AI is Not an Operating Model

—

The AI market is beginning to communicate what many enterprise leaders are observing inside their own companies.

Access to AI is not the same as value from AI.

That is an important distinction that is becoming increasingly expensive.

Goldman Sachs estimates annual AI capital expenditure could reach $765 billion in 2026 and grow to $1.6 trillion annually by 2031, with roughly $7.6 trillion in cumulative AI infrastructure investment between 2026 and 2031. The same research expects large technology companies leading the AI buildout to spend a combined $5.3 trillion on AI and data centers from 2025 through 2030.

While that level of spending does not mean AI is a bad investment, it does mean AI is not a casual one.

The largest technology companies can absorb long investment cycles, uncertain returns, heavy infrastructure costs, and years of implementation work. Most enterprises cannot operate that way. Most do not have unlimited room to turn AI into a running experiment and wait for the economics to settle later.

That is why we assess that the AI conversation should become more practical. Leaders must assess whether their business has the operating foundation required for AI to produce value.

 

Models Are Not Always the Problem

Many companies have given their employees access to AI tools. That was the easy part. The harder part starts when those same leaders look to answer a basic question: where is the return?

MIT’s NANDA initiative found that despite $30–40 billion in enterprise GenAI investment, 95% of organizations were getting zero return, while only 5% of integrated AI pilots were extracting millions in value.

The same report found that the divide was not mainly about model quality or regulation. It was about deployment, integration, workflow fit, and whether the tools learned from the enterprise context around them.

That should get every executive’s attention. The issue is not that AI cannot do impressive things. It can. The issue is that impressive capabilities do not automatically translate into operational value.

That is where many companies are getting stuck.

 

The Market is Moving Toward Deployment Because Deployment is the Hard Part

The biggest AI vendors appear to understand this.

AWS announced a $1 billion Forward Deployed Engineering organization designed to embed thousands of experts with customers and help them co-develop agentic AI solutions. AWS says the goal is not just to build something, but to leave customers with documentation, runbooks, architecture decisions, and trained internal champions.

OpenAI launched the OpenAI Deployment Company with more than $4 billion of initial investment and approximately 150 Forward Deployed Engineers and Deployment Specialists through its planned acquisition of Tomoro.

Anthropic announced a new enterprise AI services company with Blackstone, Hellman & Friedman, and Goldman Sachs to help bring Claude into core business operations. Just this week, Anthropic and Salesforce announced a partnership.

The job market is moving in the same direction. Business Insider, citing Indeed data, reported that forward-deployed engineer (FDE) job postings were roughly 729% higher year over year in April 2026.

That is not just a hiring trend but rather a signal that AI has reached the implementation layer.

The market is learning that the hard part is not giving employees and teams access to a model. The hard part is connecting AI to the company’s existing systems, workflows, records, decisions, and exceptions that already run the business.

 

Telecom AI Has to Deal with the Real Operation

Full disclosure: For over fifteen years, our team at OneVizion has thrived in this implementation layer, serving telecom and utility operators every day and hence why we monitor this shift so closely.

Telecom does not run on clean abstractions. Telecom work moves through sites, assets, leases, permits, engineering drawings, construction milestones, vendors, materials, field crews, inspections, closeout packages, finance reviews, compliance requirements, and customer commitments.

A missed handoff does not stay inside a workflow diagram. It turns into a delayed build, a bad forecast, a customer escalation, or a field team standing in the wrong place with the wrong information.

That is the environment AI is deployed.

If a project delay is tied to a permitting issue, a vendor dependency, a missing field update, and a downstream revenue commitment, AI has to understand those relationships before anyone should trust it to recommend action.

A generic tool sitting outside the operating environment will not see enough of the picture. It may summarize a document. It may answer a question. It may automate a small administrative task.

Those uses have value but are not the same as changing how infrastructure work gets governed, executed, and measured.

 

The Risk is AI Without Context

AI can either make a strong operating model faster or make a weak operating model more confusing.

If project data is fragmented, AI is working from fragmentation. If the system of record is not trusted, AI has to learn and operate from information the business already questions. If field updates, asset records, approvals, documents, forecasts, vendors, and compliance obligations are not connected, AI may produce an answer that sounds complete but misses the operational context that matters. That is a real risk in infrastructure.

A wrong answer in a general knowledge environment is one kind of problem. A wrong answer tied to a field crew, a compliance obligation, a customer commitment, or a capital forecast is another. That is why AI cannot be treated as a replacement for the operating model. It has to be integrated into one.

 

The Hybrid Approach is the Practical One

For infrastructure companies, the right model is not AI instead of people. It is also not people continuing to reconcile disconnected systems manually forever.

The practical model is a hybrid one: structured operational data, accountable workflows, human judgment, and AI working inside systems of record that preserve the relationships between the work.

That is the approach OneVizion is focused on. The platform must still manage the structure: the data model, workflow logic, business rules, permissions, relationships, audit trails, and reporting requirements.

AI can then enable people to interact with that structure more effectively and efficiently. A project manager can see what changed. An executive can understand where risk is building. A field team can get better context before making an update. A compliance team can surface evidence without reconstructing history after the fact.

The value comes from the combination. AI is not valuable because it is impressive in a demo. It is valuable when it improves the work inside the business.

 

Buyers Should Make Self-Sufficiency Enforceable

There is real value in embedded AI services and FDEs. There is also real risk.

A vendor’s FDE team usually represents the vendor that sent it. AWS will naturally help customers build inside AWS. OpenAI will naturally help customers build around OpenAI technology. Other vendors will do the same.

That does not make the model wrong; it makes the buyer’s responsibility clearer.

Leaders should ask what they are actually gaining. Are they building durable capability inside their own operating model, or are they creating a new dependency that will be hard to unwind later?

A strong AI deployment should make the customer smarter.

It should clarify architecture, improve governance, expose weak data assumptions, document key decisions, train internal champions, and leave the customer more capable than before.

A poor deployment leaves behind a black box, a backlog, and a team that still has to call the vendor every time the business changes.

Telecom has seen that movie before.

 

AI ROI Will Come from Operational Discipline

The companies that get value from AI will not be the ones with the longest list of pilots.

They will be the ones willing to look directly at the operational friction that already costs them money.

• Where do teams re-enter the same information?
• Where do field updates disappear?
• Where does leadership get a clean report that everyone knows is incomplete?
• Where do approvals stall?
• Where does the customer feel the delay before the system shows the risk?
• Where does compliance depend on people reconstructing evidence after the fact?

That is where AI belongs first. Not because those problems are glamorous. Because they are expensive.

This is also where return is easier to prove: fewer manual reconciliations, faster exception handling, cleaner handoffs, better asset visibility, stronger audit readiness, more reliable executive reporting, and earlier identification of risk.

Those are measurable outcomes along with the discipline AI needs.

 

The Practical Test

AI will be a core component the future of infrastructure operations. But AI alone is not the future.

The near-term way ahead is not about giving every employee another tool and hoping value appears later. It is about deploying AI into the systems, workflows, records, and decisions that already run the business.

For telecom and utility leaders, the practical test is simple.

• Can AI see how the work is connected? Does it understand all the relationships between all enterprise data?
• Can AI operate from trusted records?
• Can AI support human judgment without replacing accountability? Does it meet responsible AI criteria?
• Can AI make the organization more capable after the first deployment, as opposed to more dependent?

If the answer is no, the organization is not ready to deploy AI to improve how the work gets done.

If the answer is yes, AI can become what it should be: not a standalone bet, but a practical capability inside a stronger operating model.