AI Waypoints — Week of May 10, 2026 — Edition #9
The week the AI labs walked into the consulting business — and ServiceNow declared itself the control plane.
Good morning. This week the AI labs are jumping into the consulting business with $5.5B in committed capital. Anthropic locked down 220,000 GPUs from the same data center built for Grok. ServiceNow is positioning itself as the referee for every other vendor’s agents. NIST got first look at Google, Microsoft, and xAI models before they ship. The EU rewrote its own AI rules at 4:30 a.m. And Coinbase fired 14% of its people in an early-morning memo about being “AI-native.” The bills from 2024-2025 spending are coming due. The supplier base underneath enterprise AI is reshaping itself.
1. Anthropic’s financial-services sweep: consulting partnership, M365 seat, Moody’s data — all in one week
What happened: Anthropic used the week of May 5 to make three moves at once. A new $1.5B AI services partnership with Blackstone, Hellman & Friedman, and Goldman Sachs puts Anthropic engineers inside the companies they invest in, redesigning how work gets done rather than just selling access. A full Microsoft 365 integration puts Claude directly into M365 for the first time. A Moody’s partnership feeds credit ratings and financial data straight into Claude without duplicating it. Apollo, General Atlantic, GIC, Leonard Green, and Sequoia are also backing the partnership.
Why it matters: For two years we’ve watched the AI labs and consulting firms operate as separate worlds. That’s changing. The Goldman/Blackstone partnership routes directly to the companies they own; the Moody’s deal solves the “where did that number come from” question that has killed dozens of pilots; the M365 seat gets Claude into the desktops where Microsoft and OpenAI have been the only options. Three separate moves that add up to Anthropic saying: we don’t just sell AI, we’re building the whole system.
What to do: Look at your current consulting contracts. If you’re paying for design work (planning how AI fits into your workflow), that’s getting cheaper and Anthropic wants to bid on it directly. If you’re in finance, ask Anthropic when Moody’s data comes through M365, then decide if your current OpenAI setup still makes sense.
2. OpenAI raises $4B at a $10B valuation for “The Development Company”
What happened: Bloomberg reported on May 4 that OpenAI is raising $4 billion from 19 investors against a $10 billion valuation for a separate enterprise-services entity called The Development Company. The announcement arrived within hours of the Anthropic-Goldman-Blackstone announcement. It’s structured to put OpenAI engineers and consultants inside enterprise customers to move generative AI from pilots into production.
Why it matters: Two of the three leading AI companies now have consulting arms. That’s different from how Microsoft, Salesforce, and ServiceNow worked, they built partner networks instead of doing the work in-house. The AI labs are saying the partner channel is too slow and the money is too good to ignore. Every conversation with a CIO in the second half of 2026 will have a new question: who actually does the work?
What to do: Before signing a deal with an AI lab, ask who’s going to do the actual work. Get it in writing. The same project costs and runs completely differently depending on the answer.
3. ServiceNow ships Action Fabric and turns into the referee for all agents
What happened: ServiceNow announced Action Fabric on May 5 — a protocol that lets any agent (Claude, Copilot, custom-built) do work inside ServiceNow. The expanded AI Control Tower watches across five areas: Discover (what agents are running), Govern (enforcing rules), Secure (keeping it safe), Observe (seeing what’s happening), and Measure (counting results). New partnerships with Microsoft Agent 365, Lenovo, and Nvidia extend the watch across more systems. New AI specialists for IT, customer service, HR, and security all went live the same day.
Why it matters: ServiceNow just said it’s the place where all agents answer to, not just ServiceNow’s own agents, but anyone’s. Salesforce and Microsoft only police their own (with claims of extending to others). If this works, ServiceNow becomes the line item every CIO budgets for when running multiple AI agents, the way Okta became essential for login.
What to do: If you use ServiceNow, ask for written confirmation that Action Fabric treats Anthropic, OpenAI, and Microsoft agents the same way. If not, decide before Q3 whether you want ServiceNow, Microsoft, or your own system for oversight. Running all three is a headache you don’t want in 2027.
4. NIST gets to test Google, Microsoft, and xAI’s models before they ship
What happened: NIST announced agreements on May 5 with Google DeepMind, Microsoft, and xAI to evaluate their AI models in classified labs before release. They already have the same deal with OpenAI and Anthropic. NIST has now tested more than 40 models, some still unreleased. The deals started after national security concerns about Anthropic’s Mythos model for cybersecurity.
Why it matters: The Trump administration is quietly restoring pre-launch review that it dismantled in 2025 not through law, but through NIST agreements. Every company running US AI models now knows the federal government tested them first. That changes who knows what: the government has test results before you do.
What to do: When you renew with an AI vendor, ask them to confirm whether NIST tested the model and what they found. They’ll say no to details. Push for at least a yes/no on whether NIST flagged issues that matter to your business. That’s more than most boards have today.
5. EU delays AI rules by 16 months in a 4:30 a.m. deal
What happened: At 4:30 a.m. on May 7, the European Parliament and Council reached provisional agreement on changes to the AI Act. High-risk AI systems got a 16-month reprieve (from August 2026 to December 2027). AI used in safety-critical areas (like healthcare or finance) has until August 2028. Smaller companies with fewer than 250 people are exempted.
Why it matters: This is the resolution to last week’s collapse. Sixteen months is real difference for companies that were scrambling to be ready by August. But don’t read this as a break as the EU is actually strengthening enforcement, and guidance is still coming. The deadline moved; the homework reprieve looks questionable.
