Everything About AI
NextFin News - The 10-year Treasury yield has pushed decisively above 5% for the first time since 2023, and Ruchir Sharma, chairman of Rockefeller International, says the threshold he flagged as the AI bubble's tripwire is now in play. "This breach would signal the start of a new era of tighter money, in which AI mega projects will be harder to fund," Sharma wrote in a recent column, arguing that rising public borrowing costs will "hit the bubbly AI markets harder" than the rest of the economy.
The warning, first published in his analysis and repeated on television this month, lands as the bond market completes a two-year migration away from the ultra-cheap money that financed the AI buildout. The 10-year yield finished September 25 at 5.17%, after reaching as high as 5.23% during the week - the highest levels since 2007. For a market that priced a decade of AI capital spending on the assumption that 4% was a ceiling, 5% is not just another basis point. It is the rate at which the math on the largest data-center projects stops working without fresh equity, and fresh equity is exactly what Sharma says is about to flood in at the worst possible time.
The Warning and the Trigger: Why 5% Is the Line
Sharma's argument is deliberately mechanical, not apocalyptic. The AI boom is not being killed by a single bad earnings print or a regulatory crackdown. It is being squeezed by the price of money itself, and 5% is the hinge. Below it, the hyperscalers can fund $100 billion capital-spending programs with a mix of cash flow and cheap debt, and the market rewards them for growth. Above it, every incremental dollar of AI investment competes with a risk-free asset paying five percent, and the hurdle rate for "mega projects" resets higher.
The mechanism runs through two channels, and the second is the one most investors are missing.
The first channel is the discount-rate channel that every strategist knows: higher long-term yields compress the present value of future cash flows, and AI stocks, with the longest duration of any sector, take the hit first. That is the trade everyone is positioned for, and it is already priced into the 5.9% one-day drop in the PHLX Semiconductor Index on September 14.
The second channel is the funding channel, and it is slower and more damaging. Sharma's point is that the AI buildout is not just valued on cheap money - it is financed by it, and by a capital market that is currently in late-stage bubble mode. "Equity issuance is booming and consistent with a late-stage bubble," he wrote. "Given the pipeline of IPOs and follow-on offerings, another gusher of stocks is on the way. In the past, such activity has signaled a bubble's end is just months away, not years." When the 10-year yield sits at 5%, the government's own borrowing absorbs a larger share of available savings - U.S. debt has exceeded 100% of GDP, and debt-servicing costs are much higher now - and private borrowers get crowded out. The squeeze does not arrive as a crash on a headline day. It arrives as a gradual inability to roll projects at the old cost of capital.
"Rising public borrowing costs will squeeze other borrowers sooner, and hit the bubbly AI markets harder."
That is the sentence that separates Sharma's call from the routine "bonds are expensive, stocks are expensive" commentary. He is not saying AI is a fraud. He is saying the funding structure that carried it is regime-dependent, and the regime has changed.
In his framework for spotting bubbles, the AI-led U.S. market now ticks every box: it is, in his words, "overvalued, overinvested, overleveraged and, perhaps most strikingly, over-owned." The last condition is the one that makes the unwind violent when it comes. When nearly everyone already owns the trade, the only remaining buyer is leverage - and leverage is precisely what a 5% risk-free rate prices out of the market.
The Evidence: A Market That Already Flinched
The market's first reaction to the 5% breach was immediate and concentrated. On September 14, as the 10-year yield briefly crossed the threshold ahead of the Federal Reserve's meeting, AI-linked equities sold off worldwide. Nvidia fell 3.4%, Micron dropped more than 5%, and Broadcom and Advanced Micro Devices each lost more than 4%. The Nasdaq closed down 0.56% at 26,186.41, the S&P 500 down 0.48% at 7,619.94, and the Dow Jones Industrial Average down 0.29% at 52,421.17. Eight of the 11 S&P 500 sectors declined, led by information technology at -1.68%.
