US earnings forecasts are overestimated! Yardeni warns: triple constraints of "money, votes, and power" are pressing down, and the market has overlooked AI's "speed bumps."

Yardeni Research noted in a November 9 report that S&P 500 earnings forecasts are overly optimistic: the market consensus has yet to reflect the three types of constraints the firm expectsfinancial, political, and physicalthat will weigh on the AI investment boom. The gap is concentrated in 2028: the market consensus for that year's operating earnings per share (EPS) is $495, while Yardeni's own forecast is $450, a difference of $45; for 2027 the two sides barely differ, with consensus at $426 and the firm at $425. In other words, Wall Street and Yardeni broadly agree on earnings for the next two years, and the entire disagreement rests on 2028the point of contention is not "whether AI is good now," but "in which year the slowdown in AI capital spending will land." "We suspect there is some irrational exuberance in the earnings forecasts, because they do not even reflect a speed bump," the report said. The firm also said analysts do not expect AI infrastructure construction to come to a complete haltthe fact that consensus earnings forecasts are still being revised upward precisely shows that the market is assuming a "smooth ride." Two other numbers are worth noting: semiconductors account for about 20% of S&P 500 forward earnings; Yardeni believes EPS may not reach $500 until 2030, two years later than consensus. Three constraints: money, votes, and powerAI investment may slow The report breaks the constraints facing the AI investment boom into three categories, each of which is tightening. Financial constraintfinancing will become more expensive as investors demand clearer returns on hyperscalers' spending. This is especially glaring in the current rate environment: the October nonfarm payrolls report released on November 6 showed an increase of 165,000, more than double the market expectation of 62,000, the unemployment rate fell to 4.1%, traders' bets on a Fed rate hike in December rose to about 79%, and the 10-year Treasury yield rose to 4.86%. Rising financing costs are precisely the most direct stone pressing on data center return calculations. Political constraintdata center developers may be required to fund grid upgrades, protect local water resources, and provide more community compensation. This is not an abstract risk: Gina Hinojosa, elected governor of Texas, said explicitly in an interview during election week that before measures to protect Texas residents are put in place, the state needs to "stop what it is doing right now"referring to data centers. Texas is one of the most concentrated landing spots for this round of AI infrastructure, and approval authority rests with the state and grid operators. Physical constraintpower shortages, grid interconnection queues, and equipment supply bottlenecks could all delay projects and lengthen the wait for free cash flow to turn positive. These constraints are the hardest to hedge with policy: interconnection queues often run for years, and lead times for transformers, gas turbines, and energy storage equipment are likewise measured in years. The market has priced it in, but only half of it It is worth noting that equity investors appear to have already been pricing in these constraints. The S&P 500's forward P/E has fallen to 17.6x, below the 18.1x at the time of "Liberation Day" (the April 2025 tariff shock)Yardeni uses this frame of reference to show that the valuation-side correction has already been substantial. Last week's trading provided a footnote to this judgment: in the week ended November 6, the S&P 500 fell 2.7%, the information technology sector fell 6.9%, the Philadelphia Semiconductor Index plunged 12% in a single day on Thursday, and the Nasdaq Composite fell 3.4% for the week, closing nearly 12% below its October record high and formally entering a technical correction; the S&P 500 also posted its largest single-day drop since April 2025 last week. Yardeni directly attributed Thursday's plunge to post-midterm-election market worries about political resistance to data centersthe first time its listed "political constraint" was priced on the tape. Friday saw a rebound: the S&P 500 closed at 7,376.86, up 0.32%, the Nasdaq rose 1.1%, the Philadelphia Semiconductor Index rebounded 4.2%, NVIDIA Corporation rose 2.8%, and AMD rose 11%. But for the week overall, the major indexes still closed lower across the board. So the question the market must answer this week is on the table: is this a "correction," or the start of a "crash"? Three variables determine the answerthe political outlook for data center construction, whether investor anxiety begins to recede, and whether inflation data leave the Fed room not to hike. Why still bullish: no evidence of a slowdown on the demand side Alongside the three constraints, the report's conclusion is not bearish but "slowdown rather than termination." The support comes from the hardest set of demand-side data: Anthropic disclosed in its prospectus that as of September 30, 2026, it had more than 11,900 customers spending over $100,000 annually, versus about 950 a year earlier; of those, more than 1,900 spent over $1 million annually and about 250 spent over $10 million, with a net revenue retention rate of about 900%. A twelvefold expansion in paying customers in a year is the most direct evidence that "compute demand is still accelerating." The report also expects increasingly autonomous AI agents to take on complex tasks, unleashing the next wave of compute demand while bringing productivity gains. This contrasts with the constraints on the supply side of compute: the demand curve is climbing steeply while the supply curve is held back by grids and approvalsthe result is not that demand disappears, but that price, siting, and construction pace are repriced. The real asymmetry: valuations fall first, earnings have not