Google used its Cloud Next conference in Las Vegas to do something quietly dramatic: it retired Vertex AI as the primary enterprise development environment and replaced it with the Gemini Enterprise Agent Platform — a single surface for building, deploying, governing, and monitoring AI agents. The platform bundles identity management, an agent registry, observability tools, and a low-code Agent Studio. Google also committed $750 million to help its 120,000-strong partner ecosystem build agentic products. The subtext was hard to miss: Google wants to be the operating system for enterprise AI agents, not just the model provider. According to their own survey, the average organisation now runs 12 agents — which either means agents have already gone mainstream or that "agent" has become a very elastic word.

The same day Google was pitching agent infrastructure, DeepSeek released what it's calling V4 — preview versions of the Flash and Pro models, exactly one year after the Chinese lab first rattled Silicon Valley. Pro is a 1.6 trillion-parameter mixture-of-experts model with 49 billion active parameters; Flash is leaner at 284 billion total, 13 billion active. Both ship with a 1 million-token context window and a new Hybrid Attention Architecture that DeepSeek says improves memory across long conversations. And as before, it's all open-source under Apache 2.0. Early benchmarks put it close to — though not quite at — the frontier set by GPT-5.4 and Claude Opus 4.6, but at a fraction of the cost. The pattern from January 2025 repeats: a Chinese lab delivers roughly comparable performance for dramatically less money, and the US incumbents have to decide what that means for pricing.

The capital side of AI had a big week too. Amazon agreed to invest up to $25 billion in Anthropic, on top of the $8 billion it had already committed, bringing the total to $33 billion. The structure is $5 billion upfront at Anthropic's $380 billion valuation, with $20 billion more tied to commercial milestones. In return, Anthropic committed to spending over $100 billion on AWS technologies over the coming decade and plans to bring nearly 1 gigawatt of Trainium capacity online by year-end. The timing is notable — Amazon made a similar $50 billion bet on OpenAI just two months earlier. It's hedging at a scale that would have seemed absurd 18 months ago. Separately, Cognition, the startup behind the Devin AI coding agent, is in early talks to more than double its valuation to $25 billion. The appetite for autonomous developer tools shows no sign of slowing, even as the space gets more crowded.

On the influence front, Anthropic outspent OpenAI on lobbying for the first time. It dropped $1.6 million in Q1 2026 — more than four times its spend a year ago — focused on Defence Department procurement, acceptable use policy, export controls, and energy infrastructure. OpenAI spent $1 million. Across the industry, eleven major tech companies spent a combined $20 million in the quarter, averaging $226,000 per day. Meta topped the list at $7.1 million. The lobbying surge arrives as the EU AI Act approaches full application in August 2026, and as the US continues its lighter-touch approach under an executive order that blocks state-level AI laws deemed incompatible with federal policy. Nobel laureate Geoffrey Hinton addressed the UN this week, likening AI to "a very fast car with no steering wheel" — a metaphor that feels more apt the bigger the funding rounds get.

And in the broader landscape: Q1 2026 venture funding hit $300 billion globally, with $242 billion — 80% of the total — going to AI. OpenAI alone raised $122 billion at an $852 billion valuation. The company also completed its first media acquisition, buying Silicon Valley talk show TBPN for a reported low hundreds of millions; the show now reports to OpenAI's chief political operative Chris Lehane, though it retains editorial independence. Samsung announced plans to double its mobile devices running Google's Gemini AI to 800 million units by year-end. The distribution plays keep getting bigger, the models keep getting cheaper, and the question of who captures value in this stack remains wide open.

Signals

Week in Review: April 24