NextFin News - Technology is no longer the quiet disinflationary force central bankers grew accustomed to over the past two decades, a shift that Wealth Enhancement strategist Ayako Yoshioka addressed in a Sept. 4 television interview, where she weighed how artificial intelligence and the wider tech buildout are feeding through to inflation and consumer prices. The timing matters: U.S. consumer prices rose just 0.1% in July, putting the annual rate at 3.4%, yet beneath that tame headline a quiet reversal is underway in the very categories - computers, software, electronics - that historically pulled inflation down. The question the market is now forced to confront is whether the AI boom is a temporary supply bottleneck in an otherwise disinflationary story, or the first sign that the two-decade "Amazon effect" doctrine has broken.
The Situation: The "Amazon Effect" Is No Longer Automatic
For most of the post-2008 era, the consensus among policymakers was nearly doctrinal: technology is structurally disinflationary. The logic ran that digital competition, automation, and relentless productivity gains would keep unit costs falling, letting central banks hold rates lower for longer. That view is now being stress-tested. In the May consumer-price report from the Bureau of Labor Statistics, released in June, the "computer software and accessories" category jumped a record 14.5% from a year earlier, and Federal Reserve economists themselves flagged surging flash-memory chip prices - which more than doubled over the past year - as a key contributor to the spike. The category that once subtracted from inflation is now adding to it.
This is the backdrop Yoshioka engaged with: a market where the AI buildout is simultaneously a growth engine and a price pressure, where the same data-center spending that powers the Nasdaq rally also tightens supply for memory, power, and components used in ordinary consumer goods. The tension is not theoretical. Goldman Sachs economists estimate AI-related price pressures already added roughly 0.3 percentage points to annual core PCE inflation over the past year, with a similar increment expected over the next 12 months, and they project the boost to core PCE could reach 50 basis points by year-end. Software and accessories prices, they expect, will peak at a 30% year-over-year pace in November 2026.
"Higher electricity demand to power data centers is increasing electricity prices in some US regions, and we expect it to continue boosting inflation over the next couple of years," Goldman Sachs said in research this year.
The scale of the spending behind that pressure is unprecedented. The five largest hyperscalers - Amazon, Microsoft, Alphabet, Meta, and Oracle - are on track to spend between $700 billion and $900 billion on capital expenditures in 2026, a 36% increase over 2025, according to CreditSights estimates, with roughly $450 billion of that tied directly to AI infrastructure. Amazon alone has guided for $200 billion in capex this year. Sequoia Capital's David Cahn has calculated an annual revenue gap of roughly $600 billion between what hyperscalers are spending on AI infrastructure and what the AI ecosystem is actually generating in sales - a gap that is widening as spending accelerates faster than revenue. When capital formation runs that far ahead of monetization, the inflationary impulse comes first by construction; the disinflationary payoff is a promise that matures later, if at all.
The Federal Reserve's own language has shifted to match. In its July monetary policy report to Congress, the central bank said inflation "stepped up further this spring" and cited the "booming artificial intelligence buildout" alongside tariffs and war-related energy costs as a source of price pressure, noting that demand for high-tech equipment associated with AI pushed up prices for computers, software, and electronics. That is a meaningful change in emphasis for an institution that spent the 2010s treating tech-driven price declines as a structural tailwind.
Why Technology Stopped Being Disinflationary - For Now
The mechanism is straightforward once you separate the demand side of the technology cycle from the supply side. In the first phase of a general-purpose technology rollout, the money flows in before the productivity arrives. Firms pour capital into chips, servers, data centers, grid connections, and specialized labor. That demand hits a supply chain - memory wafers, advanced packaging, transformers, copper, skilled electricians - that cannot expand at the same speed. Prices rise for the inputs, and because those inputs sit inside everything from laptops to cars to cloud subscriptions, the pressure diffuses into the broader price level. Only later, if the technology actually delivers, do unit costs fall and the disinflationary payoff arrive.
