TLDRocket
6 August 2026
The capital pouring into AI has fundamentally shifted strategy. Sequoia's $10 billion mega-fund—its largest commitment in 54 years—signals the industry's new orthodoxy: concentrate bets on fewer, larger rounds rather than diversify. The firm's backing of Anthropic at a $965 billion valuation after a near-tripling in five months shows where the real money flows. Yet the headline story is subtler: nearly every major model release this week demonstrated real capability gains while the business model questions linger. OpenAI's GPT-5.5 tops benchmarks but hallucinates more than Claude and Gemini. Anthropic's Claude Opus 5 costs $2.03 per task yet somehow still underbids OpenAI while outperforming its predecessor. Z.ai's GLM-5.1 works autonomously for eight hours at $1.40 per million tokens. ByteDance's Seedance 2.0 reached 736 million CapCut users at 24 cents per second. The proliferation isn't chaos—it's proof the technology is real and spreading fast. Meanwhile, the infrastructure problem got darker: Microsoft, Alphabet, Amazon, and Meta are all building natural-gas power plants to feed AI compute, straining their net-zero pledges. DeepMind lost four senior researchers—Jeff Dean, Sanjay Ghemawat, Oriol Vinyals, Quoc Le—to Discovery Loop, a startup automating ML research itself. The shift signals where smart money sees the next frontier: not better models, but using AI to build the systems that build models. The venture capital is no longer chasing applications. It's chasing leverage.
Read the full briefing →
