Key Highlights
- Uber capped employee AI tool spending at $1,500/month after burning through its entire annual AI budget in four months — the bluntest signal yet that enterprise AI ROI remains hard to prove, with Uber’s own COO admitting it’s “very hard to draw a line” between tool usage and business outcomes.
- Microsoft made a major move toward AI independence at Build 2026, unveiling seven proprietary MAI models, a “Scout” Autopilot agent in Teams, and an AI-designed quantum chip (Majorana 2) — loosening its reliance on OpenAI.
- NVIDIA and Microsoft unified the agentic AI stack from Windows devices to cloud — RTX Spark and DGX Station hardware, the NemoClaw agent blueprint, and the OpenShell security runtime — with Anthropic’s Claude models now running natively on Blackwell systems in Azure.
- Google gave website owners a Search Console toggle to control whether their content appears in generative AI Search, as AI Overviews reaches 2.5 billion monthly users and AI Mode passes one billion — a meaningful concession on publisher control.
- Cursor shared a year of lessons building cloud agents, which now generate 40% of its internal pull requests and process 50M+ daily actions after a migration to Temporal for reliability.
Analysis & Opinion
Uber Caps Employee AI Spending After Blowing Through Its Budget — TechCrunch
Uber has imposed a $1,500 monthly per-employee cap on AI tools including Claude Code and Cursor, tracked through an internal usage dashboard, after its CTO disclosed in April that the company had exhausted its entire annual AI budget in just four months. The reversal is striking because Uber had previously urged staff to “use AI as much as possible,” even running internal leaderboards to gamify consumption. COO Andrew Macdonald conceded it remains “very hard to draw a line” between AI usage and tangible business outcomes. The episode crystallizes a sector-wide anxiety: enterprises are spending heavily on AI tooling while ROI stays largely theoretical, and unmetered “use it all you can” policies collide quickly with real budgets. Expect more companies to move from encouragement to metering as finance teams demand accountability.
Advancing Youth Safety and Opportunity Through Global Leadership — OpenAI
OpenAI laid out its commitments on youth safety, framing responsible deployment for younger users as a core part of how it ships products. The post emphasizes safeguards and age-appropriate experiences alongside expanding educational opportunity, positioning safety and access as complementary rather than competing goals. The move lands amid intensifying regulatory and legal scrutiny of AI’s effects on minors — and follows recent litigation pressure on the company over harms tied to its products. The substance to watch is whether these commitments translate into enforced age verification, default protections, and measurable transparency rather than principles statements.
Microsoft Paves Its Own AI Way at Build — The Rundown
Microsoft used Build 2026 to assert independence from OpenAI, unveiling seven proprietary MAI models spanning reasoning, coding, imagery, voice, and transcription through Microsoft Foundry, plus “Scout,” its first Autopilot agent built on OpenClaw that runs inside Teams. It also showed Majorana 2, a quantum chip that AI agents helped design with a claimed 1,000x reliability gain, and previewed Project Solara for agentic devices. The throughline: Microsoft wants to own more of its AI stack end to end.
Cyera Eyes $12B Valuation at 80x ARR Despite Operating Losses — TechCrunch
Data security firm Cyera is closing a round at a $12 billion valuation — an 80x multiple on its $150M+ ARR — just five months after a $9B Series F, despite remaining unprofitable and burning cash on aggressive sales hiring (500 employees added this year). The figures underscore how far AI-adjacent security valuations have run ahead of profitability; Cyera disputed the reported numbers as “factually and significantly inaccurate.”
New Products & Tools
NVIDIA Partners With Microsoft on a Unified Stack for Agentic AI — NVIDIA Blog
NVIDIA and Microsoft announced an end-to-end agentic stack spanning Windows devices, Azure, and local deployments — including RTX Spark laptops/desktops (1 petaflop, up to 128GB unified memory) and the GB300-based DGX Station for Windows (up to 748GB coherent memory, 20 petaflops FP4). Anthropic’s Claude models now run natively on Blackwell in Azure, and the new Nemotron 3 Ultra ships this month for long-running agent reasoning.
Build Personal AI Agents on Windows PCs With New Microsoft and NVIDIA Tools — NVIDIA Developer
Microsoft’s new eXecution Containers (MXC) and NVIDIA’s OpenShell runtime bring policy-based isolation and PII obfuscation to always-on, on-device Windows agents, with a Surface RTX Spark Dev Box for developers. NVIDIA claims 2x faster agentic inference via Multi-Token Prediction and Programmatic Dependent Launch on llama.cpp and vLLM.
Industrial Software Leaders Build Autonomous AI Engineers With NVIDIA NemoClaw — NVIDIA Blog
A dozen-plus engineering software vendors are building specialized “AI engineers” on NVIDIA NemoClaw — an open blueprint for long-running agents with a secure runtime — to automate the full CAD/meshing/simulation/reporting workflow, with security enforced by NVIDIA OpenShell.
Codex for Every Role, Tool, and Workflow — OpenAI
OpenAI expanded Codex beyond software development into a broader set of professional roles, tools, and workflows, integrating it across more platforms to support diverse use cases.
