This Week in AI & Automation

Week of April 27 – May 2, 2026

This was the week AI vendor selection turned into a procurement-with-stakes decision. The Pentagon signed classified-network AI deals with eight companies — SpaceX, OpenAI, Google, NVIDIA, Microsoft, AWS, Reflection, and Oracle — and excluded Anthropic by name, citing a "supply chain risk" designation that until now had been reserved for vendors tied to foreign adversaries. Two days earlier, Alphabet, Microsoft, Meta, and Amazon all reported Q1 2026 earnings on the same day and collectively raised their 2026 AI capex guidance to $650-700 billion, the largest concentrated infrastructure cycle in tech history. Microsoft 365 E7 — the first new enterprise license tier since E5 launched in 2015 — ships May 1 with Agent 365 inside it. NVIDIA dropped Nemotron 3 Nano Omni, a 30B-parameter (3B active) multimodal model with 9x throughput on video and document workloads. And GE Appliances confirmed 800 AI agents in production across manufacturing, logistics, and supply chain. The "pick a stack" moment for enterprise AI buyers has arrived.

The Big Story

Pentagon Signs Classified-Network AI Deals With 8 Vendors — Anthropic Excluded by Name

On May 1, the Department of War announced production AI deals with SpaceX, OpenAI, Google, NVIDIA, Microsoft, AWS, Reflection, and Oracle for use across classified networks. Anthropic — which has the strongest enterprise AI revenue trajectory in the market — was excluded by name. The Pentagon designated the company a "supply chain risk," a label historically applied only to vendors with foreign-adversary ties. The trigger: Anthropic refused to amend its usage policy to allow Claude to be used for "all lawful purposes," including autonomous weapons and mass domestic surveillance. Every other vendor on the list signed terms that the Pentagon was satisfied with.

Anthropic sued the Trump administration earlier in the dispute. A federal judge in California issued a preliminary injunction blocking parts of the government's action, but a D.C. appeals court let the supply-chain designation itself stand — which is the part that locks Anthropic out of defense procurement. OpenAI publicly stated it will not allow its tools to be used for mass domestic surveillance or autonomous weapons, language that echoes Anthropic's earlier position but which the Pentagon evidently accepted from OpenAI and rejected from Anthropic. The disparity in treatment is now itself a procurement signal.

Source: CNN Business | DefenseScoop | SiliconANGLE

Our Take: This is the first time a top-three model vendor has been blocked from a major U.S. government procurement track on safety-policy grounds. For enterprise procurement teams, it means two things. First, the "Anthropic is the safe enterprise default" narrative — which had been hardening since the $40B Google deal closed last week — now has a real asterisk in regulated industries. Second, the Pentagon just demonstrated that vendor-policy alignment is a durable selection axis, not a fringe concern. CIOs at financial institutions, healthcare systems, and critical-infrastructure operators should expect their compliance teams to start asking the same questions Defense did. The procurement question shifts from "which model is best?" to "whose policy stance survives our regulators?" — and that question doesn't have a single answer across industries. Multi-model architectures just got more defensible, not less.

Notable Developments

Hyperscaler Q1 Earnings: 2026 AI Capex Tracking $650-700 Billion

On April 29, Alphabet, Microsoft, Meta, and Amazon reported Q1 2026 earnings on the same day and all four raised AI capex guidance. Alphabet posted $62.6B in net income (up 81% year-over-year), with Google Cloud revenue up 63% and crossing $20B in quarterly revenue for the first time. Alphabet raised 2026 capex guidance to $180-190B and warned 2027 will go "significantly higher." Microsoft guided to roughly $190B for 2026 with Q4 alone exceeding $40B, two-thirds of it on GPUs and CPUs. Amazon's AWS hit $37.6B in quarterly revenue, growing 28% — its fastest pace in 15 quarters — and Amazon projected $200B in 2026 capex. Meta raised its range to $125-145B and saw shares drop 6% on the announcement.

