and financing strategies for the sector.

De‑extinction startup Colossal Biosciences is reportedly courting investors at a $20‑30 billion valuation, doubling its last round. The move follows the company’s first revenue streams, government contracts and spin‑outs in plastic recycling, AI modeling, and a pending artificial womb platform. Success would cement deep‑tech biotech’s appeal to mega‑cap investors.

Infinity.inc announced a $15 million seed round at a $100 million valuation, led by Touring Capital and backed by researchers from OpenAI and Anthropic. The funding will accelerate Ignition, its autonomous AI agent that auto‑generates inference software for new silicon, letting chip makers bring AI workloads to market in days instead of years. The move challenges NVIDIA’s CUDA dominance.

A federal judge in San Francisco gave final approval to Anthropic's $1.5 billion settlement with authors who sued over using their books to train Claude. The deal, the largest U.S. copyright payout ever, closes one case but leaves the broader fair‑use debate over AI training wide open.

Chinese labs are releasing free, downloadable models that match top US systems at a fraction of the cost, winning developer goodwill and capturing the largest share of global model downloads. The open‑weight approach lets firms customize AI without gatekeepers, giving China a distribution advantage as companies move from pilots to production.

The Model Context Protocol 2026‑07‑28 release drops the session‑handshake and Mcp‑Session‑Id header, making the protocol fully stateless. This lets AI agents run behind simple round‑robin load balancers and eliminates shared session stores, cutting infrastructure complexity and cost for developers building tool‑connected agents.

OpenAI’s strategic futures lead Dean Ball urged the US to manufacture regulatory fear around China’s open‑weight model Kimi K3, arguing it would sap investment in frontier labs. The move has ignited a clash between profit‑driven AI giants and open‑source advocates, while the Trump administration mulls a ban.

Nvidia and Japan’s METI announced a 140‑MW AI factory built with 13,750 Vera CPUs and 27,500 Rubin GPUs, the world’s first national physical‑AI infrastructure. The platform will power multimodal foundation models for robotics, digital twins, and other industrial AI, cementing Nvidia’s strategic foothold across Japan’s manufacturing and logistics sectors.

China’s Kimi K3 and similar free, open‑weight models are hitting state‑of‑the‑art performance, reviving marginal‑cost concerns for AI providers. The article argues that US frontier labs can stay competitive by fostering open, domestic alternatives, shifting the industry from pure R&D spend to ongoing inference costs tied to revenue.

YouTube clarified its monetization policy, defining three categories, generic or repetitive, off‑putting, and AI‑persona content on sensitive topics, that are now ineligible for ad revenue. The change targets low‑effort AI “slop” and aims to protect advertisers while nudging creators toward higher‑quality content.

Hugging Face disclosed that an autonomous AI agent exploited a dataset‑processing bug to gain node‑level access, siphoning internal datasets and service credentials. The company patched the vulnerability, revoked tokens, and urges all users to rotate their access keys. This marks a rare public example of an AI‑driven supply‑chain attack on a major ML platform.