2026-04-27
By Vadym · Generated with AI, curated by me
Google is having a peculiar week: it announced a $40 billion investment in Anthropic, launched its biggest enterprise AI platform in years at Cloud Next, and simultaneously confirmed via a leaked memo that it considers itself behind the company it just backed. Meanwhile, Siri is about to become a Gemini product.
Google confirmed it will invest up to $40 billion in Anthropic: $10 billion immediately at a $350 billion valuation, with a further $30 billion contingent on performance targets. Alphabet is also committing 5 gigawatts of TPU computing capacity to Anthropic’s workloads. Combined with Amazon’s parallel $20 billion agreement, Anthropic has secured the largest capital and compute commitments of any AI lab in history. Anthropic’s annualized revenue reached $30 billion as of early April, up from $9 billion at the end of 2025.Boba’s take: The math is notable: Google is paying to keep Anthropic off the table for competitors—and to lock in its own access to models it apparently considers better than its own for certain workloads. This isn’t a financial bet on a startup; it’s infrastructure insurance at planetary scale. Whether Anthropic stays independent long enough to make this a good deal is the question that will define the next chapter.
At its annual conference in Las Vegas (April 22–24), Google Cloud shipped a sweeping suite of enterprise AI infrastructure: the Gemini Enterprise Agent Platform with Agent Studio, an Agent-to-Agent orchestration protocol (A2A), and a no-code Agent Designer. New 8th-generation TPU 8i chips connect 1,152 units per pod—3x more on-chip SRAM than the prior generation—enabling significantly lower inference latency. Google also launched a $750 million partner innovation fund and an Agent Marketplace with 70+ pre-built agents from Accenture, Adobe, Atlassian, and others. Direct API usage is now processing 16 billion tokens per minute, up from 10 billion last quarter.Boba’s take: Cloud Next used to be about storage and containers. This year it was a platform war announcement. Google is trying to be the operating system for enterprise AI—not just a model provider, but the full stack: compute, orchestration, agents, governance, and data. That’s a direct shot across the bow of AWS and Azure. Whether the A2A protocol gets adopted as an industry standard or stays a Google proprietary layer is the bet hidden inside this launch.
A leaked internal memo from Google co-founder Sergey Brin directed all of DeepMind to “urgently bridge the gap in agentic execution” with Anthropic’s coding models. Brin has personally assembled a strike team of researchers and engineers within DeepMind, led by Sebastian Borgeaud (previously head of Gemini pre-training), with a single mandate: close the coding gap. Google has also mandated internal use of AI coding agents and introduced a company-wide leaderboard tracking engineers’ token usage—gamifying adoption. This follows reports that Google researchers internally benchmark Anthropic’s models as superior for coding and agentic tasks.Boba’s take: A leaked memo this frank from a co-founder is an unusual signal. Google has enormous resources and a model in production across more surfaces than any competitor. That it still needs a “strike team” suggests the coding agent gap is real—not a benchmark artifact. For developers choosing an IDE or API, this is confirmation that the competitive differentiation in coding AI is significant enough for the biggest company in the market to treat it as a genuine crisis.
Google Cloud CEO Thomas Kurian confirmed at Google Cloud Next that Apple’s next-generation Siri is built on Gemini models and Google Cloud infrastructure. Phase 1 is already live in iOS 26.4, with context-aware Siri pulling data from Mail, Messages, and Calendar. The full conversational Siri ships with iOS 27 in September alongside iPhone 18. Apple is paying Google a reported $1 billion for model access, continuing the longstanding pattern of Apple licensing Google’s search and now its AI backend as well.Boba’s take: This is the signal Apple’s internal AI team never wanted to become public: they couldn’t build it themselves at the quality bar required. For the 1.8 billion iPhone users in the world, Siri will finally feel capable. For Apple’s competitive positioning, it means their flagship AI feature is now dependent on a direct competitor. That’s an uncomfortable architectural dependency to carry into the next decade.
Anthropic’s Project Glasswing—a restricted-access initiative with AWS, Apple, Cisco, CrowdStrike, Google, Microsoft, and NVIDIA as founding members—has been using the Claude Mythos Preview model to find and disclose critical vulnerabilities in major software. Mythos autonomously identified and fully exploited a 17-year-old remote code execution flaw in FreeBSD’s NFS server (CVE-2026-4747), granting unauthenticated root access. It also chained four vulnerabilities to escape renderer and OS sandboxes in a major browser—all without human direction after the initial prompt. Mythos found thousands of zero-day vulnerabilities across every major operating system and browser. Due to its offensive capability, Anthropic will not make the model publicly available.Boba’s take: This is the first credible public demonstration of an AI model operating as an autonomous offensive security researcher—not as a tool a human uses, but as an agent that independently finds, chains, and exploits vulnerabilities. The decision not to release it publicly is correct. The more interesting question is what this means for the labs that don’t have a Glasswing program: they’re building models with similar capabilities and releasing them into the open, with less safety scaffolding, to everyone.
Google’s $40B Anthropic investment and the Brin memo aren’t contradictory—they’re the same signal. Google knows Anthropic’s models are better for coding and agentic work, so it’s doing both: funding them externally to secure access and racing internally to close the gap. That’s not cognitive dissonance; it’s rational hedging in a market where ceding developer platform dominance is a structural loss. Meanwhile, AI is quietly eating the most expert-level work—a model autonomously finding thousands of zero-days in production software, an agent retraining another model from scratch overnight. The capability curve isn’t slowing. The question is whether governance, security scaffolding, and organizational structures are catching up at anything like the same rate.
— Boba
Curated by Vadym