If AI stocks are in a bubble, why hasn’t the real economy caught up yet?

The data says only one in four mid-market firms has fully deployed AI however stock valuations price in full-scale transformation. Four potential futures explain the gap: Pilot Purgatory, Rapid Deployment, The Measured Path, and Sectoral Divergence.

Like many of you I have bought a few US Tech stocks over the last 5 years. The constant bullish market has been slightly discordant to what we are seeing when we talk with our clients who are in the trenches investigating how top make AI work for their businesses though. Some are streaks ahead ( the minority), many are testing but most have not meaningfully started. Our own firm is currently in the phase of focusing on governance, ensuring data privacy, cybersecurity and cleaning, organising and anonymising data. This is a long and frustrating phase for us!

Bears scream bubble. 91% experimenting, 25% integrated , yet AI stocks trade at valuations that assume full-scale deployment. The data and the market can't both be right. Here are four potential futures that could explain the gap: stall-out, acceleration, grind, or sectoral split.

"AI is the next dot-com bubble," they say. The 2025 RSM Middle Market AI Survey shows 91% of firms dabbling in AI, but only 25% with full integration into core operations. US Census data shows 3% to 4% adoption for small firms with 5 to 99 employees. Yet AI infrastructure stocks trade at revenue multiples of 18 to 30 times forward sales. Goldman Sachs warns that if hyperscalers cut capital expenditure back to 2022 levels, the S&P 500 could face a 15%-20% downside. The deployment data doesn't match the hype.

Future scenario one: Pilot purgatory - AI as theatre

The majority experiment but never operationalise - value creation stalls before it starts.

Companies stay trapped in experimentation. The leap to production proves too difficult for most mid-market businesses. Adoption plateaus around current levels. AI becomes another enterprise software category where 20% of companies capture 80% of the value.

Integration barriers are structural. Mid-market firms lack the data infrastructure, technical talent, and organisational capability to embed AI into mission-critical workflows. Data quality issues take 18 months to fix. Model drift requires ongoing monitoring. Integration with legacy systems creates fragility. Regulatory uncertainty grows rather than shrinks. Risk committees decide the juice isn't worth the squeeze.

If mid-market deployment stalls at 25% to 30% full integration, the total addressable market contracts by two-thirds. Revenue growth for AI vendors drops from 40% to 50% annually down to 15%–20%. AI stocks fall 60% to 70% from current levels as markets reprice for enterprise software economics rather than transformational growth.

The counter-argument is competitive pressure. Companies deploying AI effectively gain measurable cost advantages. Their competitors can't ignore that. The technology is also getting easier to use. Five years ago, deploying machine learning required PhD-level expertise. Today, pre-trained models lower the skill barrier substantially. AI delivers value incrementally, unlike ERP systems that required massive upfront investment before any payback.

Future scenario two: Rapid deployment - The everything-right bet

The bull case assumes momentum compounds faster than friction.

Best practices emerge and propagate quickly. The gap between pilot and production shrinks from 18 months to 6 months. Mid-market adoption accelerates from 25% full integration today to 60% to 70% within three years.

Network effects are real. As more companies deploy AI, more integration partners develop expertise. The ecosystem effect drives down implementation costs. What took 12 months in 2024 takes 4 months in 2027. Model capabilities continue improving. Platform companies are investing billions in making deployment easier. Microsoft, Google, and Amazon recognise that their cloud growth depends on customers actually using AI at scale.

This scenario justifies current valuations and supports further multiple expansion. Revenue growth for AI vendors remains at 40% plus for the next five years. AI stocks rally another 40% to 60% as revenue beats expectations for eight consecutive quarters over the same period.

The problem is this requires almost everything going right simultaneously. Tools need to improve faster than historical patterns suggest. Best practices need to propagate across industries at unprecedented speed. Regulatory frameworks need to emerge coherently rather than fragmenting by jurisdiction.

