May 2nd 2026
Google Cloud revenue grew 63% YoY to $20B with Sundar Pichai explicitly stating the company is "compute-constrained," while Microsoft's AI business runs at $37B annualized revenue (up 123% YoY) and AWS Trainium chips alone hit a $20B run rate. This demonstrates that despite stratospheric infrastructure spending, demand still exceeds supply across the stack.
Cerebras is pursuing one of the largest AI hardware IPOs, seeking a ~$40B valuation as it positions itself as a major player in AI infrastructure competing against Nvidia and other chip makers. This is significant as a concrete signal of investor appetite for specialized AI silicon and the economics of the AI hardware stack.
Sources indicate Huawei's Ascend 950PR chip orders are surging while Nvidia stalls in the Chinese market, with Huawei's AI chip revenue projected to reach ~$12B in 2026 versus $7.5B in 2025. This reflects the competitive dynamics and geopolitical fragmentation reshaping the AI infrastructure market.
Lux Capital-backed Majestic Labs has built server infrastructure designed to address the "memory wall" constraint limiting large model inference speed, offering roughly 1,000x more memory per server than competitors. This directly addresses a critical infrastructure constraint in AI deployment economics.
By releasing the same runtime and harness that powers their $50B product, Cursor validated that the moat lies in tool-calling, context management, and error recovery infrastructure—not the underlying LLM—enabling developers to build autonomous agents billed token-by-token.
The Department of Defense finalized agreements with eight major AI companies to deploy frontier capabilities directly on classified defense networks for operational use, representing a significant acceleration of AI adoption in military decision-making. This signals both the maturation of enterprise AI deployment and potential concentration of defense AI infrastructure among a narrow set of vendors.
I read the full Anthropic report on Project Deal, a controlled experiment where 69 employees deployed AI agents (Claude Opus vs. Haiku) to negotiate real marketplace transactions for physical goods over one week in December 2025. The report discloses concrete metrics: 186 deals struck at $4,000+ total value, Opus agents completing ~2 more deals than Haiku on average, Opus extracting $3.64 more per item sold and paying $2.45 less as buyers, with a striking finding that participants with weaker models experienced measurable disadvantage but failed to perceive it. This directly addresses AI economics, agent capability stratification, and real-world implications for AI-mediated commerce.
Read moreI read the full CNBC piece on Intel's 114% April stock surge and underlying business turnaround. The article provides concrete metrics on Intel's recovery (18A chips ramping from Arizona, 14A in development, CPU demand exceeding supply), strategic shifts under new CEO Lip-Bu Tan (workforce cuts, fab delays, foundry focus), and emerging revenue drivers including advanced packaging (EMIB technology competing with TSMC's CoWoS) and new foundry commitments from SpaceX/Tesla/xAI. The piece includes specific analysis of how agentic AI is driving CPU demand resurgence and how Intel's advanced packaging could become a multi-billion dollar business, with concrete customer commitments from Amazon, Cisco, and potential Google/Nvidia deals.
Read moreI read the full technical post on Cloudflare's new Dynamic Workflows feature, which extends their durable execution engine to support multi-tenant, dynamically-loaded workflow code. The post details the architecture for routing tenant-specific TypeScript workflows through a Worker Loader to the Workflows engine, with concrete code examples showing how platforms can now let customers ship their own workflow implementations without redeployment. This is directly relevant to infrastructure economics and platform strategy: it demonstrates how Cloudflare is positioning Workers as a foundational primitive for agentic and multi-tenant applications, with implications for how AI agents and customer-built logic can be deployed at scale with minimal idle costs.
Read more