At Advancing AI 2026, AMD revealed EPYC Venice (256 Zen 6 cores, TSMC 2nm) and the Helios AI Rack at roughly $5-5.5M with 72 MI455X GPUs. Anthropic committed to 2 GW of AMD MI450-series silicon. That is the first credible rack-scale answer to Nvidia's GB300, and it should pressure inference pricing.

Google Cloud hit $24.77B in Q2, up 82 percent year over year, with Gemini at 950M monthly users. Alphabet then raised full-year capex guidance to $180-190B and the stock fell after hours. The bet only pencils out if AI monetization bends up by 2027.

Banks began selling portions of the $35B financing (led by Apollo and Blackstone) that leases co-designed TPUs to Anthropic across 16 capital releases toward 20-plus GW by 2028. AI infra debt is now a secondary-market asset class, letting Anthropic scale compute without dilutive equity.

Pillar Security disclosed CVE-2026-48124 (CVSS 8.5): an agent writes a shell script into the workspace and registers it as a Stop hook, which Cursor then runs locally with no approval. If you run Cursor in agent mode, upgrade to v3.0.0 today. Upgrade Cursor →

OpenAI now deploys agents for support, sales, and IT through Forward Deployed Engineers, not self-serve, with no new base model. If you want the same enterprise agents without waiting on OpenAI's engineers, self-serve builders like Lindy cover the same ground.

AI Agents are now generally available on all paid Make plans, with a credit-pricing drop that cut our 8-scenario stack from $69.95 to $18.82 a month per Make's own pricing page. Solid non-Zapier automation home. Try Make →

Self-serve enterprise agent builder, no Forward Deployed Engineers required. The obvious first stop if OpenAI Presence's white-glove model is more than your team needs. Try Lindy →

Bionic is a free local-hardware coding agent: it runs at no software cost and loses on context window and multi-file edits. The pick for compliance-bound shops that cannot cloud-stream code. Read it →

Refreshed with the SpaceX acquisition, Kimi K2.7 context window, and browser agent modes. The verdict comes down to team size, and we call it. Read it →

K3 leads agentic coding benchmarks at sub-Fable-5 pricing. Here is the cache-hit cost math and when to swap it into your agent before the July 27 weights drop. Read it →

MCP tools, a no-code scenario builder, and a credit-pricing drop from $69.95 to $18.82 a month per Make's own pricing page, bumping the rating to 4.4. We explain who should buy. Read it →

Try Make →