2026-07-19
By Vadym · Generated with AI, curated by me
Today’s stories are about who controls the layer underneath the model, not the model itself. Apple just knocked Nvidia off the top of the market-cap podium, Databricks priced itself at $188 billion without shipping a foundation model of its own, and Microsoft is aiming a cut-rate security tool squarely at Anthropic’s newest product — while Beijing quietly finished building the institution meant to write AI’s international rulebook.
Apple’s market cap hit roughly $4.91 trillion on July 17, edging past Nvidia’s $4.83 trillion after Nvidia shares fell 3.5% on the day. Apple stock is up 22% in 2026 versus Nvidia’s 7%, with investors rewarding Apple’s AI strategy and comparatively modest capital spending — even though Nvidia was first to cross $5 trillion earlier this year and remains the far bigger AI infrastructure bet.
Databricks is raising roughly $3 billion in a strategic round led by existing investor Coatue, pushing its valuation to $188 billion — up from $134 billion five months ago and $100 billion the September before that. The company has repositioned from a big-data analytics platform into an AI enterprise stack, selling Unity as a multi-model gateway, Genie as an AI coworker, and Lakebase as a database built for AI agents.
Microsoft is building Project Perception, an AI security product that routes vulnerability scanning and patching across models from Anthropic, OpenAI, and its own stack, picking whichever is cheapest and best suited to each step of the job. It’s aimed squarely at Anthropic’s Mythos, which reportedly costs roughly double what Opus runs. Security chief Hayete Gallot, who joined in February, is said to be leading the effort, with a launch possible this month.
Moonshot AI’s Kimi K3, a 2.8-trillion-parameter open-weight model, took the top spot on Arena’s Frontend Code leaderboard with 1,679 points, leading in six of seven front-end domains and edging out rival closed frontier models in blind developer testing. It still trails on broad general-purpose benchmarks, but at $3 per million input tokens and $15 per million output tokens, it undercuts every closed frontier model on the workloads where it wins. Full open weights are due July 27.
Xi Jinping formally announced the World AI Cooperation Organization at the World AI Conference in Shanghai on July 17, a body first proposed a year earlier and now headquartered in Shanghai with 29 founding countries, including Indonesia, Brazil, Malaysia, South Africa, and Russia. Xi framed it as building shared AI regulation for the benefit of humanity, positioning China as a convener of international AI rules rather than a rule-taker.
Zoom out and every story today is about who controls the layer underneath the model, not the model itself. Apple beat Nvidia on capital discipline, not AI supremacy. Databricks got priced as infrastructure, not a lab. Microsoft is trying to out-cheap Anthropic on security tooling the same way it’s already doing in Office. Moonshot won a coding benchmark by being affordable, not by being the smartest model in the room. And China just built the governing body it wants writing the rules everyone else follows. The frontier-model race is still the headline, but today’s actual money and power moved one layer down — into infrastructure, pricing, and governance.
— Boba
Curated by Vadym