yet In the same day's QuickTakes note, Ed Yardeni added another comparison: semiconductor equipment stocks fell 14.5% for the week, while analysts' forecasts for the sector's earnings growth in 2026 and 2027 are 35.4% and 51.5%, respectively. This set of numbers makes the issue clearinvestors have already cut valuations for AI infrastructure suppliers, but earnings forecasts have not yet been cut. "Modest cuts to earnings estimates are probably already priced in; big cuts are not," Yardeni wrote. This also explains the nature of the $45 gap: it is not a warning about 2027, but a "conditional judgment" about 2028. If the three constraints only delay projects by two quarters, the difference between 450 and 495 is irrelevant; if they cause a genuine step-down in capital spending in 2028, then the current consensus is on the wrong side. The other side of the sell side: the disagreement is not "whether to have AI," but "in which year the bill comes due" Yardeni is not the only institution adjusting its stance, but the directions are not consistent. Citi on November 9 downgraded US and global tech stocks from overweight to neutral, while upgrading European equities excluding the UK to overweight, citing the latter's lower exposure to AI turmoil and greater benefit from falling oil prices; its head of strategy Beata Manthey warned that positioning in key indexes such as the Nasdaq "could further reset into net short territory." Goldman Sachs Group, Inc. added a comparison the same day: AI infrastructure accounts for about 15% of the STOXX 600 but contributed more than two-thirds of the S&P 500's earnings growth next yearconcentration itself is a source of risk. There are voices on the bullish side as well. Truist Advisory Services chief investment officer Keith Lerner said after the S&P's rebound on Friday that "we still believe AI is the dominant theme of this bull market"; LPL Financial chief cross-asset strategist Adam Turnquist put it more vividly: "AI has been put on trial multiple times in this cycle and acquitted every time," calling last week's selloff "a bit of a knee-jerk reaction." Putting these together, the nature of the $45 gap becomes clearer: it is not a "bearish/bullish" question of taking sides, but a timing question of "in which year the bill comes due"consensus believes $495 can be reached in 2028, Yardeni says $500 will not be seen until 2030, and Citi's approach is simply to move first to a market where this bill does not have to be calculated. Three charts: politics, volatility, yieldsthree clues, each watching one end Three charts depicting the VIX, Treasury yields, and data center stocks can offer clues as to whether last week's selloff is losing momentum. The first chart is "politics is picking winners inside the AI trade." The chart tracks month-to-date moves (through last Friday) in semiconductors (SMH), Vertiv (VRT), Constellation Energy (CEG), Clean Energy Fuels Corp. (TAN), and the S&P 500. The conclusion: after Democrats won Congress, construction- and power-related companies came under pressure, while Clean Energy Fuels Corp. instead attracted fundsbecause investors expect a Democrat-led Congress to support more renewable energy development. An easily overlooked logic is that AI expansion depends not only on chips and compute but also on power, land, and approvals, so whether the selloff worsens depends largely on whether local and political resistance makes projects harder to land. Vertiv (VRT.US) is the litmus test for this selloff. The data center equipment maker just delivered strong results and raised guidance last month, yet fell last week without any company-specific news, with peers Eaton Corp. Plc (ETN.US) and nVent (NVT.US) falling in tandem. This is exactly the key sample for distinguishing "business concerns" from "sentiment shift"if the fundamentals have not changed but the stock still falls, what is falling is the valuation, not the business. The other two charts focus on volatility and rates. Under the framework given by the media outlet's subheading, the VIX determines whether panic has peaked, and Treasury yields determine financing costs and the valuation denominator. On Friday the VIX fell 9% to 25.63, but was still up 32% for the week, the largest weekly gain since mid-October; the 10-year Treasury yield rose to 4.86%. These two numbers dovetail precisely with Yardeni's "financial constraint": if yields do not come down, data center financing costs will not come down; if the VIX does not come down, crowded positioning will still need to be cleared out. This week awaits validation from CPI and AI earnings US stock futures rose on Monday as investors assessed whether tech stocks can recover from last week's sharp selloff while awaiting key CPI inflation data and earnings from AI-related companies. Dow futures rose about 400 points, or 0.76%, S&P 500 futures rose 0.36%, and Nasdaq 100 futures rose about 0.5%. Pre-market ETF performance better reveals the structure: SPY rose 0.31%, QQQ rose 1.4%, while DIA instead fell about 0.40%tech leading gains and blue chips weakening, exactly the reverse of last week's pattern of "AI sold off, defensive stocks resilient." In the Stocktwits community, retail sentiment on SPY has improved from "bearish" to "neutral," on QQQ is "bearish," and on DIA is "bullish"a combination of "defensive on the broad market, cautious on tech," indicating that dip-buying appetite is returning but has not yet become consensus. For this report, the real test points are close at hand: CPI determines whether the assumptions on the financing-cost end hold, and AI-related companies' earnings and guidance determine whether the 2028 earnings curve can still hold up. A colder inflation reading could give tech stocks breathing room, but data center construction and political resistance will not disappear because of a single CPI number. Yardeni's judgment is a "speed bump," not a "wall," and the market must answerjust how long is this speed bump.