In every technological revolution, there comes a moment when the narrative shifts. The early phase is about ambition, capital expenditure, and scale. The later phase is about efficiency, returns, and competitive advantage. Kai Wu, Founder and CIO of Sparkline Capital, argues that artificial intelligence has reached precisely this inflection point.
In Sparkline’s 2025 ETF Investor Letter1, Wu makes a clear and provocative case: the AI boom is transitioning from infrastructure buildout to enterprise adoption — and investors who fail to adjust risk being positioned for the wrong phase of the cycle .
The implications are profound.
I. The Inflection Point: From Buildout to Adoption
Wu frames the transition in historical context. Technology revolutions follow a diffusion curve. Early capital floods into foundational infrastructure. Only later does widespread economic value creation emerge.
He anchors the moment with a quote from Microsoft CEO Satya Nadella:
“For this not to be a bubble by definition, it requires that the benefits of [AI] are much more evenly spread … not just economic growth driven by capital expenses.”
Wu argues:
“We believe the AI boom has reached a pivotal moment, with the focus now shifting from buildout to adoption.”
The distinction matters.
Hundreds of billions of dollars have gone into AI data centers and chips. Infrastructure stocks — especially the Magnificent 7 — dominated performance. But Wu notes early signs of fatigue:
- Heavily indebted infrastructure players fell sharply from highs.
- Even the Magnificent 7 have lagged in recent months.
- Adoption, not capex, will determine the sustainability of the boom.
Enterprise AI adoption remains early:
“According to the U.S. Census Business Trends and Outlook Survey, only around 10% of businesses currently use AI in production.”
Despite three years since ChatGPT’s launch, we are, in Wu’s words, “still in the early innings” .
The boom now demands proof.
II. ROI: The Proof That Changes Everything
The letter’s most original contribution is Sparkline’s systematic study of AI-driven return on investment (ROI) in earnings calls.
Wu writes:
“Ultimately, the long-term success of the AI boom hinges on whether AI tools are able to generate a meaningful return on investment (ROI) for their users.”
Sparkline trained an LLM to classify corporate mentions of AI into three increasingly strict tiers:
- AI usage
- AI-driven economic gains
- AI-driven ROI
The progression is crucial. Vague experimentation does not equal value creation. But quantifiable savings and returns do.
Since 2017:
- Firms reporting AI-driven ROI rose from near zero to 155.
- Firms reporting AI-driven economic gains rose to 675.
- 7% of AI mentions now include quantified ROI.
- 32% reference economic gains.
The market impact is measurable:
“Since 2017, firms merely discussing AI usage on earnings calls have outperformed the market by 3.2% per year. More importantly, companies able to point to specific AI-driven economic gains or ROI have done even better, earning excess returns of 4.8% and 5.2%, respectively.”
This is not hype. It is earnings leverage.
AI gains cluster into three categories:
- Revenue growth
- Productivity gains
- Cost savings
The message: tangible value creation is spreading beyond chipmakers.
III. Measuring Adoption Before It Shows Up in Earnings
Earnings disclosures are backward-looking. Sparkline’s next step was to measure AI adoption directly.
Wu explains:
“For the past several years, Sparkline has been continually tracking corporate investment in AI innovation, products, and talent across thousands of firms — parsing millions of corporate communications, patents, trademarks, employee LinkedIn profiles, and other documents using LLMs and other tools.”
From this they derive a composite AI adoption score.
Two findings stand out:
- Adoption varies widely across industries.
- It varies even more within industries.
Wu notes:
“Most industries feature a handful of early adopters investing aggressively in AI, while the vast majority of firms lag far behind.”
Sparkline categorizes firms into:
- AI Infrastructure
- AI Early Adopters
- AI Laggards
The breakdown is striking:
“Of the roughly 4,500 companies in our global stock universe, 80% are considered AI laggards… The remaining 20% is split roughly evenly between AI infrastructure firms… and AI early adopters.”
Even more striking:
“AI early adopters report even more gains than the infrastructure providers themselves.”
In other words: the companies buying the tools may ultimately benefit more than those selling them.
IV. The Market Is Not Pricing This
Despite superior economics, early adopters remain underappreciated.
Wu writes:
“Interestingly, the market does not appear to be giving AI early adopters credit for their AI investments in the form of higher valuations. As the next exhibit shows, there is virtually no correlation between AI adoption and valuation.”
Meanwhile:
“AI infrastructure firms have been richly rewarded… they now trade at nearly twice the valuations of the broader market.”
Here lies the core divergence.
Infrastructure valuations embed aggressive assumptions about near-term demand. Wu warns:
“We believe risk skews to the downside, as these stocks could see significant losses from multiple compression if demand does not materialize as quickly as investors expect.”
