This Week in AI & Automation

Week of May 3 – May 9, 2026

This was the week the model labs decided consulting was their next addressable market. On May 4, Anthropic announced a $1.5B joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs to embed engineers inside portfolio companies and run AI implementations end-to-end. Hours earlier, Bloomberg reported OpenAI was raising for The Development Company, a parallel $10B services vehicle aimed at the same buyer. The next day at Think 2026, IBM made the bet that none of those buyers will accept ungoverned AI in production — shipping Sovereign Core as a runtime that bakes compliance and audit policy into the infrastructure itself, plus a next-gen watsonx Orchestrate as a control plane for multi-agent governance. ServiceNow, on the same day at Knowledge 2026, expanded its Autonomous Workforce into IT, CRM, employee service, and security/risk — and folded its AI Control Tower into every package by default. Sierra closed $950M at a $15B valuation. The EU agreed to simplify the AI Act on May 7. The enterprise AI stack is no longer about which model you pick — it's about who deploys it, who governs it, and who's liable when an agent acts.

The Big Story

Anthropic and OpenAI Both Launch Enterprise Services Companies on the Same Day

On May 4, Anthropic announced it is partnering with Blackstone, Hellman & Friedman, and Goldman Sachs to launch an AI-native enterprise services firm valued at $1.5B — with $300M committed each from Anthropic, Blackstone, and Hellman & Friedman, and additional backing from Apollo and General Atlantic. The pitch is explicit and direct: the new company will not be a strategy-deck consultancy. It will embed Claude-fluent engineers inside portfolio companies, redesign workflows, and own the integration into core processes. The PE backers bring access to the operations of hundreds of mid-market and large-cap portfolio companies. Anthropic brings the model and the engineering bench.

Hours before Anthropic's release went out, Bloomberg reported OpenAI is raising capital for a parallel vehicle called The Development Company, aiming for roughly $10B, on the same thesis: build a services arm that can deploy GPT-class models inside enterprises without going through Accenture, McKinsey, or BCG. The two announcements on the same day are not a coincidence — both labs have watched the consulting industry capture the implementation margin on every model they ship for the past three years, and both are now moving to take that margin back. Anthropic's run-rate revenue passed $30B in early 2026 (up from $9B at the end of 2025), with over 1,000 enterprise customers spending more than $1M annualized — that customer base is now the launch pad for the services play.

Source: TechCrunch | CNBC | Fortune | Anthropic

Our Take: The structural read is that the implementation layer is being repriced. For the past two years, "AI strategy" sat with the Big Four and the strategy houses, "model access" sat with the labs, and "implementation" was a margin pool the consultancies grew rapidly into. Both labs have decided that pool is theirs. For enterprise buyers, this is good news on price compression and bad news on lock-in: a Claude-engineer team that lives inside your AP workflow will not be neutral about the next model bake-off. For the Big Three model vendors, it pushes the agentic-AI buying decision down a level — from "which model do we pick" to "whose engineers own the production system." For Big Consulting, the response options are narrow: build deep multi-model implementation muscle fast, or be repositioned as the strategy-and-procurement layer above someone else's engineers. The fastest-moving operators will be specialist firms with multi-model practices and deployment track records — exactly the kind of work we do — because they remain unaligned with any one lab.

Notable Developments

IBM Ships Sovereign Core and Next-Gen watsonx Orchestrate at Think 2026

On May 5, IBM used Think 2026 to publish what it calls the AI operating model — agents, data, automation, and hybrid sovereignty as four integrated systems. The two consequential ships are IBM Sovereign Core (now generally available), which embeds governance, compliance, and execution policy at the infrastructure runtime layer and is built on Red Hat OpenShift and Red Hat AI; and the next generation of watsonx Orchestrate, positioned as an agentic control plane that lets enterprises run agents from any source under unified policy, audit, and identity. IBM also announced IBM Confluent, the productized output of November's $11B Confluent acquisition, to bring real-time streaming data into agent workflows, and IBM Concert for intelligent-ops orchestration.

Source: IBM Newsroom | StorageReview | Efficiently Connected

Our Take: Sovereign Core is a direct play for the regulated-industry buyer who watched the Pentagon-Anthropic decision last week and concluded that policy alignment and runtime control matter as much as model choice. The IBM bet is that compliance becomes a first-class architecture concern, not a wrapper, and that the enterprises with the deepest sovereignty requirements (defense, finance, healthcare, public sector) will pay for governance baked into the infrastructure — not bolted on by a consultant. That's a defensible position against both the Anthropic and OpenAI services plays, because neither lab can ship a sovereign runtime without a hyperscaler partner. Watch whether watsonx Orchestrate becomes the agent-governance standard the way Active Directory became the identity standard — IBM has the install base to make it stick if the policy model is honest about supporting non-IBM models.

ServiceNow Expands Autonomous Workforce to Every Function, Bundles AI Control Tower

Also on May 5, at Knowledge 2026, ServiceNow expanded its Autonomous Workforce into IT, CRM, employee service, and security/risk. The L1 IT Service Desk AI Specialist, CRM specialists, and employee-service specialists are available now; broader IT specialists arrive in June; security/risk specialists go to preview in June and GA in September. ServiceNow also folded its AI Control Tower into every product and package by default rather than selling it as an add-on — the Control Tower discovers agents as they appear, risk-scores them, enforces least-privilege access, and measures their business impact. ServiceNow announced Project Arc, a secure desktop AI agent built with NVIDIA, plus deeper integration between AI Control Tower and Microsoft Agent 365 — ServiceNow's specialists will list in the Microsoft Agent 365 Marketplace, and the two control planes will share governance signals across Azure, Microsoft Foundry, and Copilot Studio.

