Descriptions:
Bloomberg Technology’s April 22, 2026 broadcast leads with Google Cloud’s announcement of its latest-generation Tensor Processing Units, introduced in two purpose-built variants: one optimized for training AI models and one for inference. The reveal came alongside a dense set of partnership announcements — including agreements with Oracle, Nvidia, CrowdStrike, and private equity firm Vista — signaling that Google Cloud is aggressively pursuing vertical integration across the AI stack. Bloomberg Intelligence analyst Mandeep Singh notes that Google’s ability to bundle its own TPUs with cloud services gives it a structurally distinct position versus hyperscalers relying on Nvidia hardware, and points to ~50% cloud growth last quarter as evidence of accelerating market share gains.
The broadcast also covers an Anthropic security incident: according to Bloomberg sources and reviewed documentation, a small group of unauthorized users gained access to Mythos, a restricted Anthropic model the company has withheld from public release due to its potential to enable dangerous cyberattacks. Investigators found the group — a Discord community with one member holding a prior contractor relationship with Anthropic — used knowledge of the company’s historical model storage patterns to locate the model rather than exploiting a formal breach.
Additional segments touch on the competitive dynamics between Google, Amazon, and OpenAI in cloud AI infrastructure, including a comparison of deal structures: Amazon provided Anthropic $5 billion in cloud credits, while Google’s $2.5 gigawatt power deal with Anthropic involved no cash transfer — a detail Singh reads as reflecting Google’s stronger negotiating position.
📺 Source: Bloomberg Technology · Published April 22, 2026
🏷️ Format: News Analysis