History offers a template, and it is not the comforting one. When then-Fed Chair Alan Greenspan became convinced in the 1990s that the technology boom would drive inflation lower, he was, by several accounts, early. The productivity acceleration of the late 1990s was real, but it took years to show up in the data, and the investment boom itself ran hot first. Chicago Fed President Austan Goolsbee has drawn the distinction sharply this year: the appropriate policy response depends on whether the economy is in a 1995 moment - a productivity surprise that justifies lower rates - or a 1999 moment, when the boom is widely expected, fully reflected in asset prices, and the Fed should be on alert for inflation. "It really isn't the same situation," Goolsbee said in February, noting that Greenspan merely delayed eventual rate hikes rather than cutting into above-target inflation.
That is the fork in the road the market is trying to price. On one side sits Fed Chair Kevin Warsh, who has argued AI will prove disinflationary through productivity gains and has built formal machinery inside the Fed to study it. In July he named the leaders of five task forces, including an AI panel co-led by venture capitalist Marc Andreessen and Stanford economist Charles Jones, charged with assessing "the economic impact of new general-purpose technologies, including artificial intelligence" and delivering recommendations to the Federal Open Market Committee by year-end. The membership matters: every named participant on the AI panel appears to believe the technology will be transformative for growth and productivity, which is precisely why critics worry the Fed is predisposed to see the payoff before the data confirms it.
On the other side sits the incoming price data, which so far shows the investment phase dominating the payoff phase. Core CPI excluding food and energy came in at 2.5% annually in July, matching the slowest pace since March 2021, and the monthly core print of 0.2% suggests the energy-driven burst from earlier in the year is easing. But the composition of that moderation is important: it reflects falling gasoline and moderating shelter, not a broad-based collapse in goods prices. Personal computers and peripherals climbed 3.5% in July alone, a pickup that Fed researchers attributed partly to Apple's price increases. The disinflation is coming from the categories least connected to the AI buildout; the inflation is coming from the categories most connected to it.
The Second-Order Question: What If the Fed Is Pricing the Payoff Too Early?
The conventional read of the AI-inflation debate is binary - either AI is inflationary or it is disinflationary - and that framing misses the more important transmission channel. The real question is not the direction of prices but the level of the neutral interest rate, the borrowing cost that neither stimulates nor restrains the economy. If AI genuinely lifts the economy's productive capacity, it raises the return on capital, which means the neutral rate should rise, not fall. In that world, a disinflationary AI boom does not give the Fed permission to cut; it gives the market a reason to demand higher real rates.
This is the subtler point that separates a 1995-style supply boom from a demand-led investment cycle. In the 1990s, Greenspan's insight was that productivity gains meant the Fed could hold off on rate hikes, not that it should slash them. Today, with core inflation still above the Fed's 2% target and the AI payoff largely anticipated rather than realized, cutting aggressively into a capital-spending boom risks adding fuel to the very demand that is pushing prices up. The bond market appears to be wrestling with exactly this: the 10-year Treasury yield rose above 4.81% in early September, its highest level since October 2023, even as inflation prints moderated - a signal that term premium, deficit concerns, and growth expectations are offsetting disinflation hopes. A deluge of corporate bond supply to fund the AI buildout has added to the pressure on long-duration debt.
The cross-asset implication is uncomfortable for the cleanest version of the AI trade. If the neutral rate is rising, the discount rate applied to long-duration growth stocks does not fall - it stays elevated. That caps multiple expansion even as earnings grow, which is precisely the regime Wealth Enhancement's year-end commentary described for 2026: returns driven by earnings rather than the multiple expansion that powered 2023-2025. In that environment, the AI winners are picked by cash-flow delivery, not by the promise of future productivity. The practical consequence is a narrowing of the trade: owning the hyperscalers is no longer a substitute for owning the entire technology complex, because the cost of capital that lifted all multiples in the low-rate era is no longer falling.