Travelers Deploys AI-Powered Claims Countrywide With OpenAI — OpenAI
Insurer Travelers rolled out OpenAI-powered claims processing nationwide, using AI to streamline evaluation and improve turnaround — a notable enterprise deployment in financial services.
New Opportunities, Control and Insights for Website Owners — Google
As AI Overviews hits 2.5B monthly users and AI Mode passes 1B, Google is testing a Search Console toggle that lets owners opt their content out of generative AI Search features, plus “Preferred Sources,” more prominent links, and subscription labels. It’s a meaningful response to publisher complaints about AI-era traffic, though the real test is how much control the toggle actually grants without tanking visibility.
June Android Drop: New Personalization and Safety Features — Google
Android’s June drop adds fake-call detection that flags scammers impersonating known contacts, a Google Photos “digital wardrobe,” Circle to Search outfit matching, and expanded Personal Safety features for children under 13.
What We’ve Learned Building Cloud Agents — Cursor
A year in, Cursor’s cloud agents run on dedicated VMs for parallel, long-running tasks, and the team found development-environment quality to be the single biggest factor in output quality — cloud agents must rebuild the whole setup from scratch. A migration to Temporal pushed reliability past two nines; cloud agents now drive 40% of Cursor’s internal PRs across 50M+ daily actions and 7M workflows.
Google Signs Smart-Grid Capacity Agreement With Voltus — Google
Google and Voltus struck a three-year deal to unlock up to 100MW of flexible grid capacity across the PJM region (67M people) by orchestrating batteries and smart thermostats, paying participating homes and businesses to ease demand and helping power Google’s data centers.
Google’s New Water Stewardship Commitments — Google
Google pledged to replenish more water than it consumes at its sites by 2030, having already replenished 7B+ gallons in 2025 across 165 projects and 97 watersheds, plus $500M+ invested in community water infrastructure — a response to data-center water-use scrutiny.
Research
Deploy Self-Evolving Agents With a Hermes Agent and NVIDIA NemoClaw — NVIDIA Developer
A blueprint pairs the Nemotron 3 Super model, the Hermes agent harness, and OpenShell’s sandboxed runtime to build agents that learn new skills (e.g., report formats) through conversation without code changes, while credentials stay isolated and external network access is blocked even if the agent is compromised.
Introducing RadixAttention to Trellis — Lobsters
Trellis adds RadixAttention, using a radix-tree structure plus block-paged KV caching to share common prompt prefixes across concurrent sessions — cutting redundant prefill work for repeated system prompts in privacy-preserving, on-your-own-hardware LLM inference.
thunderbolt-ibverbs: “We Have InfiniBand at Home” — Lobsters
An experimental Linux kernel module turns USB4/Thunderbolt ports on AMD mini PCs into InfiniBand-like RDMA devices, hitting ~48 Gb/s per direction at ~7µs latency and slashing a Gemma 3 27B LoRA training step from 1,359s over Ethernet to 126s — enabling tensor-parallel inference across cheap consumer hardware.
References
- Uber Caps Employee AI Spending After Blowing Through Its Budget — TechCrunch, 2026-06-02 [blog]
- Advancing Youth Safety and Opportunity Through Global Leadership — OpenAI, 2026-06-02 [blog]
- Microsoft Paves Its Own AI Way at Build — The Rundown, 2026-06-03 [blog]
- Cyera Eyes $12B Valuation at 80x ARR Despite Operating Losses — TechCrunch, 2026-06-02 [blog]
- NVIDIA Partners With Microsoft on a Unified Stack for Agentic AI — NVIDIA Blog, 2026-06-02 [blog]
- Build Personal AI Agents on Windows PCs With New Microsoft and NVIDIA Tools — NVIDIA Developer, 2026-06-02 [blog]
- Industrial Software Leaders Build Autonomous AI Engineers With NVIDIA NemoClaw — NVIDIA Blog, 2026-06-02 [blog]
- Codex for Every Role, Tool, and Workflow — OpenAI, 2026-06-02 [blog]
- Travelers Deploys AI-Powered Claims Countrywide With OpenAI — OpenAI, 2026-06-02 [blog]
- New Opportunities, Control and Insights for Website Owners — Google, 2026-06-03 [blog]
- June Android Drop: New Personalization and Safety Features — Google, 2026-06-02 [blog]
- What We’ve Learned Building Cloud Agents — Cursor, 2026-06-02 [blog]
- Google Signs Smart-Grid Capacity Agreement With Voltus — Google, 2026-06-02 [blog]
- Google’s New Water Stewardship Commitments — Google, 2026-06-03 [blog]
- Deploy Self-Evolving Agents With a Hermes Agent and NVIDIA NemoClaw — NVIDIA Developer, 2026-06-02 [blog]
- Introducing RadixAttention to Trellis — Lobsters, 2026-06-03 [blog]
- thunderbolt-ibverbs: We Have InfiniBand at Home — Lobsters, 2026-06-02 [blog]