Jefferies strategist Steven DeSanctis highlighted a growing divergence in artificial intelligence investment outcomes in a July 10 note. He identified a critical stress test for AI return on investment looming within the next 12-18 months. The analysis favors suppliers of AI infrastructure, known as pick-and-shovel stocks, over the hyperscale cloud companies making the massive upfront investments. This bifurcation stems from the immediate revenue recognition for suppliers versus the multi-year uncertainty for ultimate payoffs on AI spend.
Context — why this matters now
Massive capital expenditure cycles often precede a reassessment of value creation. The dot-com bubble of the late 1990s serves as a historical comparable, where infrastructure providers like Cisco Systems saw revenue surge before a dramatic collapse in telecom carrier spending triggered a broad market correction. The current AI investment wave is unprecedented in scale, with the Magnificent Seven tech companies alone guiding for combined capex exceeding $200 billion in 2024.
The macro backdrop features elevated interest rates, with the Fed funds rate at 5.25-5.50%. Higher financing costs increase the scrutiny on any long-duration project with an uncertain payoff. The catalyst for this concern is a wave of earnings calls where management teams from Microsoft, Amazon, and Google have faced intense questioning from analysts demanding clearer timelines for monetizing generative AI features. This quarter marks a shift from celebrating spending to demanding accountability.
Data — what the numbers show
Hyperscaler capital expenditure reached an aggregate $178 billion in 2025, a 42% year-over-year increase. Nvidia, the quintessential pick-and-shovel play, reported data center revenue of $47.5 billion in its last fiscal year, a 279% increase. Broadcom's AI-related revenue is projected to surpass $20 billion in fiscal 2026, representing over 40% of its semiconductor business.
A comparison reveals the valuation gap driving the thesis. The iShares Semiconductor ETF (SOXX) trades at a forward P/E of 25.5x. The Cloud Computing ETF (WCLD) trades at a forward P/E of 38.7x, indicating higher growth expectations for software and services. Key AI infrastructure stocks show significant outperformance versus the broader market. The S&P 500 is up 8.2% year-to-date, while the Philadelphia Semiconductor Index (SOX) has gained 24%.
| Metric | Hyperscaler Capex (2025) | Nvidia Data Center Revenue (FY2025) | SOX YTD Performance |
| | | | |
| Value | $178B | $47.5B | +24% |
Analysis — what it means for markets / sectors / tickers
The immediate beneficiaries are semiconductor capital equipment makers, chip designers, and specialized data center real estate investment trusts. ASML Holdings, Synopsys, and Digital Realty Trust capture revenue early in the investment cycle. Their earnings are largely insulated from whether an AI model ultimately achieves commercial success. The hyperscalers—Microsoft, Amazon, Google-parent Alphabet, and Meta Platforms—must wait years to see returns through cloud service adoption and new software product sales.
A key risk to this thesis is a sudden contraction in hyperscaler spending, which would immediately hurt supplier revenues despite their favored positioning. A slowdown in cloud revenue growth could trigger capex cuts. Flow data indicates institutional investors are already positioning for this bifurcation. Recent ETF flow analysis shows net inflows into semiconductor funds while dedicated cloud computing ETFs have seen modest outflows over the past month.
Outlook — what to watch next
The Q2 2026 earnings season, commencing in mid-July with major bank reports, will provide the next crucial data point. Microsoft, Alphabet, and Meta Platforms report during the week of July 24th. Analysts will scrutinize any changes to full-year capex guidance and question management on AI monetization metrics. Any guidance reduction would immediately pressure hyperscaler valuations.
Key technical levels to monitor include the 50-day moving average for the VanEck Semiconductor ETF (SMH) at approximately $265. A sustained break below could signal a momentum shift for the entire theme. For the Nasdaq-100 index, the 19,500 level represents critical support; a breach would indicate broad tech weakness outweighing AI optimism. The next Federal Open Market Committee meeting on September 20th will dictate the cost of capital for continued high investment.
Frequently Asked Questions
What are pick-and-shovel stocks in AI?
Pick-and-shovel stocks are companies that provide the essential tools and infrastructure needed for AI development rather than developing the end-use AI applications themselves. This category includes semiconductor manufacturers like Nvidia and AMD, chip equipment makers like Applied Materials, and data center operators like Equinix. They generate revenue upfront from the hyperscalers' capital expenditure, making their financials less dependent on the ultimate success of specific AI models.
How does current AI spending compare to past tech bubbles?
Current AI spending as a percentage of market capitalization for major tech firms remains below peak dot-com levels but is growing at a faster absolute rate. During the 1999-2000 bubble, telecom capex reached nearly 15% of S&P 500 market cap. Today, tech capex is approximately 6% of the index but involves sums exceeding $200 billion annually due to the market's larger size. The key difference is that current spending is backed by strong corporate balance sheets rather than debt-fueled speculation.
What is the biggest risk to the pick-and-shovel investment thesis?
The largest risk is a sudden and sharp pullback in capital expenditure by the major cloud providers. If hyperscalers like Amazon Web Services or Microsoft Azure see a slowdown in cloud revenue growth or encounter economic headwinds, they could delay or cancel expansion plans. This would immediately impact orders for semiconductors, servers, and data center capacity, causing a rapid de-rating of supplier stocks despite their structural positioning in the value chain.
Bottom Line
Suppliers monetizing the AI infrastructure build-out face less execution risk than the hyperscalers funding it.
Disclaimer: This article is for informational purposes only and does not constitute investment advice. CFD trading carries high risk of capital loss.
