NextFin News - Greg Fleming, chief executive of Rockefeller Capital Management, is making a deliberately two-sided bet on the American economy: artificial intelligence is the strongest reason for optimism in a generation, and the federal deficit is the single biggest threat hanging over it. Speaking in a television interview on Sept. 2, 2026, Fleming argued that AI is already lifting growth and productivity across the economy, even as he warned that Washington's fiscal path - deficits he puts at 5% to 7% of gross domestic product with no political impetus to change course - risks crowding out the very expansion he expects AI to deliver. The tension between those two convictions is the story investors need to price. It is a view delivered from a position of scale: Rockefeller Capital Management oversees $224 billion in client assets as of June 30, 2026, across a 140-year legacy that began with John D. Rockefeller's original 1882 family office.
The Setup: A Boom Built on Borrowed Money
Fleming's optimism is not abstract. It rests on the largest corporate capital-spending cycle in history. The five largest U.S. cloud and AI infrastructure providers - Microsoft, Alphabet, Amazon, Meta and Oracle - have collectively committed to between $660 billion and $690 billion of capital expenditure in 2026, nearly double the approximately $380 billion deployed in 2025. Amazon alone is guiding toward $200 billion; Alphabet toward $175 billion to $185 billion; Meta toward $115 billion to $135 billion; Microsoft toward $120 billion or more; Oracle toward $50 billion. That is not a forecast about what AI might do someday. It is money leaving corporate treasuries this year.
The equity market has priced that conviction aggressively. The S&P 500 closed at a record 7,798.99 on Aug. 13, 2026, and the index is up roughly 14% for the year while the Nasdaq has gained about 15%. On the day of the record close, seven of the 11 S&P 500 sector indexes rose, a sign the rally has broadened beyond the megacap technology names that led earlier in the year.
But the labor market is sending a softer signal, and it is the softening that makes Fleming's two-sided view necessary. The July employment report showed nonfarm payrolls falling by 23,000 against a median economist forecast of a gain of 80,000, with the unemployment rate at 4.1%. Average monthly payroll growth has slowed to 20,000 over the past three months from a year-to-date average of 60,900. And the fiscal backdrop keeps tightening: the federal debt crossed $39 trillion in mid-March, debt held by the public passed 100% of GDP in April for the first time since World War II, and net interest payments are projected to exceed $1 trillion in fiscal 2026 - nearly triple the $345 billion paid in 2020. The federal government now spends more on interest than on national defense.
So the setup is a three-way contest: an AI investment boom pulling growth up, a cooling labor market pulling it down, and a deficit large enough to push long-term interest rates higher regardless of what the Federal Reserve does. Fleming's view is that the first force wins - but only if the third does not break it.
The AI Capex Engine Is Real, and It Is Debt-Funded
The first question investors should ask about any capital-spending supercycle is: where does the money come from? In this one, a meaningful share comes from borrowed dollars. The five largest hyperscalers have issued more than $150 billion of U.S. investment-grade debt in 2026, with more than $60 billion issued in other currencies, according to Bank of America. Across the broader corporate sector, companies have borrowed roughly $600 billion to fund the AI build-out since last year.
That matters because it changes the risk profile of the boom. A capex cycle funded from free cash flow can be slowed quietly when returns disappoint. A capex cycle funded with debt carries fixed obligations that do not adjust when the revenue curve bends. If AI-driven revenue fails to materialize at the pace the infrastructure build assumes, the unwind is not just a reduction in spending - it is a credit event in the making. Fleming's optimism implicitly assumes the revenue follows the steel and silicon. The bond market will be the first to tell him if it does not.
The Productivity Transmission: Disinflation Is the Prize
The economic case for AI does not rest on chatbots. It rests on whether AI agents embedded inside companies raise output per hour - and, critically, whether they do it faster than they raise wages. Fleming said in an August interview with David Rubenstein that AI adoption is "moving faster than even I thought" and could have a disinflationary effect as companies embed AI agents to boost productivity.
"It's a fantastic amount of money to have borrowed, even for an economy this robust and this big."