Two details in that tape matter more than the index points. First, volume ran at 15.3 billion shares against a 20-session average of 14.8 billion - a heavy-volume decline, which is distribution, not a thin-market wobble. Second, the Nasdaq recorded 47 new 52-week highs against 202 new lows. That 4.3-to-1 ratio of new lows to new highs inside a market that is still near record levels is the signature of a narrowing leadership base - the exact condition that precedes a broader unwind when the leaders finally break.
The selloff was not caused by yields alone. The same session saw Brent crude settle 1% higher at $105.68 a barrel after strikes on Saudi energy infrastructure, and the leaders of Anthropic, OpenAI and xAI had just called for a slowdown in AI development over safety concerns. Three forces - energy-driven inflation, a yield breach, and a credibility shock from inside the industry - arrived in the same week. That convergence is why the 5% level deserves more attention than a round-number technical break normally would.
Then the Fed acted. On September 17, the Federal Open Market Committee voted 12-0 to raise the federal funds target by 25 basis points to a range of 3.75%-4.00%, the first rate increase since July 2023. "The plain fact is that inflation is too high and has been for too long," Fed Chair Kevin Warsh said. The move was widely anticipated - a futures-based probability tool showed a 90% implied chance of the hike as of September 14 - but the accompanying signal was not benign. The median projection in the Fed's September Summary of Economic Projections pointed to one more quarter-point increase in 2026, and 16 of 18 participants indicated they expect at least one more hike this year. A central bank that is still hiking into a slowing growth backdrop is not providing the safety net that carried the 2023-2024 rally.
Cyclical or Structural: The Regime Question
This is where the analysis has to make a call, because the wrong one inverts the conclusion. The correct reading is that both forces are present, and they run on different clocks.
The AI valuation excess is cyclical. It is a classic late-stage bubble pattern - overvaluation, overinvestment, overleverage, and over-ownership - and it will mean-revert. The evidence for the cyclical read is the historical pattern: equity issuance surges and IPO gushers have preceded bubble ends by months, not years, in every major cycle from the 1929 utilities boom to the 2000 tech peak. The 10-year yield's behavior through the dot-com era reinforces this: it traded in a range with 5% as the upper bound for two decades, and each breach was followed by a repricing of risk assets. Cyclical calls require mean-reversion evidence, and here it is abundant.
But the rate regime itself is structural. This is not a cyclical spike that will fade when the next recession hits. The drivers are permanent: a U.S. debt burden above 100% of GDP that requires continuous large-scale issuance; a term premium that has reappeared after years of suppression by quantitative easing; and an inflation floor lifted by deglobalization, energy shocks, and fiscal dominance. Luis Alvarado, co-head of global fixed-income strategy at Wells Fargo Investment Institute, told reporters that clients are being told "normal for longer," because the factors pushing yields higher "are here to stay." The 10-year yield traded at 1.3% five years ago. It is now above 5%. That is not a cycle reverting; it is a regime changing.
The practical implication is uncomfortable for portfolio construction. If the yield move were cyclical, the playbook would be simple: wait it out, because cheap money returns. If the regime is structural, cheap money does not return on its own, and every asset class priced on the old regime - not just AI stocks, but commercial real estate, long-duration growth, and leveraged private equity - has to reprice to a new equilibrium. Sharma's warning works precisely because he is pairing a cyclical bubble with a structural rate shock. The bubble would be survivable in a 2% world. It is not survivable, in his view, in a 5% world.
The Counter-Thesis: Why the Bubble Call Could Be Wrong
The strongest case against Sharma is not that yields won't stay high. It is that the AI boom is fundamentally different from the dot-com bubble, and that earnings, not hype, are carrying it now.
Christophe Fouquet, chief executive of ASML - the near-monopoly supplier of the extreme-ultraviolet lithography machines required to make advanced chips - said in a May 2026 interview that "demand on AI is coming so strongly that we will be in a supply-limited market for quite a while." The revenue is real: AI-related server demand is showing up in data-center construction, utility interconnection queues, and the earnings of the chipmakers themselves. The PHLX Semiconductor Index, even after the September pullback, still carried a 57% gain for the year as of mid-September. A sector that is merely frothy does not deliver that kind of fundamental follow-through.
