With five major cloud operators expected to spend $1.2 trillion on artificial intelligence infrastructure in 2027, New Constructs says investors should separate purchased compute capacity from profit. Compute capacity can create an advantage, but only when it produces measurable customer value, sustainable cash flow, and returns that justify market expectations.

NASHVILLE, Tenn., Sept. 1, 2026 /PRNewswire/ -- Morgan Stanley estimated that a small group of the world's largest cloud operators could spend $1.2 trillion on AI infrastructure next year. But David Trainer, CEO of New Constructs, a financial technology firm specializing in fundamental investment research and valuation analysis, argues that the market is still asking the wrong question: not who is building the most AI capacity, but who owns the proprietary data and workflows capable of turning that investment into measurable economic profit.

"AI infrastructure is an input, not a moat," Trainer said. "The investor's job is to determine whether a company has an advantage competitors cannot replicate and whether that advantage is producing returns above the cost of capital."

AI Capacity Buildout Is Not a Moat

Chips, servers, data centers, power, and access to general-purpose models are necessary for AI adoption. They do not automatically create pricing power, customer value, or economic profitability.

Demand for AI is not in question. A 2026 survey of 178 global asset-management executives found that 95% expect AI to be important to meeting their investment management goals. Trainer said widespread adoption does not establish which providers will retain pricing power as models converge, compute becomes easier to rent, and public information remains available to every competitor.

The investment test is no longer capacity; it is economic return. An AI deployment should show that it changes the business by producing differentiated research outcomes, lowering costs, improving margins, or increasing revenue. Without those results, the market may be overpaying for AI capacity before knowing whether adequate future cash flows will emerge.

The Scarce Asset Is Not the Model

Recent comments from Hims & Hers CEO Andrew Dudum and investor Chamath Palihapitiya support that distinction. Dudum has argued that foundational models have limited value without a closed-loop dataset, meaning proprietary information generated inside a business and unavailable to competitors. Palihapitiya has similarly argued that value will accrue in the "harness" and application layers that organize proprietary data, workflows, evaluations, and business rules around models.

That shifts the competitive question from access to compute to access to information others cannot easily obtain or reproduce. As models become more widely available, the ability to apply proprietary data to specialized problems may become a more durable source of differentiation.

"Public models learn from information everyone can access," Trainer said. "The data nobody else has is where a company can create an edge. Giving it away can mean giving away the business model."

Data Moats Still Have to Earn Their Valuation

Palantir, a software and data analytics company serving government and commercial customers, offers one example of a data-first approach. Its model organizes customer data into operational workflows designed to facilitate specific operations and address specific problems. Its $10 billion Army software and data contract illustrates the scale of demand for systems that can structure and apply AI tool to create value with proprietary information.

Demand, however, does not settle the valuation question. According to New Constructs' analysis as of August 29, 2022, at $187/share Palantir's stock price implies profit growth of approximately 20% compounded annually for more than 30 years, or 35% compounded annually for more than 15 years.

"The right question is not whether a company is good," Trainer said. "The question is what the market is already pricing in. A strong business can still be a bad stock when the expectations embedded in its price are too difficult to meet."

The distinction between AI infrastructure and AI value also shapes New Constructs' approach to AI. FinSights pairs Google Cloud technology with New Constructs' proprietary investment research data, using the infrastructure as a delivery layer rather than the source of the investment insight.

Trainer sees AI following the path of electricity: essential to society, but not necessarily a profit pool captured by the companies that build the underlying infrastructure. For investors, the question is not whether AI will be widely used. It is whether a company can turn AI into services, workflows, or insights customers will pay for at margins that justify the investment. "Trillions of dollars in AI spending are not profits," Trainer said. "They are costs that still must be earned back. The winner is not whoever spends the most. It is whoever can turn proprietary data into services competitors cannot easily replicate."

About New Constructs

New Constructs is an independent financial technology firm bringing transparency and accountability to investment research. Combining forensic accounting expertise with patented AI technology, the company analyzes SEC filings and financial disclosures with unprecedented speed and accuracy to uncover the true profitability and valuation of public companies. The firm's research drives live-traded indices that dramatically out-perform the S&P 500. Its Robo-Analyst platform automates financial modeling and investment ratings across more than 10,000 securities, while FinSights, built with Google Cloud technology, demonstrates how AI can support investment analysis when grounded in trusted, auditable data. For more information, visit www.newconstructs.com.

References

• Bharade, A. (2026, August 19). Companies with huge datasets can save up to 80% in AI costs by using open-weight models: Hims CEO. Business Insider. businessinsider.com/hims-hers-ceo-andrew-dudum-datasets-ai-costs-open-weight-2026-8
• CNBC. (2026, August 18). Hims & Hers CEO Andrew Dudum on the company's approach to AI agents [Video]. cnbc.com/video/2026/08/18/him-hers-ceo-andrew-dudum-on-the-companys-approach-to-ai-agents.html
• Comtois, J. (2026, July 22). Asset managers are doubling down on AI, but data quality is emerging as the real differentiator. Institutional Investor. institutionalinvestor.com/article/asset-managers-are-doubling-down-ai-data-quality-emerging-real-differentiator
• MacFadden, D. (2026, July 17). Who pays for AI? Financial Times. ft.com/content/05976c31-3a30-4d25-b1cb-6a2559014c1f
• Morgan Stanley. (2026, June 3). The new AI credit playbook [Audio podcast episode]. In Thoughts on the Market. morganstanley.com/insights/podcasts/thoughts-on-the-market/credit-markets-ai-capex-boom-vishy-tirupattur
• Palihapitiya, C. [@chamath]. (2026, August 22). Here is my AI investing guide [Substack note]. Substack. substack.com/@chamath/note/c-319432183
• Subin, S. (2025, August 1). Palantir lands $10 billion Army software and data contract. CNBC. cnbc.com/2025/08/01/palantir-lands-10-billion-army-software-and-data-contract.html
   

Media Inquiries:

Karla Jo Helms
JOTO PR
727-777-4629
Jotopr.com

View original content to download multimedia:https://www.prnewswire.com/news-releases/ai-markets-are-mis-pricing-capacity-as-profit-new-constructs-warns-302865732.html

SOURCE New Constructs