TLDRocket
6 August 2026
The AI industry is fracturing along business-model lines. ByteDance's Seedance 2.0 demonstrates the staying power of integrated platforms: by embedding its video generator directly into CapCut's 736 million monthly users, the company bypassed the computational cost spiral that killed OpenAI's Sora. Meanwhile, Anthropic's Claude Opus 5 undercuts rivals on price while addressing user frustration with refusals, and Z.ai's GLM-5.1 can now autonomously plan and execute single tasks across eight-hour sessions—capabilities that suggested theoretical but now prove practical. The common thread: models matter less than deployment architecture and unit economics. OpenAI's Greg Brockman conceded uncertainty about sustainable business models even as his company's ChatGPT approaches one billion users. Microsoft's filing reveals the extent of its OpenAI dependence: $24.1 billion in OpenAI revenue represents roughly 70 percent of its total AI sales. The infrastructure race meanwhile accelerated into contradiction. Major AI companies are now building natural-gas power plants to meet demand, straining the net-zero climate pledges they made years ago. ChangXin Memory Technologies' Shanghai IPO—raising $8.6 billion with shares surging 466 percent on the first day—signals capital flooding toward AI supply chains. Yet trust is fragmenting. Medical AI systems tested by Stanford researchers omitted critical information in 76.6 percent of harmful errors across OpenAI, Anthropic, and Doximity models. Humanoid robots now working factory floors revealed another flaw: expressive robots that fail trigger greater suspicion than motionless ones, undermining the design assumption that lifelike behavior builds confidence. The industry has built powerful tools. It has not yet solved how to deploy them safely, cheaply, or reliably at scale.
Read the full briefing →