Who Owns Enterprise AI in 2030? Three ‘Anchor Deals’ Boards Can’t Afford to Ignore

22 May 2026 | AIMG

Enterprise AI is no longer being defined by model launches and chatbot demos, but by a handful of architecture‑setting partnerships that determine who controls physical operations, agent orchestration, and the underlying compute grid.

Hitachi–Anthropic, ServiceNow–NVIDIA, and Project Stargate look less like isolated announcements and more like a coordinated blueprint for how AI will actually run critical industries, enterprise workflows, and sovereign infrastructure by 2030.

1. Three anchor deals, one emerging architecture

Hitachi is using Anthropic’s Claude inside Lumada 3.0 and HMAX to move from “digital AI” into physical AI that directly touches energy grids, rail networks, and frontline operations, targeting a total addressable market of roughly USD 640 billion by 2030 (converted from over JPY 100 trillion).
ServiceNow and NVIDIA are using Project Arc to turn copilots into governed, long‑running autonomous agents, secured by NVIDIA’s OpenShell and orchestrated by ServiceNow’s AI Control Tower, with Now Assist on track for about USD 1.5 billion in AI annual contract value (ACV) by the end of 2026.
Project Stargate is assembling a USD 500 billion, 10‑gigawatt global compute footprint with OpenAI as anchor tenant, SoftBank/MGX as financial architects, and Oracle as infrastructure operator, shifting AI build‑out from single‑vendor clouds to multi‑party consortiums.

2. What this signals for enterprise AI by 2030

Physical AI becomes the default in asset‑heavy industries: Hitachi’s “Customer Zero” strategy (290,000 employees using Claude, 100,000 AI professionals trained) is a template for how large incumbents will industrialize AI before selling it to their own customers.
An enterprise “agent OS” beats standalone copilots: ServiceNow’s open, token‑based Action Fabric lets enterprises plug in any model or agent while keeping deterministic execution and governance in the Now platform, positioning it as the “AI of AIs” rather than just another SaaS app.
The economics of SaaS flip from seats to tokens: early data shows Now Assist customers expanding ACV by 3x and a projected ~30% of total ACV from AI by 2030 (around USD 9–10 billion of ACV if ServiceNow reaches USD 30 billion total), but AIMG flags a key risk that token growth may not fully offset seat compression if agent adoption under‑delivers.

3. What this signals for sovereign AI and infrastructure

Stargate is the reference design for sovereign AI infrastructure: 5 GW in the UAE, 500 MW in Argentina, and 520 MW in Norway illustrate how data residency, geopolitics, and stranded renewables are now primary siting variables for AI data centers.
Financing is becoming “Private Credit 2.0”: structures like OpenAI’s USD 38 billion Project Jacquard use compute‑contract‑backed delayed‑draw term loans and chip‑leasing special purpose vehicles (SPVs) to bridge the capex–revenue gap, pulling banks, private credit, and sovereign wealth funds directly into AI infrastructure risk.
Agentic AI is rewiring the hardware mix: moving from batch inference to continuous orchestration roughly quadruples CPU demand from about 30 million to about 120 million cores per gigawatt and compresses the CPU:GPU ratio toward 1:2, creating a quiet CPU renaissance that benefits Intel, AMD EPYC, and ARM‑based servers.

4. AIMG view: how this reshapes power in enterprise AI

These three deals effectively define the control points of the 2030 stack: physical AI at the edge (Hitachi), governed agent orchestration in the enterprise core (ServiceNow), and capital‑intensive, sovereign‑aligned compute at the base layer (Stargate/Oracle).
Power is consolidating into an oligopoly of architecture owners: incumbents that can combine operational domain expertise, agent‑safe control planes, and multi‑hundred‑billion‑dollar financing networks will set de facto standards that are very hard for pure‑play model vendors or startups to dislodge.
The dominant strategic question for boards over 2026–2030 shifts from “Which model do we choose?” to “Which architecture do we want to be locked into — and at which layer of this new AI stack do we intend to compete?”.

Source: AIMG Research