NextFin News - Sebastien Page, chief investment officer and co-head of global investments at T. Rowe Price, laid out the three forces now setting the terms for portfolios in an August 31, 2026 television interview: the AI investment boom, sticky inflation risks, and a bond market that has pushed the 30-year Treasury yield above 5% for its longest stretch since 2007. The combination is the story. The same capital spending lifting AI earnings is also competing with government debt for bond buyers - and keeping the discount rate that values those earnings uncomfortably high.
The interview landed on a day of cross-currents. The S&P 500 fell to 7,683.52 on August 31, down 0.25%, still within reach of the record set earlier in the month, when the index touched an intraday high of 7,816.70 and a closing peak of 7,798.99 on August 13. Nvidia, the bellwether for the AI trade, closed at $217.55 on August 28 with a market capitalization of roughly $5.25 trillion, up about 16.79% year-to-date against 12.65% for the S&P 500. Page took over as co-head of global investments and CIO on June 1, 2026, and oversees the allocation of approximately $1.80 trillion in client assets.
The question the interview framed - and that his firm's published positioning answers - is whether the AI trade can survive the very yields its own financing needs are helping to push up.
The AI Trade Has Become a Bond-Market Story
The transmission channel is no longer subtle. Artificial-intelligence capital spending has moved from a digital earnings story to a fixed-income supply story. When the builders of AI tap debt markets to fund data centers, power connections, and the physical infrastructure of the buildout, they add to a Treasury market already absorbing record government issuance. More supply chasing the same pool of bond capital means a higher term premium; a higher term premium means a higher discount rate; and a higher discount rate feeds straight back into the valuations of the long-duration growth stocks that led this cycle.
The scale of the spending is what makes the channel matter. Hyperscalers spent about $260 billion on capital expenditure in 2024. Analysts at Goldman Sachs and Morgan Stanley estimate AI-related capex will reach roughly $800 billion in 2026, and Morgan Stanley has raised its 2027 outlook by 17% to $1.12 trillion. JPMorgan's strategists put sell-side 2026 capex estimates at $697 billion for just five US companies, up $173 billion since the start of the year. Industry analysis suggests that capex will absorb around 94% of hyperscaler operating cash flow across 2026 and 2027.
Put those figures against the fixed-income market they are entering. The US 30-year Treasury yield held above 5% for its longest stretch since 2007, a level last seen near 5.32% in mid-August, while the benchmark 10-year yield traded in the 4.68% to 4.74% range. The 30-year yield briefly touched 5.197% in late July, its highest level since 2007. A survey of global fund managers conducted by Bank of America found 62% of respondents expect 30-year yields to reach 6% - a level last seen in late 1999 - with only 20% targeting 4%.
T. Rowe Price has been explicit about the shift. In its 2026 Global Market Outlook, the firm wrote:
AI is evolving from a purely digital story into a real-world infrastructure boom.
The firm added that investment is spreading beyond large technology platforms into areas such as power. That is the point at which an equity narrative acquires a bond-market footprint.
Page's own published view on the funding competition is direct:
Value will be harder to find in high-quality government bonds, where significant fiscal expansion and economic growth are pushing supply and yields higher in a competition to attract buyers.
Note what is being described: not a cyclical wobble in rates, but a structural change in who is issuing, and why.
Cyclical or Structural: The Sell-Off Is Temporary, the Capex Cycle Is Not
Getting this call right decides the conclusion. The AI-driven pullback in August is cyclical; the capex cycle behind it is structural. They must be argued separately, because blending them produces the wrong portfolio answer.
The cyclical case rests on sentiment and rotation, and T. Rowe Price's August 2026 asset-allocation update states it plainly: "We believe the AI-driven sell-off is temporary." The firm remains overweight equities and recently increased that position, even as it trimmed its overweight to credit because spreads have become "increasingly compressed." Page himself framed the rotation in February 2026, describing a market moving "from the builders of AI to the beneficiaries of AI, to the victims of AI" - a pattern of leadership rotation within a theme, not the end of the theme.