The data so far are consistent with that view, though not yet conclusive. Nonfarm business productivity rose 2.9% year over year in the first quarter of 2026, continuing a recovery from the post-pandemic slump. In the third quarter of 2025, productivity surged 4.9% as real value-added output grew 5.4% while hours worked rose only 0.5% - companies produced materially more without hiring proportionally more. The St. Louis Fed, analyzing nearly 490,000 earnings-call transcripts, has found that AI references are proliferating across industries, though the measurable productivity payoff remains uneven.
This is the mechanism that makes Fleming's optimism coherent rather than hopeful. If AI raises productivity growth by even a percentage point or two on a sustained basis, the economy can grow faster without inflation accelerating - which gives the Federal Reserve room to hold policy steady or cut gradually rather than fight price pressure. That is the "grow our way out" scenario Fleming has floated: AI expands the numerator of GDP faster than the denominator of debt accumulates.
The Fiscal Counterweight: Deficits That Do Not Self-Correct
Here is the counterforce. Fleming's own words are the clearest statement of the risk. "It's a fantastic amount of money to have borrowed, even for an economy this robust and this big," he said of the national debt, which is approaching $40 trillion. "I'm most focused on the fiscal situation in this country," he added. "We're still at relatively full employment…We have been in a good time. We run these 5, 6, 7% of GDP deficits annually, and there seems to be no impetus to change that."
A deficit of 5% to 7% of GDP with the economy at or near full employment is historically unusual - it is stimulus applied when stimulus is not needed. The consequence shows up in the bond market as a term premium: extra yield investors demand for holding long-dated government debt when the supply of that debt is expanding without a credible plan to slow it. That pushes up the long end of the yield curve independently of the Fed's policy rate, which stands at 3.50% to 3.75% after the July FOMC meeting - where three of 12 committee members dissented in favor of a quarter-point hike.
The political economy makes this worse. Deficit reduction requires either higher taxes or lower spending, both of which subtract from GDP growth in the near term. No party has offered a credible path, which is why Fleming says there is "no impetus to change that." The result is a structural bid under long-term yields that competes directly with the structural disinflation AI might deliver.
Cyclical or Structural? Both Forces Are Present, and They Point in Opposite Directions
This is where the analysis has to separate the forces rather than blend them - because getting this wrong flips the conclusion.
The AI investment cycle has a cyclical leg and a structural leg. The cyclical leg is the capex surge itself - the rush to order graphics processors, build data centers, and sign power contracts. Capex cycles overshoot. They are financed by debt and sentiment, they cluster in time, and they frequently produce periods of overcapacity before demand catches up. The 1990s fiber-optic build is the textbook case: the infrastructure was real and transformative, but the investors who paid peak prices for it lost money for a decade. If 2026 is the peak of AI infrastructure spending, the cyclical leg will revert - orders will slow, equipment prices will fall, and the earnings growth that has carried the market will decelerate.
The structural leg is different. Once AI agents are embedded in business processes - in coding, customer service, accounting, logistics, design - they do not get un-embedded when the capex cycle turns. The productivity gains, if real, persist and compound. That is the regime-shift component: a permanent lift to the economy's potential growth rate, not a temporary sugar rush. Fleming's optimism is a bet that the structural leg is larger than the cyclical one.
The labor market and the deficit, by contrast, are more clearly cyclical in the near term but structural in the fiscal dimension. Payroll softening can reverse with a rate cut or a rebound in demand. The deficit, however, is structural: it will not self-correct without a political decision, and the interest burden grows automatically as debt rolls over at higher rates.
The judgment: the AI productivity shift is structural; the 2026 capex boom and the equity rally riding it are cyclical and already extended. That combination argues for optimism about the economy's medium-term trajectory and caution about the market's near-term risk premium. The market is pricing the structural payoff now, at record valuations, while the cyclical capex peak and the fiscal drag are still ahead.
The Second-Order Question the Market Is Not Asking
The consensus view is simple: AI capex drives earnings, earnings drive the market. The second-order question is different: who absorbs the cost of the deficit while the AI boom unfolds, and does it come out of the equity risk premium?
