Slowdown or Not, Enterprise AI Budgets Should Be Built on Proven ROI

This week gave enterprise AI leaders two scenarios to plan for at once. After Tuesday's midterm elections, several incoming governors and congressional leaders signaled tougher terms for new data centers, and AI stocks sold off hard before partly recovering. One camp now argues the AI buildout is about to stall. Another says the dip is noise. Our view is that enterprise AI strategy should not depend on which camp is right. The companies that come through either outcome in good shape will be the ones that fund AI on proven returns, keep their data house in order and avoid locking themselves into a single bet on cheap, abundant compute.

What actually changed this week?

The facts are still early. Voters approved restrictions in 44 of 45 completed local data center referendums, the AP reported, and governors-elect in Texas, Ohio and Wisconsin have backed moratoriums or tighter permitting. Senator Chuck Schumer said protecting consumers from data center energy costs will be a priority, according to Reuters. Goldman Sachs estimated about 45% of planned U.S. capacity could face restrictions from newly elected governors, Yahoo Finance reported, and the S&P 500 had its worst week since April 2025, the AP reported.

None of that has changed a single cloud contract yet. President Donald Trump retains the veto, and analysts disagree on how much will pass. J.P. Morgan said most capacity due in 2027 is already under construction, Techzine reported, so any slowdown would bite later.

The case for pressing ahead

The strongest argument against caution deserves a fair hearing. Demand for compute still looks strong. AMD chief executive Lisa Su told CNBC the election does not change her outlook, The National reported, and Yardeni Research described the political backlash as speed bumps rather than roadblocks, Yahoo Finance reported. If capacity does get scarcer, companies that secure it early and build skills now could gain an edge over cautious rivals. Pausing AI programs because of a stock market wobble would repeat the mistake of every company that cut digital investment in a downturn and spent years catching up.

We agree with much of that. The answer is not to stop. It is to be precise about what you are buying and why.

Why ROI discipline wins in both scenarios

If a slowdown comes, compute and model access could get more expensive. Siebert Financial's Mark Malek told Reuters the buildout is likelier to get costlier than to vanish. In that world, every workload that cannot show a return becomes a liability. If no slowdown comes and valuations correct anyway, CFOs will ask the same question for a different reason.

AI Connect speakers saw this coming. At The AI Enterprise Conference 2026, Verisk Global Chief Data Officer Louis Dimodugno warned that token prices are likely to rise once frontier model companies go public, and that organizations without visibility into AI consumption will be caught off guard. "The financial piece is the piece that's always going to get the most attention," he said. Verisk used efficiencies found in its existing data platform to pay for the tooling that found them, an approach we covered in the self-funding AI strategy.

Disciplined does not mean small. It means each use case is tied to a business metric, as panelists from The Home Depot and Equifax described in how to tie every AI use case to an OKR, and each has a known cost per outcome.

How should leaders manage compute risk?

Right-size first. "You don't need premium models for everything that you're doing," Pushkar Rege, Vice President and Co-Founder of the Business Intelligence Program at Mastercard, said at The AI Enterprise Conference 2026, in a panel on AI cost per outcome. Smaller models and better routing reduce exposure to whatever happens to GPU prices.

Then ask providers harder questions. Where is the capacity behind your 2027 and 2028 commitments being built? What happens to your pricing if grid upgrade costs shift onto developers, as some lawmakers and analysts expect? Can workloads move between regions or providers without a rebuild? Contract terms and architecture choices made now are cheaper than migrations made under pressure.

Why data readiness is the safest bet

Whatever happens to data centers, the work that makes AI useful inside a company is mostly not about compute. Clean, well-governed, well-described data makes smaller models more accurate, reduces wasted tokens and shortens the path from pilot to production. That work keeps its value if compute gets scarce, and it pays off faster if compute stays cheap.

It also keeps the focus on impact rather than activity. "Adoption without impact is activity," Kanchan Soman, VP of AI, Data & Analytics at Crum & Forster, said at The AI Enterprise Conference 2026, describing how the insurer scaled AI adoption. A market correction is a good moment to apply that test to every line in the AI budget.

Enterprise leaders cannot control permitting fights in Austin or Columbus, the votes in Congress or Nvidia's share price. They can control which projects they fund, how they measure them and how dependent they are on any one supplier's expansion plans. A portfolio built on proven returns and ready data does not need the boom to continue, and it does not need a crash to justify itself. That is the position we would want to be in, whichever way the next year breaks.

Frequently asked questions

Should enterprises pause AI investment because of a possible data center slowdown?

We argue no. Instead, fund use cases with measured returns and known cost per outcome, and avoid depending on a single provider's expansion plans.

How could a data center slowdown affect enterprise AI costs?

If new capacity becomes scarcer or costlier, compute and model access prices could rise, making low-return workloads harder to justify.

What is the safest AI investment in an uncertain market?

Data readiness: well-governed, well-described data improves accuracy with smaller models and pays off whether compute is scarce or cheap.

Opinion pieces reflect the views of The AI Enterprise Magazine Team.

Filed under: AI Strategy & ROI, Data & AI Infrastructure