The Strongest Counter-Thesis: This Is Just a Supply Bottleneck, Not a Regime Change
The most credible case against the "tech is inflationary now" reading is that we are looking at a transient supply constraint, not a structural break. Memory prices double, flash drives spike, electricity grids strain - but these are the predictable pains of a capacity cycle that will correct once new fabs, new power plants, and new data centers come online. Historically, technology categories have a relentless deflationary drift: semiconductors follow learning curves, computing power per dollar rises, and competition eventually restores the old order. Under this view, the 14.5% jump in software and accessories inflation is a blip in a decades-long downtrend, and the disinflationary doctrine remains intact - it is simply running behind schedule.
There is real evidence for this camp. The 0.2% monthly core CPI print in July, the easing in gasoline, and the prospect that software and accessories inflation peaks before year-end all point to a self-limiting impulse. If the AI inflation contribution tops out near 50 basis points of core PCE and then fades, the disinflationary second half of the story could still arrive within the Fed's policy horizon. The 1990s analogy, properly understood, supports the optimists: productivity did eventually accelerate, inflation did fall, and the economy achieved a rare combination of fast growth and stable prices. Goldman Sachs Vice Chairman Robert Kaplan has argued that AI adoption will ultimately prove disinflationary, and the firm's own research frames the episode as "up then down" - inflationary during the buildout, disinflationary once the productivity gains diffuse.
But the counter-thesis has a vulnerability: it assumes the supply response will be fast enough, and it assumes the productivity payoff will be large enough to offset the demand shock. Neither is guaranteed. The $600 billion annual revenue gap identified by Sequoia means the AI buildout is, for now, consuming capital faster than it is generating the cash flows that would fund the next wave of capacity without price pressure. And unlike the 1990s internet buildout - which produced consumer-facing deflation in retail, media, and communications almost immediately - today's AI spending is concentrated in upstream infrastructure whose cost savings may take years to reach the consumer price index. A bottleneck that lasts three years is not a blip in a two-year inflation-targeting framework, and the Fed's own task force will not deliver its recommendations until year-end - long after several more inflation prints have crossed the desk.
What to Watch: The Signal That Settles the Debate
The falsifying test is concrete and near-term. If "computer software and accessories" inflation - which hit a record 14.5% year-over-year in May - does not decelerate materially by the end of 2026, and if core PCE remains at or above 0.3% month-over-month for two consecutive readings into early 2027, the structural-disinflation thesis for AI is wrong for this cycle, and the "investment-first, payoff-later" regime extends further than the consensus expects. Conversely, if memory prices roll over and the category prints flat-to-negative by mid-2027 while productivity data accelerates, the 1990s playbook is back in force.
For investors, the practical split runs by time horizon. In the short term, the inflation impulse from AI spending is a headwind for rate-sensitive assets and a tailwind for the suppliers of the buildout - memory makers, power equipment, data-center real estate, and the utilities feeding the grid. Over the medium term, the earnings-delivery test separates the hyperscalers who can monetize capacity from those carrying stranded assets; the $600 billion revenue gap is the yardstick. Over the long term, if the productivity payoff arrives, the beneficiaries shift to the broad economy through lower unit costs and higher real wages - but that is a conditional outcome, not a base case.
The base case is a two-speed inflation story: technology remains disinflationary in the categories where competition is fierce and capacity is abundant, and inflationary in the categories where the AI buildout is bidding for scarce inputs. That split is more useful than a single verdict, because it tells you where to look rather than what to believe. Three catalysts will determine which speed dominates: the September and October inflation prints, which will show whether the software-and-accessories surge is rolling over; the Fed task force's recommendations, due by year-end, which will reveal whether the central bank is formally embedding AI productivity assumptions into its policy framework; and hyperscaler earnings guidance for 2027 capital expenditure, which will show whether the spending cycle is accelerating or finally plateauing.
Yoshioka's appearance lands squarely in that uncertainty: technology's impact on prices is no longer a one-directional force, and the market that priced AI as pure disinflation is learning that the bill for the future arrives before the discount. The investors who treat the AI trade as a single bet on lower rates are the ones most exposed to that sequencing risk.
Explore more exclusive insights at nextfin.ai.