The structural case rests on regime change. Three years after the launch of ChatGPT, T. Rowe Price's published research argues that AI remains on track to become "the biggest productivity driver for the global economy since electricity," with the potential to transform nearly every sector rather than just technology. The capex is shifting from software and chips into power generation, transmission, and construction - a multi-year physical buildout, not a one-quarter spending spike. And critically, the spending is funded from the operating cash flow of profitable hyperscalers, not from speculative equity issuance by loss-making startups, as it was in the dot-com era.
The evidence floor for the structural call is met: the driver is a permanent change in technology and industry structure, the funding base is durable cash flow, and the buildout has multi-year visibility. T. Rowe Price's January 2026 analysis of the AI capex cycle reached the same conclusion from a different angle, arguing the cycle can persist for another two to three years before facing its first true test, sustained by competitive incentives, supply bottlenecks, and measurable progress in monetization. The evidence floor for the cyclical call is also met: the firm's own positioning language, the historical pattern of sector rotations within secular themes, and the concentration-and-valuation anxiety that typically mean-reverts.
History offers three comparable cycles, and the differences matter. The dot-com bust of 2000-2002 broke when the Federal Reserve tightened and capital became expensive for loss-making issuers; today's AI builders are cash-generative with net cash balance sheets. The 2013 taper tantrum pushed the 10-year yield from 1.6% to 3% in five months and compressed duration-rich valuations without ending the technology cycle; today's yield move is slower but starting from a higher base. The 2018 volatility episode saw the S&P 500 fall nearly 20% into Christmas Eve on growth-and-rates fears before a sharp reversal; today's earnings backdrop is far stronger. The common thread across all three is that rate shocks test valuation, not necessarily the underlying investment thesis.
Inflation Is the Second-Order Risk the Market Is Still Underpricing
The first-order effect of AI capex is higher earnings for the builders. The second-order effect - the one the market has not fully priced - is that the capex itself is inflationary before it is productive. A demand shock to power capacity, construction, and industrial materials arrives now; the supply-side productivity gain arrives later. In between, the Federal Reserve sits on its hands.
Page's firm has made its inflation stance unmistakable. Its August 2026 update describes "inflation remaining the more important risk than slowing growth." In a late-August television appearance, Page said, "We continue to see the risks in inflation." And in the firm's 2026 outlook, he wrote:
Given the potential for persistent inflation against a backdrop of improving growth, tight labor markets, and supply chain friction from tariffs, we believe it is reasonable to hedge the inflation risk through tilts to inflation-sensitive parts of the market, such as real assets equities and short-term Treasury inflation protected securities within fixed income.
The 2026 Global Market Outlook goes further: "The fight against inflation looks increasingly difficult as a manufacturing revival, fiscal expansion, and policy shifts introduce new risks." That is not a passing concern. It is the reason the firm's highest-conviction fixed-income ideas are to keep duration low, overweight public credit versus government bonds, underweight the US versus the rest of the world, seek exposure to inflation-linked bonds, and overweight emerging markets with a quality bias.
Run the propagation chain to its end. AI capex raises demand for power and construction. Power and construction run into capacity constraints and tight labor - power infrastructure lead times currently stretch three to five years, according to T. Rowe Price's capex-cycle analysis. Tariffs add supply-chain friction. Inflation stays above the Federal Reserve's 2% target; the combined average of CPI and PCE core inflation was running near 3% through mid-2026, roughly Transamerica's forecast for year-end. The Fed holds rates where they are, or signals that cuts are further away than the market wants. Yields stay high. Equity duration stays under pressure. The market has priced a Fed on hold; it has not fully priced that the AI boom itself is part of what keeps the Fed there.