Source: CNBC — Alphabet earnings | Fortune | Benzinga

Our Take: AWS's 28% growth — its fastest in nearly four years — is the metric that matters most. Cloud growth had been decelerating across all three hyperscalers for the better part of two years; AI workloads have visibly reversed it. The flip side is that AWS is still capacity-constrained. So is Azure (Microsoft has flagged supply tightness for three consecutive quarters). The $650-700B annual capex isn't speculative buildout — it's a response to demand that already exists and isn't being met. For enterprise AI buyers, the operational read is simple: factor compute availability into your 2026 deployment plans. The model you want to run may not have the inference capacity you need on the cloud you've standardized on.

Microsoft 365 E7 and Agent 365 Ship May 1 — First New Enterprise Tier Since 2015

May 1 was the general-availability date for Microsoft 365 E7, the first new enterprise license tier Microsoft has launched since E5 shipped in 2015. List price is $99 per user per month, and the bundle includes Office 365 E5, Enterprise Mobility + Security E5, Windows 11 Enterprise E5, Microsoft 365 Copilot ($30 standalone), Microsoft Entra Suite ($12 standalone), and the new Microsoft Agent 365 ($15 standalone). Bought separately, the components run $117/user/month, so E7 is a 15% bundle discount. Agent 365 — also available as a standalone add-on starting May 1 — is Microsoft's new control plane for governing AI agents at enterprise scale: identity, access policy, audit, and lifecycle management for agents the same way Entra handles those primitives for human users.

Source: Directions on Microsoft | SAMexpert | Trustmarque

Our Take: Agent 365 is the consequential piece, not the bundle math. Microsoft has just declared that AI agents are first-class identities that need their own governance plane — separate from but parallel to human identity in Entra. Every enterprise that has been quietly piloting agents in shadow IT now has an officially supported track to bring them under policy. The pricing matters too: at $15/user/month standalone, Agent 365 is the cheapest path to credentialed, audited agent operations for any Microsoft-shop CIO who's been told by legal that ungoverned agents aren't acceptable. Expect every other identity vendor (Okta, Ping, ForgeRock) to ship a competitive answer within two quarters.

NVIDIA Drops Nemotron 3 Nano Omni — Multimodal at the Edge

On April 28, NVIDIA released Nemotron 3 Nano Omni, an open multimodal model that unifies vision, audio, video, and text in a single architecture. The headline number: 30B parameters total, only 3B active per token (a hybrid mixture-of-experts), with a 256K context window and up to 9x higher throughput than other open omni models on video and document workloads. The model tops six leaderboards for document intelligence, video understanding, and audio understanding. It is available on Hugging Face, OpenRouter, build.nvidia.com, and 25+ partner platforms. Early adopters include Foxconn, Palantir, Eka Care, Aible, and Applied Scientific Intelligence.

Source: NVIDIA Blog | SiliconANGLE | Hugging Face

Our Take: The 3B-active footprint is the operations-relevant detail. A model that handles vision, speech, and document parsing in a single forward pass — and runs on the kind of edge hardware that already sits in factories, distribution centers, and contact centers — collapses the cost stack for computer vision plus speech plus retrieval workflows that currently require three stitched pipelines. For Operations AI deployments specifically, this is the model class that lets you stop running two different inference clusters on the factory floor.

GE Appliances Puts 800 AI Agents Into Production on Gemini Enterprise

GE Appliances confirmed it has deployed more than 800 AI agents across its manufacturing, logistics, and supply-chain operations, all running on Google's Gemini Enterprise platform. The agents handle production-line decisioning, inventory routing, and quality inspection workflows. Separately at Hannover Messe in late April, Lenovo and NVIDIA showed an AI deployment at Lenovo's largest North American manufacturing site that delivered 85% faster lead times, 42% lower logistics costs, and 58% higher productivity. Terex, an industrial-equipment manufacturer, is projecting a 3% yield increase and 10% rework reduction from its agent deployment. Across the Hannover Messe survey base, 94% of manufacturers said they will increase AI investment in 2026.