The second leg of the counter-thesis is that the yield move is normalization, not regime change. After years of central-bank suppression, a return to 5% on the 10-year is a return to the long-run average, not an anomaly. Bond strategists have argued exactly this: yields are merely normalizing, and equities have spent two years adjusting to a higher-rate world. The S&P 500, they note, traded at 19 times expected earnings as of mid-September - its cheapest multiple since April 2025 - which leaves little valuation risk in the broad index. On this read, the damage is confined to the unprofitable fringe, and the AI leaders with real cash flow simply rotate into the leadership slot that quality always takes in a higher-rate environment.
There is force in both arguments, and they deserve weight. But they both rest on one assumption: that AI revenue growth continues to outrun the rising cost of capital. That is the hinge. If the return on AI investment fails to materialize at the scale the capex plans assume - if enterprise adoption plateaus, if the marginal dollar of GPU spending produces diminishing revenue - then the "real earnings" defense collapses, because the earnings were underwritten by the next round of spending, which is now more expensive to fund. The bubble is not in today's earnings. It is in the assumption that today's capex produces tomorrow's earnings at the same rate.
What to Watch: The Signals That Settle It
The thesis is falsifiable, and the signals are concrete. The primary trigger is the yield itself: if the 10-year Treasury yield closes above 5% for two consecutive month-ends while hyperscaler capital-spending guidance remains intact, the "transitory breach" defense dies and Sharma's funding-squeeze scenario becomes the base case. The yield finished September 25 at 5.17%; one more month like that, and the breach is no longer a headline - it is the new floor.
The second signal is issuance. Sharma's bubble framework depends on the "gusher of stocks" - IPOs and follow-on offerings - arriving at the top. Watch the IPO calendar and secondary supply in the AI supply chain through the fourth quarter. A wave of insider selling and late-cycle offerings at elevated multiples would confirm the late-stage bubble read. Conversely, if issuance stays dormant and the leaders fund capex from cash flow, the bubble thesis loses its funding-channel mechanism.
The third signal is the Fed's reaction function. A central bank that hikes into a growth slowdown is the classic policy-error setup, and it is what turned the 2000 and 2007 cycles into crashes rather than corrections. If inflation prints cool and the Fed pauses after the next hike, the soft-landing path remains open. If it keeps tightening as growth rolls over, the funding squeeze accelerates.
Outlook: Three Scenarios, Not One Line
Base case. The 10-year yield holds in the 5%-5.5% range, the Fed delivers one more hike in 2026 and pauses, and AI leaders report solid earnings but guide to slower capex growth. The result is a 15%-25% correction in the most extended AI names, with the PHLX Semiconductor Index giving back a meaningful share of its 57% year-to-date gain as of mid-September, while the broader market grinds sideways. The bubble deflates in the unprofitable fringe; the cash-generative leaders survive but rerate lower.
Downside case. Yields push toward 6% - a level Ben Emons, founder and CIO of FedWatch Advisors, said in late September is highly probable by January 2027 - while AI revenue growth disappoints and a wave of IPOs and secondary offerings hits the market simultaneously. That is the 2000 analog: a funding squeeze meeting an earnings disappointment, producing a 30%-40% peak-to-trough decline in AI-linked equities and a broad-market drawdown as the leadership base finally breaks.
Upside case. The 10-year yield fails to hold above 5%, falling back below the threshold on a growth scare that forces the Fed to pivot, while AI adoption accelerates and enterprise ROI data validates the capex plans. In that world, the September selloff becomes the buying opportunity the dip-buyers expected, and the bubble call is revealed as a timing error rather than a direction error.
Across all three scenarios, the asymmetry favors caution. The upside case requires two things to go right - yields to retreat and adoption to accelerate. The downside case only requires one thing to go wrong: the cost of capital to stay high while the returns fail to arrive.
The kicker: This is not a bet that AI is a fraud. It is a bet that the price of money is a regime variable, not a cycle - and that a boom built on 1.3% capital cannot survive a 5% world without a funding crisis. The 10-year yield has already sent the invitation. The next two months will show whether the AI market accepts it.
Explore more exclusive insights at nextfin.ai.
