The AI Absorption Deficit: Why Compute Is Outrunning Enterprise Demand

Four days before OpenAI closed the largest private funding round in history, they called in McKinsey. And BCG. And Accenture. And Capgemini. Multi-year partnerships. Dedicated practice groups. Teams certified on OpenAI technology. The $110 billion headline is what everyone covered in late February. The Frontier Alliance is what the infrastructure story looks like it needs before AI can lives up to its transformational potential.

Call it the absorption deficit: the structural gap between the rate at which AI infrastructure is being deployed and the rate at which enterprise organisations can actually consume, manage, and extract value from it. It’s not a new idea. It’s just never had a $690 billion price tag attached, unlike previous eras of major computing platform shifts such as cloud and mobile.

The four largest technology companies, plus Oracle, are projecting $690 billion in infrastructure CapEx for 2026 with around 75% directed towards AI. OpenAI’s current annual revenue is approximately $20 billion, around 3% of the CapEx being spent partly to serve it (on top of losing the leading share in enterprise LLM spend to its rival, Anthropic). The entire cohort of pure-play AI vendors is expected to generate under $35 billion in 2026 revenue combined. The infrastructure is being built for a customer that, by its own data, hasn’t arrived yet.

The scale of capital going in extends well beyond technology companies. Blackstone — primarily an asset manager — has placed a series of very large bets on becoming the landlord of the AI revolution, redirecting a substantial portion of its portfolio toward AI infrastructure. Since acquiring data centre operator QTS in 2021, it has grown its leased capacity 14 times; QTS now represents more than 20% of Blackstone’s flagship real estate trust by asset value. In Australia, it is leading a $10 billion debt facility for campuses housing tens of thousands of Nvidia GB300 GPUs (Project Southgate). In New Mexico, it is navigating an $11.5 billion acquisition of the state’s primary utility, because when the constraint for AI shifts from chips to power, electricity becomes a real estate play.  

The ripple of AI capital investment into real estate and physical infrastructure including utility network, is visible and growing. Whether the returns materialise at the scale, these positions imply is still to be determined — but a company whose primary identity is property and asset management going this deep into AI infrastructure tells you where a certain class of smart money thinks the value will sit.

However, a survey of more than 120,000 enterprise respondents published by Recon Analytics this January found that 63.7% of companies have no formalised AI initiative at all. Only 8.6% have AI agents running in production. MIT’s 2025 GenAI Divide study puts the generative AI pilot failure rate at 95%, defined as projects that haven’t returned measurable ROI within six months. The infrastructure is being built for a customer that is, by most measures, still figuring out and deciding how to ‘transform’ and make use of it.

This gap will shrink in 2026 but not completely disappear. Enterprises will continue to grapple with talent deficiencies as IDC’s IT and AI workforce readiness research shows that more than 90% of global enterprises will face critical AI skills shortages this year, with $5.5 trillion in economic exposure attached to that gap. The World Economic Forum’s survey of 1,010 C-suite executives found 94% report AI-critical skill gaps, with a third reporting gaps of 40% or more in essential roles. Bain & company projects approximately 1.3 million AI-skills roles will need to be filled in the US alone by 2027, with up to half likely to remain unfilled.  

Piloting will still define most organisations’ AI journey through this year. But the scaffolding for a different reality is actively being constructed. The Frontier Alliance is part of that scaffolding — a structural acknowledgement from OpenAI that compute alone does not get deployed, and that the path from infrastructure to enterprise value runs through people and process. The enterprises that emerge from 2026 with meaningful AI traction will not be the ones that waited for the technology to mature. They will be the ones that built absorption capacity while everyone else was still evaluating vendors.

What this means if you’re building an AI strategy

The relevant question for 2026 isn’t whether the compute exists. It does, and more is coming. The question is whether your absorption capacity in the form of talent, processes, and governance your organisation can actually deploy, is keeping pace with what’s being provisioned. Most enterprise AI strategies are paced against vendor capability, not internal readiness. Those aren’t the same number. The gap between them is where the ROI and value disappears.

Runtime Check

Three points for CXOs to bring up in your next meeting:

• Measure your absorption rate, not just your AI spend. Over 90% of enterprises face critical AI skills shortages. If your AI roadmap is calibrated to vendor capability rather than your organisation’s actual capacity to absorb and operationalise, you are planning against a ceiling you haven’t measured. That ceiling is arriving faster than the talent to raise it.

• The Frontier Alliance is a signal worth reading carefully. McKinsey, BCG, Accenture, and Capgemini didn’t receive multi-year deployment partnerships because OpenAI needed better marketing. They received them because compute alone doesn’t get deployed. The professional services layer — the firms doing strategy, workflow redesign, and production rollout — is where infrastructure investment becomes enterprise value. That scaffolding is now being formalised at the model-company level. The enterprises wiring into it early will have an advantage that compounds.

• The ISV layer is the operational wrapper most strategies are underweighting. The software that manages data centre operations, monitors model performance, governs access, and integrates AI into the systems enterprises already run isn’t a commodity. It’s the layer without which the infrastructure sits idle. Gartner projects more than 80% of ISVs will embed gen AI into enterprise applications by 2026, up from less than 5% two years ago. The ecosystem bet most organisations are missing isn’t the model. It’s the tooling built around it.

The moment OpenAI called in their partners in consulting, the infrastructure story became an ecosystem story. The gap is real, and it won’t close on its own. But for the first time, it is being designed around. The question is whether your organisation is positioned to absorb what’s coming when the lights come on.

Enjoyed the read? Let’s take it further.

Connect to unlock exclusive insights, smart AI tools, and real connections that spark action.

Schedule a chat to unlock the full experience