The industry is rapidly shifting from experimental chatbots to production-grade autonomous agents, demanding unprecedented levels of compute and capital. This week's curated links highlight the massive investments in AI infrastructure, the evolution of enterprise pricing models, and the geopolitical complexities of the global AI supply chain.
Infrastructure & Pricing
DeepSeek Slashes V4-Pro Prices by 75% — Cache Hits Cut to 1/10: DeepSeek cuts V4-Pro pricing by 75% and cache-hit fees to 1/10, making it roughly 100x cheaper than Western-frontier models.
Mag 7 Earnings Day — $700B Combined AI Capex Under the Microscope: Alphabet, Microsoft, Meta and Amazon report Q1 2026 earnings with combined AI capex approaching $700B, validating the infrastructure buildout thesis.
Google to Sell TPU 8t/8i Chips to Select Customers — Latest Anti-Nvidia Shot: Alphabet is selling custom TPUs (8th-gen 8t for training, 8i for inference) to select customers, escalating chip-supply competition with Nvidia.
Enterprise Adoption
OpenAI Ends Microsoft Exclusivity — Multi-Cloud AI Era Begins: OpenAI can now run all products on any cloud through 2032, clearing the path for Amazon's $50B investment and ending Microsoft's exclusivity.
Microsoft Agent 365 Goes GA May 1 — Enterprise AI Agent Governance Arrives: Microsoft Agent 365 launches as an enterprise control plane for AI agents, discovering shadow agents and enforcing governance at $15/user/month.
Anthropic Overtakes OpenAI in Enterprise AI — 40% Market Share vs 27%: Anthropic now commands 40% of enterprise LLM API spend (up from 12% in 2023), surpassing OpenAI's 27% due to superior long-context handling.
Amazon Launches AI Productivity Suite — Connect Decisions, Connect Talent & Quick Suite: Amazon released three AI tools: Connect Talent for AI-led hiring, Connect Decisions for supply chain analysis, and Quick Suite for business intelligence.
Microsoft says it has over 20M paid Copilot users, and they really are using it | TechCrunch: Microsoft Copilot reached 20M paid enterprise seats (up from 15M in December) with agent mode now default in Word, Excel and PowerPoint.
Robotics & Physical AI
Japan Airlines trials humanoid robots as ground handlers: Japan Airlines is deploying 130cm humanoid robots for ground handling at Tokyo Haneda starting May 2026, driven by acute labour shortages.
Physical Intelligence Releases π0.7 — Robot Skills Generalise Across Bodies: π0.7 is a robot-control model that transfers learned skills across different robot bodies, solving the long-standing cross-body skill transfer problem.
Geopolitics & Regulation
House Committees Probe Cursor-Maker Anysphere and Airbnb Over Chinese AI Use: US House committees are investigating Anysphere (Cursor) and Airbnb for using Chinese AI models, citing national security concerns.
Musk v. Altman is just getting started | TechCrunch: Elon Musk testified for three days that Sam Altman betrayed OpenAI's nonprofit mission by converting it to for-profit, with Altman's testimony still pending.
Research & Development
ICLR 2026: 'The Reasoning Trap' — Smarter AI Models Hallucinate More Tools: ICLR research reveals that training AI models for stronger reasoning increases tool-hallucination rates in lockstep, creating a reliability-capability trade-off.
Sam Altman: GPT-5.5-Cyber Begins Rollout to Critical Cyber Defenders: OpenAI is rolling out GPT-5.5-Cyber, a frontier cybersecurity model, to critical cyber defenders with government-backed trusted-access arrangements.
Karpathy at Sequoia AI Ascent 2026: 'Vibe Coding Raised the Floor; Agentic Engineering Raises the Ceiling': Karpathy outlines three themes: new LLM-native categories, the "jaggedness" of LLM capability, and the emerging agent-native economy.
Mayo Clinic REDMOD AI Detects Pancreatic Cancer Up to 3 Years Before Diagnosis: Mayo Clinic's REDMOD model detected pancreatic cancer up to 3 years early in a validation study, spotting 3x more early cases than experienced radiologists.
That’s all we have for this week. We’ll return with more curated news in our next edition.
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