What to do: Don’t shelve the work you did to map which of your AI systems are high-risk. The deadline moved, but the rules didn’t go away. Treat this as 16 months to get ready properly. And check the small-company definition before next budget — it might not apply to you.
6. Coinbase fires 14% calling it “AI-native”; PayPal plans 20% cuts
What happened: Coinbase CEO Brian Armstrong announced on May 5 that 700 people (14% of staff) are being let go, with the company reorganizing around “player-coach” managers and one-person teams running AI agents. PayPal announced plans to cut 20% of its 24,000-person workforce over two to three years using AI and automation to save $1.5 billion. Both companies joined Meta, which announced 8,000 layoffs starting May 20.
Why it matters: Layoffs blamed on AI have gone from less than 8% in 2025 to roughly 20% in early 2026, and these announcements will push it higher. Meanwhile the US labor market added 115,000 jobs in April. So two things are true at once: the overall market is hiring, but white-collar jobs in tech and finance are shrinking fast. “AI-native” is becoming buzzword cover for cutting costs, whether the AI actually works yet or not.
What to do: If anyone in your company is drafting “AI-native restructuring” language for layoffs, stop them. Name the specific workflows that AI will handle.
7. Anthropic gets access to SpaceX’s massive GPU warehouse
What happened: On May 6, SpaceX and Anthropic signed a deal giving Anthropic access to Colossus 1 — the Memphis data center xAI built for Grok. It’s the world’s biggest single data center, with 220,000 GPUs running at 300MW. Bloomberg confirmed it the same day. They’re also talking about AI satellites. The Memphis deal is real and happening now.
Why it matters: Anthropic’s been quietly running short on processing power and if Claude keeps improving but you can’t run it, companies get nervous about betting on it. Colossus being available to Anthropic (even though it was built for Grok and that training reportedly has moved to Colossus II) shows that compute has become shared infrastructure. Between Anthropic’s new deal and Google/Amazon/Broadcom expanding capacity, Anthropic now has power from four different sources instead of one. That changes whether you trust the company to stay competitive.
What to do: If you’ve been hesitant about Anthropic because of capacity questions, ask your account rep for a new timeline now that Colossus is available. If you’re using multiple AI vendors, factor Anthropic’s stronger supply position into your routing plans for 2027. The “Anthropic is constrained” discount is fading.
Also from Signal Finder this week:
When does the tokenmaxxing math actually become real? — Per-task LLM cost is 50-300x cheaper than human typing and falling ~200x per year. People are still being hired. What’s actually holding up the displacement math, and when does it stop holding.
Microsoft just put numbers on the operating-model problem — The Work Trend Index found organizational factors — culture, manager behavior, talent practices — drive 2x the AI impact of individual effort. The bottleneck in your rollout probably has a job title.
Your CISO has a Quantum deadline. Your CIO doesn’t. — Quantum in 2026 is two decisions on two desks with different timelines and different budgets. Conflating them is how cryptography work goes unfunded while the lab demos get the slide deck.
What are we missing? I deliberately skipped the RSAC 2026 agent-identity wave (Microsoft Entra Agent ID, Cisco, Palo Alto, CrowdStrike) — covered as a category in Edition #7 and the Shadow AI cheat sheet, no fresh decision-driving signal this week. I also did not include the broader hyperscaler Q1 capex digest (Microsoft $190B, Meta $125-145B, Google $180-190B) — those landed in Edition #8 last week. If you saw something I should be tracking, hit reply.
References:
Blackstone newsroom — Anthropic $1.5B services JV: https://www.blackstone.com/news/press/anthropic-partners-with-blackstone-hellman-friedman-and-goldman-sachs-to-launch-enterprise-ai-services-firm/
Fortune — Anthropic M365 + Moody’s + financial agents: https://fortune.com/2026/05/05/anthropic-wall-street-financial-services-agents-jamie-dimon/
TechCrunch — Anthropic + OpenAI joint ventures: https://techcrunch.com/2026/05/04/anthropic-and-openai-are-both-launching-joint-ventures-for-enterprise-ai-services/
ServiceNow newsroom — Action Fabric + AI Control Tower: https://newsroom.servicenow.com/press-releases/details/2026/ServiceNow-turns-enterprise-AI-chaos-into-control-with-the-platform-for-governed-autonomous-work/default.aspx
HPCwire — CAISI / NIST agreements with Google, Microsoft, xAI: https://www.hpcwire.com/off-the-wire/nists-caisi-announces-new-frontier-ai-testing-agreements-with-google-deepmind-microsoft-xai/
Council of the EU — AI Omnibus political agreement: https://www.consilium.europa.eu/en/press/press-releases/2026/05/07/artificial-intelligence-council-and-parliament-agree-to-simplify-and-streamline-rules/
CNBC — Coinbase 14% layoffs: https://www.cnbc.com/2026/05/05/coinbase-cuts-headcount-by-14percent-citing-ai-acceleration-the-shares-are-gaining.html
xAI — New Compute Partnership with Anthropic: https://x.ai/news/anthropic-compute-partnership
Bloomberg — Anthropic, SpaceX Sign Deal to Boost AI Computing Power: https://www.bloomberg.com/news/articles/2026-05-06/anthropic-inks-computing-deal-with-spacex-to-meet-ai-demand
How this was made: Claude Opus 4.7 researched and drafted this against May 3-10 sources. Claude verified every claim against live sources and checked the quality of each one. Karthik read it for voice (it’s his byline) and picked the headlines. Full details: AIWaypoints-Edition9-2026-05-10-MANIFEST.json.