The Pace of Progress and the Price of Expectations

Today’s market is walking a tightrope, attempting to balance conflicting expectations and mismatched timelines. As sentiment turns toward reality, market pricing will also likely change, leaving those too anchored to AI momentum in a perilous position. The underlying risk isn't just enthusiasm around AI—it's the timeline disconnect between immediate, hyped valuations and the longer horizon needed for real economic returns.

Investors are paying premium prices for AI infrastructure related companies, bidding up hardware and data centers as if every corporation in America will be fully automated using AI within a very short period of time. On the other side, investors are discounting traditional enterprise software, pricing those companies as though AI will bypass existing applications overnight and render them obsolete. Both of these views share a single, flawed assumption: that large organizations can redesign their core operations at lightning speed. History—and plain common sense—tells us that big companies simply don't move that fast.

The central question surrounding artificial intelligence is no longer whether the technology works. The real question for investors is how quickly this technological capability translates into broad, repeatable, and profitable business demand. That’s a question that very few are able to answer with clear supporting evidence.

Capability is Not Adoption

Most people using AI are using it to improve personal productivity, ourselves included. However It’s easy to mistake personal convenience for organizational transformation. Handing an employee an AI assistant to help draft an email or write a piece of code takes very little time. However, handing AI agents the keys to a company's mission-critical business process is a whole different ballgame.

McKinsey’s 2026 global survey illustrates this gap clearly. Nearly nine out of ten companies report using AI in at least one function, and 80% of workers say it helps them get things done faster. Yet only 44% report scaling AI across the broader enterprise, a mere 37% can point to any bottom-line EBIT impact, and just 6% qualify as high performers capturing meaningful financial gains. Numerous other case studies, including one from OpenAI, show similar results.

The Unglamorous Side of the Story

Why is it taking so long for all these incredible AI models to actually boost corporate bottom lines? The short answer is that big companies run on deeply entrenched systems and processes that, while dated, are exceptionally good at keeping the business running. That creates a massive disincentive to change whenever a flashy new technology comes along.

For starters, you can’t just tell a computer to go "fix customer service." You have to map out every single step, plan for edge cases, and figure out exactly where a human still needs to sign off. On top of that, enterprise data is usually a mess—scattered across legacy databases, buried in spreadsheets, or trapped in email threads. Feeding messy data to AI just makes it spit out mistakes faster. Even when an AI tool makes the right call—like deciding a customer deserves a refund—it still needs a secure, compliant path into the accounting software to actually transfer the money.

Then there’s the issue of reliability. A model that’s right 90% of the time is great for kicking around ideas, but it’s completely unusable for payroll, tax filings, or healthcare, where mistakes trigger severe financial or legal fallout. At the same time, the economics can be tough to justify. Token fees, software licenses, and integration costs add up quickly, squeezing returns before a project even gets off the ground. And even if you solve the technical and financial side, human habits change slowly. Updating software takes seconds, but getting employees to trust automated tools and change how they work requires real effort and time.

Businesses adopt technology, one messy workflow at a time, and that step-by-step reality is what truly dictates the pace.

.Why Enterprise Software Isn't Going Anywhere

There is a popular narrative that AI agents will render traditional software obsolete. In mission-critical environments, we think the opposite is closer to the truth.

AI models supply the raw intelligence, but enterprise software supplies the context—the rules, permissions, customer histories, and audit trails that allow management to trust the result. If you are running a low-stakes task, a standalone AI tool is fine. But for core processes governing money, customer records, or operations, trusted context matters far more than an increase in model horsepower.

We are already seeing this play out. In the second quarter of 2026, ServiceNow, a software IT service provider to large enterprise companies, reported that its AI offerings crossed $1 billion in annual contract value, with subscription revenue growing over 24%. Customers weren't bypassing ServiceNow to implement their own AI; they were using ServiceNow to make AI safe and operational.

Rather than being displaced, well-entrenched software companies sit in an enviable spot: they can incorporate better intelligence into their existing products without having to spend tens of billions of dollars building the underlying models themselves. This makes their offerings even more valuable to the enterprise companies they serve.

Two Different Timelines

This brings us back to the market's current expectation gap. Estimates are that total AI related capital expenditures will surpass $1 trillion USD in 2027. Massive sums of capital are being poured into building AI infrastructure on the assumption that enterprise demand will compound fast enough to absorb all that new capacity almost immediately. Keep in mind that this is a still somewhat unproven demand story in that only a small slice of global businesses are truly trying to adopt the technology into their workflows. The rest are just experimenting with or watching from the sidelines.

The risk for investors isn't that AI will fail in a binary sense. The technology is real and will create value over time (although it is still unclear how much value and to whom that value will accrue). The risk is that enterprise adoption continues along a normal, multi-year adoption curve while the infrastructure picks and shovels are priced for instant transformation. This appears to be the reality of the situation causing many securities to be mispriced (i.e.significantly overvalued or undervalued).

In the end, we expect a long, uneven transition. Simple, language-heavy tasks like basic coding or research will continue to automate quickly. Complex, multi-system workflows involving sensitive data and regulatory oversight will move much more deliberately. So far the evidence strongly supports this exact scenario and we are investing accordingly–exploiting the undervalued prices of great businesses.

For long-term investors, the pace of adoption and change matters just as much as the destination. A great technology can still make for a disappointing investment if you pay a price that demands the future arrive all at once. Conversely, durable companies that quietly do the heavy lifting to turn exciting new technology into real business utility often yield the best returns of all.

Sources

McKinsey & Company, The State of AI in 2026: On the Road to ROI, August 25, 2026.

OpenAI, The State of Enterprise AI 2025, December 17, 2025.

IBM, The Biggest AI Adoption Challenges for 2026.

Information Services Group, State of Enterprise AI Adoption Report 2025.