The AI investment narrative is undergoing a pivotal transformation. As recent quarterly results from tech titans like Microsoft and Amazon have eased concerns about the return on capital expenditure, top-tier global asset managers are pivoting their focus from "how much is being spent" to "who is actually positioned to profit."

Institutions such as Wellington Management, which oversees roughly $1.3 trillion in assets, and Capital Group, managing approximately $3.6 trillion, are actively increasing their stakes in hyperscale cloud providers while maintaining their semiconductor positions. These investors believe that accelerating cloud growth and persistently tight compute supply will disproportionately benefit companies like Amazon, Microsoft, and Google, which control vast infrastructure and deep customer relationships.

Conversely, "neocloud" service providers that depend on expensive leased compute are being flagged as potentially vulnerable, with some institutions recommending short positions. This sentiment shift is clearly reflected in asset prices. The Philadelphia Semiconductor Index has climbed roughly 75% year-to-date, while the four major AI capital expenditure giants have underperformed that rally. Meanwhile, compute leasing firms CoreWeave and Nebius have surged approximately 50% and over 200% respectively, driven by a scarcity premium on compute. However, institutional investors caution that once new compute capacity comes online en masse and leasing prices normalize, the high-debt, high-pricing models of these latter companies will face severe tests.

Hyperscalers Regain Favor with Institutions

The robust rebound in cloud demand forms the core logic behind this institutional accumulation. Brian Barbetta, co-head of the technology platform at Wellington Management, noted that hyperscalers are being recognized by the market as major beneficiaries of the AI paradigm shift, stating, "They remain a core holding in our portfolio, and in fact we have recently increased our allocations across several of these companies."

Richard Clode, fund manager at Janus Henderson's Bankers Investment Trust, lists Amazon as one of his fund's largest overweight positions, summarizing his investment thesis as "today's capital expenditure is tomorrow's sales revenue." He anticipates that by late next year through 2028, profit and cash flow growth at these companies will begin to outpace the incremental growth in capital spending.

According to Reuters estimates, the annual operating cash flow of hyperscalers in 2027 is projected to increase by approximately $340 billion compared to 2025, while capital expenditure during the same period is expected to rise by about $534 billion. The sustained release of cash flow returns is a key basis for institutions maintaining their overweight positions.

Chips and Cloud: Not an Either/Or Proposition

Some investors emphasize that AI investment should not be simplified into a binary choice between semiconductors and cloud computing. John Lamb, equity investment director at Capital Group, points out that investors should view AI as an ever-expanding ecosystem, adding, "The question isn't whether chips are a better investment than hyperscalers, but rather that both should be part of your portfolio."

Lamb also reminds that data centers typically require 12 to 18 months from breaking ground to generating revenue, noting, "We are just beginning to see this inflection point in the latest earnings reports." This suggests that the conversion of current capital expenditure into revenue is still in its early stages, and investors need to exercise patience.

Valuations for hyperscalers have already compressed somewhat from their highs but remain below post-pandemic peak levels. Among them, Microsoft trades at the highest forward P/E ratio of roughly 24.6 times, while Meta sits at the lowest with approximately 17.6 times, providing institutions with reasonable valuation room to continue building positions.

Compute Leasing Firms: The Dividend May Be Fading

The high prosperity of neocloud service providers is built on the premium of compute scarcity, a foundation that is now being shaken. Noah Weisenberger, chief US equity strategist at BCA Research, points out that once new compute capacity comes online in large scale and market pricing normalizes, compute leasing companies that rely heavily on debt financing and premium pricing will face significant risks.

Weisenberger consequently recommends a paired trading strategy of "going long hyperscalers and shorting neocloud providers." He also notes that the transition of hyperscalers to more capital-intensive business models, even with strong profitability, could place certain constraints on their valuations.

Clode further elaborates on the moat of hyperscalers from a competitive landscape perspective: companies that simultaneously control compute resources and possess the platform capabilities to help customers efficiently deploy AI across models, optimizing both cost and performance, will develop more durable competitive advantages. Amazon, Microsoft, and Google possess precisely these traits through their scale and customer relationships, while neocloud providers fall short in comparison.

Winners Are Destined to Be Scarce

Even with a bullish long-term outlook on AI, institutions remain clear-eyed about the high barriers to entry in current investing. Alberto Conca, Chief Investment Officer at Swiss wealth manager LGF+ZEST, estimates that AI monetization levels would need to increase by 5 to 13 times to justify current spending plans.

Barbetta predicts that as the market matures, competition will further weed out players, with the ultimate winners being those with the broadest technology portfolios, the deepest customer relationships, and the strongest control over their own infrastructure. More specialized competitors may fall behind in this race.

"The future winners will be far fewer in number than today's market participants," he said.