Everything About AI
NextFin News - The investment case for artificial intelligence is running on two clocks: markets are already rewarding the spending cycle, while the productivity gains that could justify it remain ahead. In a Nov. 11, 2026 interview, Citi's Olaolu Aganga said strong earnings and attractive valuations continue to support U.S. stocks despite recent volatility, while warning that inflation could stay hotter than expected and push interest rates higher. Her message is neither simply bullish nor bearish. It is a split-screen view in which long-term AI capital formation can coexist with short-term pressure on duration-sensitive assets.
Aganga highlighted the long-term AI investment cycle, opportunities in cybersecurity and critical metals, and the possibility that AI productivity gains have not yet arrived at the scale investors anticipate. She also pointed to shorter-duration fixed income and gold as areas that can be more resilient if inflation keeps rates elevated. The tension matters because the same buildout can lift corporate investment and earnings while raising demand for power, data centers, semiconductors and industrial inputs before it lowers the cost of producing goods and services.
Citi's midyear outlook describes AI investment as a structural driver of earnings growth and capital spending, but also expects inflation to moderate and forecasts a federal-funds rate of 3.25% by year-end 2026. Separately, Citi Private Bank has argued that inflation risks may be underpriced, citing a five-year inflation breakeven of 2.21% in its Aug. 19 bulletin. The Federal Reserve's research frames the central uncertainty more cautiously: investment can lead productivity by years, and the absence of a large economy-wide signal in 2026 does not settle the long-run question.
The market question is therefore narrower than whether AI is real. It is whether the economy will receive the supply-side benefit before the financing, labor and commodity demands of the buildout keep inflation and bond yields higher. The evidence points to a structural technology shift wrapped in a cyclical valuation and rates trade.
The AI Trade Has Moved From Promise to Capital Spending
The first-order effect of AI is familiar: companies are spending to build computing capacity, and suppliers are receiving orders. The more important mechanism is the conversion of those orders into economy-wide productivity. Until that conversion occurs, the market is capitalizing expected future cash flows while the economy absorbs present costs.
Citi's midyear outlook says the AI investment cycle is broadening beyond the initial group of beneficiaries. That distinction matters. A narrow infrastructure boom can support semiconductor, networking and data-center businesses; a broader cycle can lift software, cybersecurity, industrial automation, energy infrastructure and critical-material demand. Aganga's emphasis on cybersecurity and critical metals fits that transmission chain: the AI economy needs protection for increasingly valuable digital systems and physical inputs for the infrastructure that runs them.
Broadening participation is not the same as realized productivity. The Federal Reserve's July research note says measured gains from general-purpose technologies have historically lagged investment by years. It also warns that hyperscaler capital expenditure is an imperfect AI proxy because it includes non-AI spending and may understate capacity obtained through leases. Headline investment may overstate the near-term economic payoff, while the same headline may undercount the eventual scale of the buildout.
That gap explains why valuation and inflation can move in opposite directions. Equity investors can look through current spending toward future margins. Bond investors must price the near-term supply of debt, power demand, labor demand and inflation uncertainty. The longer the productivity payoff takes to appear, the more the discount rate matters.
“Some participants remarked that productivity gains associated with AI adoption would eventually reduce production costs and increase aggregate supply, which should put downward pressure on inflation, though they noted this effect would likely take time to materialize.”
The Federal Reserve Bank of Minneapolis quoted that passage from the Federal Reserve's June meeting minutes in its 2026 analysis. The wording captures the sequencing issue: the investment arrives before the full supply response.
The short-term conclusion is cyclical: the intensity of AI capex can rise and fall with financing conditions, corporate budgets and investor enthusiasm. The long-term conclusion is structural: if firms reorganize production around AI, the competitive map and the economy's potential supply can change even after a particular investment boom cools.
Inflation Is the Missing Link Between AI Spending and Asset Prices
Why does inflation matter to an AI market story? Because the transition first creates demand and only later, if adoption succeeds, creates supply. Data centers require construction, electricity, chips, networking equipment and specialized labor. Those inputs can become more expensive before AI tools lower costs in customer-facing industries.
Citi Private Bank's August bulletin argued that inflation had moderated but remained above target and that markets might not be adequately compensating investors for persistence. It cited five-year inflation breakevens of 2.21%. That figure is not a forecast of realized inflation, but it provides a market-based measure of compensation embedded in nominal bonds relative to inflation-protected securities. A higher-than-anticipated path would challenge long-duration equities and longer-maturity bonds at the same time.
Aganga's warning that inflation could be hotter than expected therefore reaches beyond the consumer-price index. Higher inflation can keep policy rates elevated; higher policy rates lift the hurdle rate applied to distant AI cash flows; and higher real financing costs can separate companies with current earnings from companies valued mainly on future scale. The second-order effect is a rotation inside the AI trade, not necessarily an end to it.
That is why shorter-duration fixed income can look more defensive than long-duration bonds in this framework. Short maturities reset sooner when rates change, while long maturities carry more price sensitivity to shifts in inflation expectations and term premiums. Gold occupies a different role: it does not pay a coupon, but it can benefit when investors seek an asset less directly tied to the real value of fiat cash or to corporate earnings multiples. These are analytical channels, not guarantees of performance.
The market may already understand the first-order relationship between hotter inflation and higher yields. The less obvious point is that AI can initially reinforce the inflation problem it is expected to solve. If electricity, construction and specialized labor are bottlenecks, the capital cycle raises marginal costs before automation expands effective supply. A productivity boom is disinflationary only after it becomes operational, not when it is merely being financed.
Cyclical Enthusiasm, Structural Technology
The evidence supports a two-part call rather than a single verdict. AI-related valuation excess and the sensitivity of technology shares to rates are cyclical; the diffusion of AI into production, cybersecurity and industrial systems is potentially structural. Confusing the two creates the wrong market conclusion.