More critically, organisational change management doesn't compress on Moore's Law timelines. Companies still need to retrain staff, redesign processes, and manage the political dynamics of automation. Those constraints are human, not technical. Competition will also intensify as adoption grows. The 70% to 80% gross margins that AI application companies enjoy today will compress towards 50% to 60% as markets mature.

Future scenario three: The measured path - the long grind

Progress happens slower than bulls expect and faster than bears fear.

Integration moves from 25% full integration to 45% to 50% over five years rather than three. Growth is real but measured.

This fits how enterprise technology adoption actually works. Companies adopt transformative technologies on their own timelines, driven by budget cycles, organisational readiness, and competitive pressure. That's measured in years, not quarters. We're 18 months into the generative AI wave. Only 25% of mid-market firms have reached full integration - a pace that's reasonably fast for enterprise tech.

This creates a painful middle period for AI stocks. Valuations were priced for acceleration. As adoption follows a more gradual path, multiples compress. Companies trading at high revenue multiples drop to more moderate valuations. AI stocks decline 30% to 40% over 18 months as growth rates decelerate from 50% to 30%, then consolidate sideways before beginning a multi-year recovery.

Then, as deployment reaches critical mass, the narrative shifts. Analysts recognise that 50% penetration with real P&L impact is more valuable than 90% pilot adoption with minimal business impact. Valuations recover based on actual earnings power rather than forward projections.

Future scenario four: Sectoral divergence - winners by industry

Adoption races ahead in some sectors and stalls in others. The variance matters more than the average.

Tech companies, retailers, and professional services reach 60% to 70% integration within three years. Healthcare, financial services, and manufacturing plateau at 15% to 20%. The aggregate numbers resemble the "grind" scenario, but the divergence defines the story.

The data already shows this. E-commerce companies report 10% to 15% conversion gains from AI recommendation engines. Software development teams using AI coding assistants ship 30% to 40% more features per sprint. Customer service operations cut costs 20% to 25% through AI-first workflows.

Contrast that with regulated sectors. Healthcare providers face complexity adding 12–18 months to any AI rollout. Banks operate under frameworks written before generative AI existed. Manufacturing firms battle decades-old systems that resist modern integration.

This creates divergence in the investment landscape. Vendors focused on high-adoption sectors see 50% plus growth and justify premium valuations. Horizontal platforms serving all industries see 20% to 25% blended growth, disappointing versus expectations. AI stocks split sharply: vertical pure plays outperform indices by 80% to 120%; horizontal platforms underperform by 25% to 35%. Investors must pick winners by industry, not just by theme.

My current verdict

The sectoral split wins in the near term. The long grind wins over five years.

Deployment data already shows massive variance. Retail AI companies report material P&L impact today. Healthcare AI companies are fighting regulatory battles that will take years. That divergence determines which stocks work in 2025-2026.

But splits don't last forever. Lagging industries eventually adopt or disappear. Healthcare will modernise regulation under pressure from AI-enabled competitors. Manufacturing will solve integration challenges as productivity gaps widen.

The stall-out requires believing competition doesn't matter but it does. Rapid acceleration assumes all sectors move in unison however they won't. The long grind fits the data, the history, and the organisational physics of change.

For investors, this means abandoning index-level thinking. The AI trade isn't monolithic. Vertical specialists serving fast-adopting sectors will compound at 60% while horizontal platforms grind at 25%. Stock picking matters more than sector allocation.

For boards and executives, your industry's adoption curve determines your timeline, not the headlines. Retailers and software companies must embed AI within 18 months or risk losing position. Healthcare and financial services have longer runways but higher hurdles.

The deployment gap is real. The variance is bigger than the average. Valuations may exhale before adoption inhales but the oxygen of real deployment will return.

"After looking at these four potential futures, I'm convinced: this isn't a bubble bursting, it's a system learning how to drive." I think I will "hold" these stocks and wait for the segments and markets to "catchup" even if there are corrections.