Descriptions:
Bloomberg Technology’s April 20 broadcast leads with a significant AI infrastructure story: Google is preparing to announce a new generation of TPUs at a Las Vegas event later that week, with Bloomberg reporting the company plans to introduce a dedicated inference chip separate from its existing general-purpose training and inference TPU. Bloomberg also reports Google is in active talks with Marvell about a future TPU collaboration, news that immediately moved Marvell’s stock.
The TPU segment, featuring Bloomberg AI infrastructure reporter Dina Bass, explores Google’s competitive positioning against Nvidia in the accelerator market. A key argument: Google is currently the only frontier model lab that both trains top-tier AI models and manufactures its own AI accelerator chips — a vertical integration advantage that lets it co-optimize hardware and software using live inference workloads from Gemini. Major confirmed TPU customers include Anthropic (a large multi-year deal described as a validation event for the technology), Meta (a multibillion-dollar multiyear agreement for its first large tranche), and quantitative hedge fund Citadel, which is scheduled to speak at the Google Las Vegas event. Despite Nvidia CEO Jensen Huang’s public claims that demand is limited to Anthropic, Bloomberg’s reporting suggests a broader and constrained supply picture — Google is prioritizing frontier lab customers who can best utilize what TPUs offer.
Other segments cover the Cerebras IPO returning to the table after prior delays, AST SpaceMobile shares falling after a Blue Origin launch failure over the weekend, and analyst Mark Gurman’s analysis of Apple’s upcoming WWDC focus on Siri — including a reported standalone chatbot app with web search summaries designed to compete directly with Gemini and ChatGPT.
📺 Source: Bloomberg Technology · Published April 20, 2026
🏷️ Format: News Analysis