Source: Google Cloud Press | NVIDIA Blog — Hannover Messe | Lenovo

Our Take: 800 agents in production is past the pilot threshold and into the "this is how the company operates" range. The interesting design choice is that GE didn't centralize the agents — they distributed them across function-specific workflows where the human operators already had decision-rights. That is the deployment pattern that actually scales on a factory floor. The single-site, three-figure productivity gain at Lenovo is a useful counterweight to the "AI ROI is fuzzy" narrative — when AI is deployed inside an operations workflow with clean baseline data, the lift is measurable in weeks, not quarters.

Quick Hits

• Anthropic's Project Deal results published (Apr 24-27). 69 Anthropic employees ran AI agents in a real marketplace; agents struck 186 deals worth over $4,000. The non-obvious finding: Opus-represented users earned $2.68 more per item sold and saved $2.45 more per item bought versus Haiku-represented users — but rated their deal fairness identically. Model quality compounds in agent-on-agent commerce, and humans cannot detect the gap. (TechCrunch | Anthropic)

• AI funding stayed hot through end of April. Aidoc closed $150M Series E led by Goldman Sachs Growth (radiology AI, 2,000 hospitals, 60M imaging cases/year). Manifest OS raised $60M Series A at $750M for legal AI. SPREAD AI raised $30M Series B for industrial-engineering AI. Scout AI raised $100M Series A for unmanned-warfare AI. (Tech Startups)

• Half of Google's and Amazon's "blowout AI profits" came from their Anthropic stakes. Fortune's analysis of the Q1 earnings beats showed that mark-to-market gains on the Anthropic equity stake — not core operating performance — drove a meaningful share of the upside. The Anthropic valuation cycle is now feeding the hyperscalers' P&L directly. (Fortune)

Numbers of the Week

What We're Watching

The Anthropic policy stance as a procurement axis. The Pentagon decision will not stay isolated. Expect at least one Fortune 500 financial institution and one healthcare system to formally evaluate whether Anthropic's usage policies create regulatory friction for them. The answer for most will be "no, our use case is fine" — but the question itself is now on the table, and the vendor-comparison framework most procurement teams are using doesn't yet have a column for "policy alignment with our regulators."

Whether $700B in capex is matched by inference demand. Two of the four hyperscalers (Microsoft, AWS) explicitly said they remain supply-constrained. Meta's stock dropped 6% because investors are not yet convinced its capex maps to revenue the way Google's does. The next two earnings cycles (Q2 in late July, Q3 in late October) will tell us whether enterprise AI inference workloads are filling the new capacity as fast as it ships. If they are, expect another round of capex guidance increases. If they aren't, expect Meta to take the cuts first.

Agent governance as the next platform war. Microsoft just made Agent 365 a $15/user/month line item. Google has the Gemini Enterprise Agent Platform. AWS has Bedrock AgentCore. Expect identity vendors (Okta, Ping) and observability vendors (Datadog, Splunk) to ship agent-governance products by end of Q3, and expect a wave of agent-specific compliance frameworks — agent inventories, agent audit logs, agent-to-agent trust policies — to start appearing in enterprise RFPs.

This Week's Reading

• What is Multimodal AI? — The architectural shift Nemotron 3 Nano Omni represents.
• What is a Vector Database? — Why edge-deployable multimodal models still need durable retrieval.
• What is Agentic AI? — The concept underneath Agent 365 and the Pentagon procurement debate.
• OpenAI vs Anthropic vs Google: Choosing an Enterprise LLM Provider in 2026 — Updated this week; now needs a procurement-policy column.
• Last Week's Roundup: Google Bets $40B on Anthropic — Worth re-reading in light of this week's Pentagon decision.

See you next week.