The cyclical evidence is visible in the timing mismatch. Capital spending can accelerate before measured productivity, and it can be delayed quickly when financing costs rise even if the underlying technology remains useful. Rate-sensitive growth assets are especially exposed to that mismatch because their valuations depend on cash flows far in the future. The proper cyclical inference is not that AI is a failed technology cycle; it is that returns can mean-revert around changes in liquidity, rates and earnings delivery.
The structural case rests on a different set of facts. AI is being incorporated into software, security, customer service, coding and industrial processes rather than remaining a single product category. Citi describes the investment cycle as broadening, and the Federal Reserve is building indicators to track adoption, labor-market effects and sector productivity. A technology that changes how firms perform tasks can survive a capex slowdown; the spending rate is cyclical, but installed capability can keep diffusing.
The critical test is whether productivity shows up outside the firms selling infrastructure. The Federal Reserve says the absence of a large aggregate signal in 2026 does not invalidate future gains, partly because adoption remains shallow and historical productivity gains can lag investment. The Minneapolis Fed, however, notes that software price indexes have been flat to rising since generative AI emerged, offering no clear ground-level evidence that AI has yet produced broad disinflation. Both observations can be true: the technology may be structurally important while its macroeconomic payoff remains unproven.
That is the expectation gap. Equity markets can price a long-run supply shock before national accounts confirm it. If the supply shock arrives, today's infrastructure spending can be defended as the cost of building a more productive economy. If it does not, investors are left with high capital intensity, uncertain monetization and rates that remain restrictive.
The Counter-Thesis: AI Could Be Disinflationary Sooner Than Expected
The strongest argument against the cautious rates view is that AI adoption could lower costs faster than infrastructure demand raises them. Software can scale at low marginal cost. Automated workflows can reduce labor requirements. Better forecasting can lower inventory waste. Competition can force firms to pass productivity gains through to customers rather than retain them as margins. Under that thesis, AI would expand aggregate supply, lower inflation and eventually allow central banks to ease without sacrificing growth.
The Federal Reserve discussion preserves this possibility, and the Minneapolis Fed describes productivity gains as a potential structural force that dampens inflation. Citi's constructive equity outlook also assumes that earnings and capital spending can broaden as inflation moderates. If AI becomes a supply shock quickly, the cautious duration argument would be too conservative: long-term yields could fall with inflation expectations, while high-growth equities could benefit from stronger earnings and a lower discount rate.
That counter-thesis is serious because it attacks the central judgment: perhaps investors do not need to wait years for the productivity payoff. Yet the current evidence does not establish that timing. The Fed's own research says measured gains have not broadly materialized, and the Minneapolis Fed finds no clear disinflation in software prices. The investment-to-productivity lag remains the key uncertainty.
The falsifying signal for the cautious view is specific: if broad labor productivity accelerates materially for several quarters while core inflation falls toward target despite continued AI-related investment, the claim that the buildout is inflationary before it is disinflationary would be wrong. Conversely, if core inflation remains persistent while long-term yields rise and AI adoption fails to broaden beyond infrastructure suppliers, the disinflationary counter-thesis would lose force.
There is a further risk to the bullish market narrative. Even successful productivity can hurt the valuation of individual companies if it spreads faster than those companies can defend margins. The winners may be users of AI rather than only its first-wave suppliers. Cybersecurity and critical-material businesses can benefit from the buildout, but their earnings still depend on prices, competition and capital discipline rather than on the label “AI.”
What Aganga's View Means Across Time Horizons
In the short term, the dominant variable is liquidity. A hotter inflation surprise or a repricing of rate expectations can pressure long-duration equities and long-maturity bonds even if AI earnings estimates remain intact. Shorter-duration fixed income and gold respond to different parts of that shock, which is why Aganga identified them alongside the AI opportunity set rather than treating the market as a single trade.
Over the medium term, earnings breadth matters more than the narrative. Citi expects AI participation to widen beyond the initial beneficiaries, but the market will need evidence in operating margins, revenue productivity, software adoption and capital returns. The key comparison is not whether AI spending is large; it is whether the earnings generated by that spending grow faster than the financing and operating costs required to sustain it.
Over the long term, the structural question is whether AI raises potential output enough to alter the relationship between growth, labor demand and inflation. A genuine supply expansion would support growth with less price pressure and could eventually improve the case for lower rates. A slower diffusion path would leave the economy with a capital-heavy investment cycle and greater sensitivity to energy, labor and financing bottlenecks.
The base case is a delayed productivity payoff: AI investment remains a major earnings and capital-spending theme, but inflation and rates stay volatile enough to punish the most duration-sensitive valuations. The upside case is faster diffusion across non-technology sectors, with productivity gains lowering unit costs and validating the structural bull case. The downside case is that infrastructure spending keeps demand pressure high while monetization disappoints, producing weaker returns and a higher-for-longer rates regime.
The next evidence should come from three places: company disclosures showing AI-driven revenue or cost savings outside infrastructure suppliers; official productivity and labor data showing whether adoption is lifting output per hour; and inflation and bond-market measures showing whether the buildout is adding persistent demand pressure. A single market rally cannot resolve the question. The mechanism must appear in the data.
Aganga's central message is best read as a sequencing argument. AI may ultimately be disinflationary, but markets must first finance the infrastructure, absorb the bottlenecks and wait for adoption to spread. The near-term trade is therefore about rates and duration; the long-term verdict depends on productivity.
AI is not yet a proven disinflation shock; it is a structural investment cycle carrying a cyclical inflation bill.
Data cutoff: Nov. 11, 2026.
Explore more exclusive insights at nextfin.ai.
