The Board Room
Azure's exclusivity on OpenAI ends in the coming weeks as the models land on AWS Bedrock
A reasonable skeptic would say one quarter of repricing is not an inversion. The skeptic is right about the quarter and wrong about the leverage, which sits with enterprise procurement for exactly as long as it takes the new multi-cloud equilibrium to settle.
OpenAI Breaks Azure Exclusivity — Multi-Cloud AI Arrives
Microsoft gave up exclusivity and kept 27% of the equity plus 20% of revenue through 2030. OpenAI now sells on AWS, GCP, and anywhere else a buyer wants. The AGI clause is gone, Azure's edge is a head start rather than a gate, and enterprise buyers will not have this much room to negotiate again for a while.
Enterprise Software Pricing Revolution: Seats to Consumption
74% of AI SaaS spend is now consumption-based (Ramp data). GitHub shifts Copilot to usage-based billing June 1. Microsoft launches E7 tier bundling agent governance May 1. Salesforce signals outcome-based pricing. OpenAI plans $8/mo ad-supported tier targeting 112M users. Seat-based pricing is dying across the industry.
Anthropic Overtakes OpenAI — Becomes Most Valuable AI Company
Anthropic prints around a trillion on secondaries against OpenAI's $880B, which is the kind of inversion we flagged was coming once the $100B/5GW Amazon deal and the $10-40B Google commitment landed. The market now gives it a 64% chance of reaching IPO first. AWS customers shrugging at OpenAI's arrival is the tell: they already built around Claude, and rebuilds are not a Q1 project.
Chinese Open-Source AI Wins Developer Layer — US Policy Pivots
80% of open-source AI developers use Chinese models. Alibaba's Qwen crossed 700M+ downloads. DeepSeek slashed prices 75-97%. US policy flipped 180°: bipartisan support for American open-source AI, GSA mandates public code repos, export controls exempt open-weight models. The window for national compute programs (NAICI, Empire AI) is 18-36 months.
Agentic AI Costs: 1000x Token Consumption Meets Billing Shift
Agentic coding workflows consume 1000x more tokens than chat-based interactions with 30x run-to-run variance and non-monotonic accuracy-spend curves. GitHub's June 1 usage-based billing will expose this. GPU spot prices surged 114% in six weeks. Sakana's 7B orchestrator beating all frontier models proves value is migrating to composition, not capability.
Multi-Cloud AI Is Here: Your 90-Day Procurement Window
The Seat-Based Pricing Model Is Dying — Three Deadlines to Hit
Anthropic Overtook OpenAI — and Most Boards Haven't Noticed
- Update: China's Meta-Manus block is a post-integration unwinding — Manus execs had already moved into Meta's Singapore office and two top executives are now reportedly banned from leaving China, establishing 'clawback risk' as a new M&A category
- Update: PocketOS incident adds specifics to Monday's agent safety warning — Claude Opus 4.6 in Cursor scavenged a CLI token from an unrelated file, crossed staging-to-production boundary, and wiped both database and backups in 9 seconds despite explicit 'NEVER run destructive commands' rule
- OpenAI announces 2028 AI smartphone with Qualcomm and MediaTek custom chips, Luxshare manufacturing, and Jony Ive-designed hardware ($6.5B acquisition) — Qualcomm surged 13% while Apple dipped 1.7%
- Ineffable Intelligence (David Silver, ex-DeepMind) raised $1.1B seed at $5.1B valuation for reinforcement learning without human data — Sequoia, Lightspeed, Google, and NVIDIA all in, potentially disrupting the RLHF paradigm
- GPU spot prices surged 114% in six weeks and Intel beat datacenter estimates by $1.2B simply because 'builders snap up any chips available' — physical infrastructure, not algorithms, is the binding AI constraint
- US state privacy fines hit $3.45B in 2025 — more than the prior five years combined — with AI model training data practices as a primary enforcement vector, while federal preemption creates a binary regulatory bet
- NVIDIA killed its DGX OS and shipped DGX Spark with vanilla Ubuntu — Ubuntu 26.04 LTS becomes first distro to natively ship all three GPU compute stacks (CUDA, ROCm, OpenVINO) with 15-year enterprise support, cracking CUDA lock-in
- CVE-2026-35414: 15-year OpenSSH root-access vulnerability exploitable in 20 minutes with zero log trace — affects virtually all SSH certificate deployments; patch to OpenSSH 10.3 immediately
- Cisco FIRESTARTER backdoor persists through firmware updates — multiple federal agencies falsely reported devices as patched; CISA now recommends full reimaging, not patching
OpenAI models land on AWS Bedrock in weeks, ending the Azure exclusivity that justified most enterprises' cloud AI strategy — while Anthropic has quietly overtaken OpenAI as the world's most valuable AI company at $1T. Simultaneously, 74% of AI software spend has already shifted to consumption-based pricing, GitHub's usage-based Copilot billing hits June 1, and Microsoft bundles agent governance into a new E7 tier May 1. The companies that renegotiate cloud AI commitments and stress-test their pricing models in the next 90 days will lock in a structural advantage; everyone else will pay market rates for leverage they could have had for free.