2026-07-03
By Vadym · Generated with AI, curated by me
Nobody shipped a new model today — instead Microsoft and open-model infrastructure players spent big to control what happens after the model, a soft jobs report landed with awkward timing, and NVIDIA quietly built the safety layer every warehouse robot will need before it's allowed near a person.
Microsoft announced Microsoft Frontier Company on July 2, a new unit committing $2.5B and roughly 6,000 engineers to work on-site inside customer organizations building and running AI systems, led by Rodrigo Kede Lima, former president of Microsoft Asia. The launch lands two days after Amazon committed $1B to a nearly identical forward-deployed engineering unit, following similar moves from OpenAI and Anthropic in May.
Together AI closed an $800M Series C on July 1, led by Aramco Ventures with Nvidia, Vista Equity Partners, and General Catalyst participating — more than doubling its valuation from $3.3B roughly 16 months ago. The open-model inference cloud said annual bookings passed $1.15B in Q2 2026 and plans to grow its compute capacity roughly 50x over the next five years.
The Bureau of Labor Statistics reported June nonfarm payrolls grew by only 57,000, well below the roughly 110,000 consensus and a downwardly revised 129,000 in May, while unemployment held at 4.2%. Leisure and hospitality shed 61,000 jobs, and April–May figures were revised down a combined 74,000.
Cognition’s Cascade — the local agent that powered Windsurf before its June rebrand to Devin Desktop — officially reached end-of-life on July 1. Devin Local replaces it as the default local agent, with Cognition claiming up to 30% better token efficiency and added parallel subagents; interactive editor use migrated automatically, but CI pipelines and scripts that explicitly call Cascade needed to be repointed before the cutover.
NVIDIA announced Halos for Robotics, billed as the industry’s first full-stack safety system for physical AI, porting its autonomous-vehicle safety architecture to humanoids and industrial robots. Agility Robotics is first to adopt it, pairing NVIDIA IGX Thor hardware with Halos Core software in its Digit humanoid — already deployed at Amazon, GXO, Schaeffler, and Toyota facilities — targeting certification against standards like IEC 61508 and ISO 13849.
TwelveLabs closed a $100M Series B on July 1, co-led by NEA and NAVER Ventures with Amazon, Radical Ventures, and others participating. The company builds video-native models — Marengo 3.0 for embeddings and Pegasus 1.5 for structured scene and entity extraction — and signed a multiyear deal to run workloads on AWS Trainium chips, alongside new offices in New York and London.
Nothing today was about a smarter model — it was about who controls the layer between the model and the result. Microsoft and Amazon are building armies to install AI inside enterprises; Together AI and TwelveLabs are proving that specialized infrastructure and modality-specific models are durable businesses on their own, separate from the frontier race; and NVIDIA is trying to make sure the robots those armies eventually deploy don’t hurt anyone. A soft jobs report landed right in the middle of it — no clean line from AI to layoffs, but unmistakable timing. The race to build smarter models has receded into the background this week; the race to actually plug AI into the world is the one making headlines.
— Boba
Curated by Vadym