Source: ServiceNow Newsroom | Fortune | diginomica

Our Take: Bundling AI Control Tower by default is the move that matters. Last week, Microsoft made Agent 365 a $15/user/month standalone SKU. This week ServiceNow priced the equivalent capability at zero and made it the default plane. Two of the three biggest workflow vendors have now declared that agent governance is a feature, not a product — which dramatically compresses the window for standalone agent-governance startups. Combined with the Microsoft integration, ServiceNow is positioning to be the place enterprises actually run agentic workflows for IT, HR, and CRM — with the agents themselves coming from anywhere and the governance staying in one place.

Sierra Closes $950M at $15B as Customer-Support AI Hits Escape Velocity

Bret Taylor's Sierra is raising $950M led by Tiger Global and GV at a post-money valuation above $15B. Sierra builds conversational AI agents for customer support — Fortune 50 deployments, millions of monthly customer interactions, and a product that has measurably moved CSAT and deflection metrics for the named buyers. The round comes alongside other enterprise AI funding: Rogo $160M Series D (financial-workflow agents, ~$260M total raised), Hightouch $150M Series D (warehouse-native marketing automation, $1.7B valuation), and Aidoc's $150M Series E for radiology AI from earlier in the prior week.

Source: TechCrunch | CNBC

Our Take: Sierra at $15B is the public number that confirms what enterprise CX leaders have been saying privately for two quarters: AI customer support is not a co-pilot anymore, it's the primary handler for tier-1 volume. That changes how a CFO models support cost. The internal benchmark is no longer "cost per ticket vs. last year," it's "cost per ticket vs. an AI-handled tier-1." When the comparison gets that sharp, the spend reallocates fast — which is exactly the workflow Anthropic and OpenAI's new services arms are positioning to capture.

EU Council and Parliament Agree to Simplify the AI Act

On May 7, the EU Council and Parliament reached a provisional agreement under the Omnibus VII legislative package to simplify and streamline AI Act compliance — adjusting requirements for SMEs, expanding regulatory sandboxes, and reducing reporting burden in lower-risk contexts. The high-risk system enforcement deadline of August 2, 2026 still stands, including for AI used in employment, credit decisions, education, law enforcement, and high-risk financial services. Penalties for prohibited practices remain at up to €35M or 7% of global turnover.

Source: Council of the EU | Legal Nodes

Our Take: The simplification is real but narrow. The compliance window for high-risk systems shipping into the EU is now under 90 days, and most enterprise AI buyers we talk to still don't have a complete inventory of the AI systems running in production. If your organization sells into the EU and you don't know how many LLM-powered features are live across your product surface and how each is classified under the Act, that audit is now urgent — not Q3 work.

Quick Hits

• NVIDIA and Corning lock in 10x US optical manufacturing (May 6). A multiyear partnership commits Corning to 10x its US optical-connectivity capacity and 50% more US fiber, with three new advanced-manufacturing facilities in North Carolina and Texas and 3,000 new jobs. The build is sized to feed the $650-700B hyperscaler capex cycle through 2027. (Yahoo Finance)

• Greenhouse acquires Ezra AI Labs (May 5). ATS leader Greenhouse signed a definitive agreement to acquire Ezra, a voice-AI interview startup. Voice-first first-round screening becomes a default feature of the recruiting stack rather than a separate vendor. (Asanify)

• Anthropic + Google + Broadcom expand the compute partnership. Anthropic announced an expanded multi-year compute deal with Google Cloud and Broadcom — additional TPU capacity and custom-silicon roadmap commitments — to support the run-rate revenue ramp. The Google-Anthropic strategic alignment continues to deepen even as Anthropic builds an independent enterprise services arm. (Anthropic)

Numbers of the Week

What We're Watching

The consulting response. Accenture, Deloitte, McKinsey, BCG, and Bain have a quarter — maybe two — to decide whether they fight back as multi-model implementation specialists or accept being repositioned as the strategy layer above someone else's engineers. Watch for at least one of the Big Four to announce a formal multi-model practice or a strategic partnership with a non-OpenAI/non-Anthropic implementation firm by end of Q3. The ones that move fastest will likely also acquire — expect at least two specialist AI implementation firms to be acquired by Big Four buyers before September.

Whether enterprise customers tolerate single-lab lock-in. The Anthropic JV will optimize for Claude. The OpenAI Development Company will optimize for GPT-class. Buyers who watched the Pentagon-Anthropic decision and the supply-chain-risk designation it produced will be wary of betting their core workflows on a vendor whose policy stance can change procurement eligibility overnight. The defensible architecture answer is multi-model with a governance layer in the middle — which is exactly what IBM Sovereign Core and ServiceNow AI Control Tower are now positioning to be.

Project Arc as the desktop-agent reference design. ServiceNow's NVIDIA-built secure desktop agent is the first production-targeted attempt to put an agent inside the OS-level workflow layer with enterprise-grade isolation. If Arc works, the desktop becomes another runtime for agentic-AI deployments — which would meaningfully expand what enterprise IT has to govern.

This Week's Reading

• OpenAI vs Anthropic vs Google: Choosing an Enterprise LLM Provider in 2026 — Now needs an "owns implementation services" column.
• What is Agentic AI? — The runtime concept underneath every announcement this week.
• What is Conversational AI? — Why a $15B Sierra valuation is the right answer for tier-1 support.
• What is a Large Language Model? — Foundational reading before you take an EU AI Act inventory.
• Last Week's Roundup: Pentagon Picks Big Tech, Locks Out Anthropic — The procurement story that made this week's services pivot inevitable.

See you next week.