The Adversarial Case: The Bulls Are Right About Earnings
The strongest argument against the caution above is simple and data-backed: the earnings are real. In the second quarter of 2026, 86% of S&P 500 companies that reported posted actual earnings per share above estimates - the highest positive-surprise rate since the second quarter of 2021, when it was 87%. Blended year-over-year earnings growth for the index reached 50.4%, the highest since that same quarter of 2021, when growth was 91.6%. Revenue grew 13.8% on $5.1 trillion of blended quarterly sales, and companies reported aggregate earnings 29.2% above estimates. This is not a multiple-driven rally pretending to be fundamental.
Nor is the AI capex a dot-com replay. The dot-com cycle broke when capital became expensive and investors stopped rewarding growth at any price; today's builders are funding expansion from cash flow generated by profitable businesses. T. Rowe Price itself is effectively betting on this: it calls the AI sell-off temporary, keeps an equity overweight, and has recently increased it. Strategists at Transamerica and Raymond James hold year-end 2026 S&P 500 targets of 7,400 and 7,650 respectively - and the index closed August at 7,683.52, implying the bull case sees room to run from here.
The answer to the bull case is not that AI fails. It is that earnings and valuation are two different clocks. A 20% required return on the AI capital stock - a hurdle rate used in industry return-on-capital analysis - implies roughly $400 billion a year of incremental revenue just to replace depreciated assets, plus an additional $800 billion to deliver the return itself, on a $4 trillion stock of capital. That revenue must arrive before the debt used to build the capacity refinances at higher rates. The risk is a timing mismatch: capex outlays land today, monetization lands later, and the cost of the capital in between is set by a bond market that is being asked to absorb both AI paper and record Treasury supply at once.
There is also a concentration problem the bull case must answer. Nvidia alone accounts for roughly $5.25 trillion of market capitalization, and the AI leaders carry a disproportionate share of index gains; when a handful of names fund the bulk of the capex and capture the bulk of the multiple, the margin for error narrows. A single disappointment in the monetization timeline does not need to be large to move the index, because the weighting concentrates the risk.
Who Benefits, Who Is Exposed, and What Would Prove This Wrong
Translating the mechanism into impact: the beneficiaries are inflation-sensitive assets - real-assets equities and short-term Treasury inflation-protected securities, the exact hedges Page recommends - and public credit over government bonds, though that overweight is being trimmed as spreads compress. Beyond the builders, the next winners are the companies that monetize AI productivity rather than those that merely finance its construction.
The exposed are long-duration US government bonds - the firm remains underweight duration, primarily in the US - and the AI builders whose capex absorbs around 94% of operating cash flow. Any company priced for flawless execution of a multiyear monetization timeline is exposed to the same repricing if the funding cost of that timeline rises.
Split by time horizon, the picture is not uniform. In the short term, sentiment and liquidity dominate: choppy, yield-driven trading, with the August AI sell-off judged temporary. Over the medium term, fundamentals decide leadership: earnings momentum is strong, but the funding mix and the level of the 10-year Treasury yield determine which parts of the market keep winning. Over the long term, the structural view prevails: AI as an electricity-scale productivity shift, with a capex cycle built to persist.
Three scenarios frame the path. In the base case, the Federal Reserve holds, the 10-year Treasury yield ranges between roughly 4.25% and 4.75%, and AI leadership broadens beyond the hardware builders. In the upside case, monetization accelerates, capex converts to cash flow faster than expected, yields stabilize, and multiples re-rate higher. In the downside case, core inflation prints at 0.3% or more month-over-month for two consecutive months, the Fed is forced to hold longer or signal hikes, the 30-year yield breaks decisively above 5.5%, and capex-heavy names de-rate on funding-cost fears.
That inflation threshold is the falsifying signal. If core PCE prints at or above 0.3% month-over-month for two consecutive months, the view that inflation is a manageable cross-current rather than the dominant regime driver is wrong - and the bond market, not the AI earnings cycle, will set the terms for equities.
The AI boom is not being killed by a lack of demand. It is being repriced by the cost of the capital it consumes, and until yields settle, the trade belongs to the beneficiaries, not just the builders.
Explore more exclusive insights at nextfin.ai.