Here is the chain. AI-driven disinflation should, in theory, support higher equity valuations by lowering the discount rate applied to future earnings. But the deficit pushes the other way: a larger supply of Treasury debt, with a term premium attached, raises the risk-free rate at the long end and forces investors to demand more compensation for holding stocks over bonds. If long-term yields rise because of fiscal supply rather than growth optimism, the discount rate rises even as inflation falls - a combination that compresses valuations even while earnings grow.
The transmission runs through the Fed as well. New Fed Chair Kevin Warsh inherits a tangled picture: AI pulling inflation down, a deficit pushing long rates up, and a softening labor market pulling toward easing. If the Fed cuts rates to support employment while the bond market prices in fiscal risk, the yield curve steepens on the long end - exactly the outcome that tightens financial conditions for mortgages, corporate debt, and small business even as the policy rate falls. Fleming's optimism about AI and his worry about the deficit are not two separate views. They are the two inputs into the same policy trap.
The third-order implication is about concentration. The AI capex is concentrated in five companies. The earnings growth that has lifted the index is concentrated in the same group. The debt funding the build-out is concentrated in investment-grade issuance from those same balance sheets. A single-engine economy is more fragile than a broad one, and Fleming's own observation that wealth creation is broadening has not yet been matched by broadening in the capex or the earnings.
The Counter-Thesis, and What Would Prove It Right
The strongest argument against this reading is that it underestimates how fast AI revenue will arrive. The bull case holds that AI agents will generate measurable enterprise revenue within quarters, not years - that the capex is not ahead of demand but barely ahead of it. In that world, productivity data will accelerate through 2027, corporate margins expand without pricing power, the Fed cuts into strength, and the deficit becomes manageable because nominal GDP grows faster than debt. The St. Louis Fed's transcript analysis and the early productivity prints are the footholds for this view.
That case is coherent. But it fails if one thing happens: if capital expenditure keeps rising while revenue per dollar of AI spend stagnates. The falsifying signal is specific and observable. Watch the hyperscalers' AI revenue growth versus their AI capex growth over the next four quarters. If combined AI-related revenue growth stays below half of combined capex growth for two consecutive quarters - while capex guidance keeps rising - the cyclical-overreach thesis is confirmed and the market's structural pricing is wrong. A secondary signal: if the 10-year Treasury yield climbs above 5% while core inflation stays at or below 2%, the fiscal term premium has overwhelmed the disinflation trade, and the discount-rate support for equities evaporates.
What to Watch, and the Bottom Line
The practical implication for investors is a split by time horizon.
In the short term - the next six to twelve months - the risk is a cyclical peak in AI capex meeting a softening labor market and a fiscal bid under long-term yields. The beneficiaries are the infrastructure suppliers with contracted backlog and the companies already monetizing AI usage. The exposed are the equity index at record valuations, rate-sensitive sectors such as housing and regional banking, and any company whose valuation assumes AI revenue arriving faster than customers can deploy it.
In the medium term - one to three years - the outcome depends on the revenue-versus-capex ratio described above. If AI revenue catches up, the productivity dividend lifts broad participation beyond technology and Fleming's optimism is vindicated. If it does not, the capex cycle turns and the deficit's interest burden becomes the dominant market narrative.
In the long term - five years and beyond - the structural case for AI is largely independent of the cycle. If AI agents embed into business processes the way prior general-purpose technologies did, the economy's potential growth rate rises, and the debt problem becomes a political choice rather than a mathematical trap. That is the version of the future Fleming is betting on.
The catalysts to watch, in order: the September 4 jobs report and whether payroll growth stabilizes or deteriorates further; the September 15-16 Federal Open Market Committee meeting and whether policymakers hold the target rate at 3.50% to 3.75% or signal a hike; the next round of hyperscaler earnings and capex guidance versus AI revenue disclosure; and the behavior of the 10-year Treasury yield relative to core inflation.
Fleming is right that AI is the best growth story the U.S. economy has had in decades, and he is right to worry that the deficit could steal it away. The market, for now, is pricing the first conviction and discounting the second. That gap - between a structural technological shift and a cyclical, debt-funded build-out - is where the risk and the opportunity both live.
Explore more exclusive insights at nextfin.ai.
