AI Valuations Don't Add Up. What Does That Mean for Your Portfolio?
Enterprises don't refresh their tech stack every six weeks, but AI valuations assume they do — and that mismatch could cause trouble. Here's how investors can get ahead.
Editor's note: This is the fourth and final article in a series examining AI concentration risk in growth portfolios. The first, Your Diversified ETF Isn't as Diversified as You Think, examined how popular ETFs concentrate exposure to the AI capital spending cycle; the second, AI Is a Supply Chain Rather Than an Industry: Here's Why, mapped the supply chain layers behind that spending; the third, How to Manage the 4 Risks Hiding Inside Every AI-Heavy Portfolio, examined the risks facing the buildout itself.
A new frontier AI model now reaches the market every few weeks. No hospital system, manufacturer, bank or retailer refreshes its production technology stack on that schedule, and none ever will.
That mismatch is not a minor operational footnote. It is a structural risk sitting underneath the valuation of nearly every company in the AI supply chain, because those valuations assume enterprise revenue will arrive on the same clock as model releases. It won't.
From just $107.88 $24.99 for Kiplinger Personal Finance
Become a smarter, better informed investor. Subscribe from just $107.88 $24.99, plus get up to 4 Special Issues
Sign up for Kiplinger’s Free Newsletters
Profit and prosper with the best of expert advice on investing, taxes, retirement, personal finance and more - straight to your e-mail.
Profit and prosper with the best of expert advice - straight to your e-mail.
Two clocks running at different speeds
Model providers compete on speed. The frontier model release cadence compressed dramatically through 2025 and into 2026: Labs that once shipped a major model roughly every six months moved to a pace of multiple substantive releases per week by early 2026, with one industry tracker counting more than a dozen meaningful frontier launches in the first quarter alone.
A separate count found five major model launches or availability milestones in a single 16-day window in July 2026.
Enterprise procurement runs on an entirely different clock, and a much slower one. A 2026 analysis of AI vendor contracts found legal review cycles for AI-specific agreements now averaging close to 15 weeks, more than triple the review time for a traditional Software-as-a-Service (SaaS) contract just a few years earlier.
Add security review, compliance sign-off and a multi-stakeholder approval process, and a real-world deployment at a large enterprise, in banking, healthcare, manufacturing or retail for example, can run nine to 10 months from initial evaluation to signed contract, even when the vendor already holds standard security certifications.
About Adviser Intel
The author of this article is a participant in Kiplinger's Adviser Intel program, a curated network of trusted financial professionals who share expert insights on wealth building and preservation. Contributors, including fiduciary financial planners, wealth managers, CEOs and attorneys, provide actionable advice about retirement planning, estate planning, tax strategies and more. Experts are invited to contribute and do not pay to be included, so you can trust their advice is honest and valuable.
Why enterprises move slowly, and why that's the correct speed
This is not enterprises being slow-footed or resistant to change. It shows up across every regulated or safety-conscious industry, each carrying its own version of the same brake.
- A bank or broker-dealer answers to examiners over data governance and vendor oversight
- A hospital system answers to HIPAA auditors over how patient data touches a new tool
- A drugmaker answers to the FDA over validating any system that touches a clinical trial or a manufacturing line
- A manufacturer answers to safety certification bodies before a new system gets anywhere near a production floor
None of these processes moves on a six-week cycle, and none of them is optional for the companies running them.
Consumer-facing companies feel a lighter version of the same friction. A retailer integrating an AI tool into its customer service or logistics stack still has to test it against real order volume, train staff and confirm it will not break during a peak selling season before betting the business on it. The specific obstacle changes by industry. The pace does not.
Deloitte's 2026 State of AI in the Enterprise report captures the resulting gap directly. Worker access to AI tools across enterprises rose 50% in 2025, yet the number of companies with at least 40% of their AI projects running in production is only on track to double within the next six months, meaning most organizations are still well short of that mark.
Most of that gap is not a technology problem. It is a governance and procurement problem, and it will not close on the timeline the model providers are shipping against.
The assumption this breaks in the supply chain
Every layer of the AI supply chain, from chip design through data center construction to hyperscaler capital spending, is financed against an assumption about how quickly enterprise customers will adopt and pay for AI capability.