Conversely, early adopters may represent a classic mispricing — productivity improvement not yet capitalized into multiples.
V. The Infrastructure Subsidy Effect
Wu introduces a powerful historical analogy.
“A key lesson of past infrastructure booms is that the builders of the underlying infrastructure often fail to capture much of the value they create. Rather, this value accrues to their customers and the rest of society.”
Examples:
- Railroads
- Fiber-optic telecom
He writes:
“Past booms have invariably resulted in overbuilding, with excess capacity effectively a subsidy from builders to their customers.”
Today:
“AI infrastructure firms are footing an expected $5 trillion bill for the AI buildout, while their customers enjoy declining capital intensity.”
And:
“Early adopters get to reap the fruits of AI progress without the risk of large speculative capital outlays.”
Competition is rising at the model layer. Wu observes:
“Once far ahead of its rivals, OpenAI now shares the frontier with Anthropic and Google – with even open-source DeepSeek models not far behind.”
If AI models commoditize, value migrates to applications and users.
History suggests this is plausible.
VI. Index Risk: The S&P 500 Is a Buildout Bet
One of Wu’s most practical warnings concerns portfolio construction.
“A whopping 46% of the S&P 500 is currently in AI infrastructure stocks, leaving investors highly exposed to the rising risks of the AI buildout.”
Alternative indexes overshoot in the other direction, concentrating in laggards.
Wu proposes a “third path”:
“Exposure to AI early adopters, which we believe combine the best features of both.”
Sparkline positions its ETFs to maintain AI intensity while avoiding infrastructure concentration risk.
VII. Value Investing in Hype Cycles
The letter returns to discipline.
Wu reminds investors:
“Valuation discipline is essential when navigating the hype around technological revolutions.”
And warns:
“While the Internet succeeded as a technology, investors who bought highly valued Internet stocks in 2000 spent the next two decades underwater.”
Traditional valuation frameworks penalize intangible investment. Sparkline incorporates intangibles into valuation models, allowing:
“A way to rotate out of stocks as they become overvalued while maintaining long-term innovation exposure.”
This dynamic rotation led ITAN to reduce infrastructure exposure as valuations expanded and reallocate toward early adopters.
VIII. Performance and Positioning
2025 results were strong:
- ITAN: +20.3%
- DTAN: +29.6%
Notably, ITAN outperformed the S&P 500 despite an underweight to infrastructure — largely through security selection .
DTAN underperformed EAFE headline numbers due to underweight financials but outperformed on an ex-financial basis .
Wu concludes that both funds are positioned for the adoption phase.
Key Insights & Points of Debate
1. Is Adoption Truly Accelerating?
The evidence of rising ROI mentions is persuasive. However, 7% ROI references remain small. Skeptics may argue adoption could stall.
2. Will Infrastructure Truly Commoditize?
If AI compute remains scarce and pricing power holds, infrastructure profits may endure longer than expected.
3. Are Early Adopters Really Underpriced?
The “no correlation” finding between adoption and valuation is striking. Markets often move ahead of fundamentals — but perhaps not here.
4. Can LLM-Based Adoption Scoring Remain Predictive?
Sparkline’s methodology is innovative. Its durability will be tested as AI disclosure becomes more common.
Key Takeaways for Advisors and Investors
1. Recognize the Phase Shift
The AI story is no longer just about chips and data centers. The marginal opportunity may lie in enterprise productivity gains.
2. Demand Measurable ROI
Companies reporting quantifiable AI ROI have historically delivered superior returns.
3. Reassess Index Concentration
Market-cap weighting embeds a heavy infrastructure bet. Diversification may require intentional rebalancing.
4. Look for Adoption Dispersion
Within-sector differences are large. Early adopters may outcompete peers.
5. Maintain Valuation Discipline
Technological revolutions reward users long-term — but punish investors who overpay.
6. Expect Volatility in the Transition
As leadership rotates, performance dispersion may increase.
Conclusion: The Real AI Trade
Wu closes with clarity:
“As the AI boom shifts from buildout to adoption, investors need a new approach.”
And:
“AI early adopters offer a path forward. Currently overlooked, they are starting to generate real AI-driven ROI while offloading capital spending risk.”
History rarely rewards those who chase the most visible layer of a technological revolution. It rewards those who identify where value ultimately accrues.
If Wu is correct, the next decade’s winners may not be those who built the AI rails — but those who quietly learned how to ride them.
Footnotes:
1 Wu, Kai. AI Adopters: Beneficiaries of the Boom. Sparkline Capital, 30 Jan. 2026.
