AI News Roundup — March 17, 2026 (Enterprise + Product)
TL;DR: Anthropic is quietly displacing OpenAI as the default enterprise AI vendor, while both labs are now courting private equity to build corporate consulting arms. Meanwhile, the "AI everywhere" push is hitting a wall — Microsoft just killed off its ambient Copilot vision for Windows 11.
Top Stories
Anthropic Becomes the New Enterprise Default — New data from Ramp's lead economist shows a surge in enterprises choosing Anthropic over OpenAI for their first AI contracts, with one analyst declaring flatly: "Anthropic is the new default for businesses." Axios
OpenAI + Anthropic Both Courting Private Equity — Both labs are in separate talks with PE groups to create enterprise AI consulting arms; OpenAI's proposed deal carries a ~$10B pre-money valuation, giving the firm a faster on-ramp into corporate portfolios. Reuters / Axios
Morgan Stanley: AI Scaling Laws Are Still Holding — The bank warns most companies aren't ready for the AI capability jump it sees coming in 2026, arguing the current compute buildout will yield bigger-than-expected gains in the near term. Fortune
Enterprise Connect 2026: CX Leaders Ask "Is AI Actually Worth It?" — The Las Vegas event marked a shift to outcomes-driven accountability, with CX executives demanding proof of ROI on AI investments rather than accepting platform promises at face value. National Today
Shipping & Platform
Anthropic Goes GA on 1M Context at No Premium — Claude Opus 4.6 and Sonnet 4.6 now include the full 1M token context window at standard API pricing with no long-context surcharge — a significant cost unlock for agentic and document-heavy workflows. Cursor Forum
Claude Opus 4.6 Adds Adaptive Thinking + Compaction API — The new "Compaction API" addresses context rot in long-running agents, letting models compress conversation history intelligently; Opus 4.6 hits 76% multi-needle retrieval accuracy at 1M tokens. InfoQ
Microsoft Kills Ambient Copilot in Windows 11 — After user pushback, Microsoft has shelved plans to embed Copilot into notifications, Settings, and File Explorer, pivoting to a more opt-in AI approach and allowing users to disable features to reduce "AI bloat." Windows Central
Apple Launches MacBook Air + Pro with M5 — Both machines ship with expanded on-device AI capabilities via M5/M5 Pro/M5 Max, advancing Apple's strategy of keeping AI inference local rather than cloud-dependent. Apple
Cognizant AI Factory Goes Live — Cognizant launched a multi-tenant, enterprise-grade AI infrastructure offering built on Dell + NVIDIA stack, targeting large orgs that want managed AI compute without building their own. TradingView
Policy & Governance
Anthropic Launches Institute for Societal AI Risk — The newly formed Anthropic Institute consolidates safety, policy, and societal-impact work under one roof — including a team actively using Claude to find and exploit code vulnerabilities as a red-teaming exercise. SiliconANGLE
US Federal Push to Preempt State AI Laws — A new executive action asserts broad federal authority to override state-level AI regulation in favor of a "minimally burdensome national standard," setting up a major legal battle with California and other active state regulators. Ropes & Gray
China's Two Sessions: AI Governance Enters the Agent Era — Chinese policymakers are grappling with autonomous agents as a new governance frontier — moving the regulatory conversation from "how do we scale AI?" to "who is legally responsible when agents act?" Forbes
One Take
The Anthropic vs. OpenAI enterprise story is the one to watch right now. Anthropic winning first contracts isn't just a market share stat — it signals that enterprise buyers are starting to optimize for reliability and reasoning depth over brand recognition. The PE deal structures are fascinating because they create a new distribution channel that bypasses the typical SaaS procurement cycle entirely. For anyone building on top of these models: if you're not already evaluating Claude for deep reasoning or long-context tasks, this week's 1M-context-at-standard-pricing announcement is your forcing function. Action item: Run a cost comparison on your current long-context API spend against the new Anthropic pricing before Q2 budgets lock.