The $700 billion to $725 billion in combined 2026 capital expenditure from Microsoft, Amazon, Alphabet and Meta depends on enterprise revenue materializing on a timeline that matches the speed of the infrastructure being built.
If enterprises take 12 to 24 months to move a single AI deployment from evaluation to production, and model providers ship a new frontier model every few weeks in the meantime, the two clocks are not just out of sync. They are structurally incompatible.
Infrastructure gets built for a demand curve that assumes fast adoption. Actual revenue arrives on a curve shaped by compliance review, legal sign-off and board-level risk tolerance.
Where this shows up first
The companies most exposed to this mismatch are not the chip designers or the data center builders. Chip and infrastructure companies get paid when hyperscalers buy equipment, regardless of how quickly that capacity actually gets sold to enterprise customers.
The exposure lands hardest on the enterprise software companies selling AI-enabled products directly into regulated industries, since they depend on enterprise contracts to convert infrastructure spending into revenue.
A vendor whose entire pitch rests on rapid iteration and constant improvement is marketing directly against the clock its actual customers are forced to run on.
Fast model refresh is a genuine advantage for a consumer application competing on user experience. It is closer to a liability for a vendor selling into a compliance function that has to document, test and defend every material change to a production system before it goes live.
Looking for expert tips to grow and preserve your wealth? Sign up for Adviser Intel, our free, twice-weekly newsletter.
What this means for your portfolio
This risk will not show up as a single dramatic event. It shows up as a slow gap between guided revenue and delivered revenue at the enterprise software layer, quarter after quarter, while the infrastructure layers beneath it keep reporting strong numbers because hyperscalers keep buying equipment regardless of how fast their customers can absorb it.
Watch the enterprise adoption data as closely as you watch the capital spending headlines. A hyperscaler raising its 2027 capex guidance tells you demand for infrastructure remains strong. It tells you nothing about whether the enterprises meant to eventually pay for that infrastructure can move fast enough to justify it.
Three places to look before the next earnings cycle.
1. Read the backlog language, not just the revenue line, for any enterprise software company you own that sells AI features into banking, healthcare or manufacturing. A growing backlog of signed but not-yet-recognized contracts tells you demand exists but hasn't cleared the procurement gauntlet yet.
A shrinking backlog next to flat AI revenue is the first sign the 12-to-24-month sales cycle is running longer than guided.
2. Compare guidance language quarter over quarter for the same company. If a company guided to AI revenue "accelerating in the back half" two quarters running, ask why the acceleration keeps getting pushed out. Procurement timelines don't compress. Guidance that assumes they will is the tell.
3. Separate your infrastructure exposure from your enterprise software exposure and size them differently. The chip and data center companies get paid on the hyperscaler capex cycle regardless of how fast banks and hospitals adopt. The enterprise software companies only get paid when adoption actually happens.
If both sit in your portfolio at equal weight, you are pricing them on the same clock. They are not on the same clock.
The market is pricing AI companies on an innovation cycle measured in weeks. Most of their actual customers are moving on a cycle measured in years, and that gap does not close because a new model shipped.
If a correction comes, it is unlikely to be triggered by AI failing to work. It is far more likely to be triggered by the market finally pricing in how long enterprises actually take to pay for it.
Related Content
- 10 Major AI Companies You Should Know
- The Risks of an AI Bubble
- What Tech Leaders Have Said About an AI Bubble
- Will AI Pay Dividends for Your Firm? To Find Out, Budget for the Whole Iceberg, Not Just the Tip
- How Businesses (and Advisers) Can Budget for AI Use When the Bill Keeps Changing
Join over half a million readers using Kiplinger's insights to make smart financial decisions. Profit and prosper with our expert guidance on investing, taxes and retirement, and more. Delivered daily.
John O'Connell is founder and CEO of The Oasis Group, an award-winning consultancy and research firm serving wealth management firms nationwide. O'Connell has more than 30 years of leadership experience in financial technology and wealth management, including North American leadership at Oracle, fintech CEO and president roles and participation in IPO and M&A transactions. He is the creator of the AI WealthTech Map (100+ firms), the developer of the Oasis AI Readiness Index and is recognized as a leading independent voice on AI adoption in wealth management.