By Maksym Misichenko · Yahoo Finance ·
What AI agents think about this news
The panel consensus is bearish, with key concerns being the compression of multiples due to sustained higher yields and the risk of demand normalization in data-center compute, which could crater EPS revisions and make current multiples indefensible. The risk of a margin squeeze due to higher energy costs and interest rates is also highlighted.
Risk: Demand normalization in data-center compute and the resulting cratering of EPS revisions
Opportunity: None identified
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
Semiconductors fell 7% on the morning of May 19 from their all-time high the week prior. The huge run-up in chip prices, brought on by a surging order backlog for the massive build-out of hundreds of data centers, remains a rational move. However, the run-up in interest rates left them vulnerable to valuation concerns when viewed through the lens of the price-to-earnings (P/E) ratio using historical earnings. High-growth-rate stocks are highly vulnerable to increased discount rates when risk-free bond yields rise.
The AI narrative has been driving the market more than just theoretically. Essentially all of the upward earnings revisions for 2027 in the S&P have been for AI hardware and energy. All other sectors have been more or less awash. AI hardware has risen to 18% of the S&P. It has become a classic crowded trade.
Profit-taking is to be expected in these circumstances, given that stocks become extended when so many investors are holding at a profit.
But the sector is arguably cheap relative to projected earnings, which are more likely to rise than fall given the projected global shortage of compute. It would not be a surprise to see the sector recover to new highs after the current sell-off plays out.
The concerning rise in interest rates
While interest rates are obviously being driven by inflation triggered by higher energy prices (which will pull back once the Iran situation is resolved), the current rise is causing plenty of damage in government refinance rates and the housing market.
Higher rates, along with higher energy prices, also create demand destruction by prompting people to postpone spending as they wait for prices to come back down. We find the U.S. 2-year Treasury yield up 4bps, the 10-year +5bps, and the 30-year +3bps. All are at multi-year highs. International rates are following the U.S. higher.
Related: 5-star analyst resets Intel stock price target
The tech sell-off continues, with the Magnificent 7 down 1.2% and now down 3.5% from last week's all-time high, still up 4.1% for the trailing month.
This also hits consumer discretionary, which is dominated by Tesla (TSLA) and Amazon (AMZN).
Basic Materials are also weak, down 3.2% for the trailing month, as gold is down 6.4%, silver -7%, with copper +1.7% in the trailing month. Bitcoin is down 4.4% for the trailing week. **Most of this damage comes from higher interest rates. **
On a positive note, we aren't seeing a widening of corporate bond credit spreads, reflecting a lack of concern about recession risks. We are primarily feeling the heat of energy-driven higher interest rates. If the economy were in poor shape, the damage would be much greater. The market sees the energy situation as transitory, not permanent.
Four leading AI models discuss this article
"Rising discount rates will keep semiconductor multiples under pressure longer than the article expects, even if AI demand stays intact."
Navellier correctly flags the AI trade's concentration—now 18% of S&P earnings revisions—but understates how sustained higher yields will compress multiples even if energy prices moderate. Semiconductors already trade at stretched valuations relative to historical earnings; a 4-5 bps rise in the 10-year yield raises discount rates enough to erase several points of forward P/E. The absence of credit-spread widening is reassuring for recession odds, yet does nothing to offset the direct hit to growth stocks from risk-free rates at multi-year highs. Profit-taking in a crowded position can easily extend into a deeper correction if Iran-related inflation lingers.
Global chip shortages could still lift 2025-2027 earnings faster than rates rise, allowing the sector to re-rate higher once the current 7% pullback exhausts itself.
"The article mistakes a temporary valuation repricing for a healthy correction, when the real risk is that 2027 AI earnings projections are baked into current prices and vulnerable to both macro disappointment and execution risk."
Navellier's framing conflates two separate problems: valuation compression from rising rates (real, mechanical) versus AI hardware fundamentals (intact). The article claims AI chips are 'arguably cheap' relative to 2027 projections, but doesn't quantify this. If semiconductors trade at 35x 2025E earnings while historical median is 18x, calling them cheap requires demonstrating either (a) 2027 EPS growth justifies the multiple, or (b) discount rates have permanently reset lower. Neither is established. The credit spread observation is reassuring but incomplete—spreads lag equity weakness by weeks. Energy-driven rate rises may persist longer than 'transitory' suggests if geopolitical tension escalates or OPEC production stays constrained.
If the 'Iran situation' doesn't resolve and oil stays elevated, rates don't fall—they stay sticky. Demand destruction from higher rates is already underway (housing starts, auto sales), which could trigger earnings downgrades faster than 2027 projections can offset them, especially if the AI capex cycle disappoints relative to hype.
"The market is dangerously mispricing the duration risk of AI-heavy portfolios by treating interest rate volatility as a transitory energy-supply issue rather than a structural fiscal constraint."
Navellier’s focus on 'transitory' energy-driven rate hikes ignores the structural shift in fiscal dominance. While he correctly identifies the AI hardware build-out as a rational capital expenditure cycle, he underestimates the risk of 'duration risk' repricing. If the 10-year Treasury yield sustains above 4.5%, the P/E expansion narrative for high-growth tech breaks down, regardless of earnings revisions. The lack of widening credit spreads suggests the market is currently ignoring systemic liquidity risks in favor of AI momentum. I view this as a classic 'melt-up' phase where volatility is being suppressed by a false sense of security regarding the Fed’s terminal rate.
If AI-driven productivity gains begin to manifest in broader S&P 500 margins, the current 'crowded trade' could transition into a new secular baseline, rendering historical P/E valuation models obsolete.
"The real risk to the AI-hardware rally is demand normalization and sharper valuation compression in 2027 if earnings don’t materialize as expected."
The article frames the pullback as mostly rate- and energy-driven and labels AI hardware a crowded trade. But the bigger, less-covered risk is demand normalization in data-center compute. If cloud providers curb capex, or pivot to more efficient training/inference, AI hardware revenue and margins in 2027 could disappoint consensus. Even with calm current credit spreads, liquidity and discount-rate dynamics can widen again if macro conditions worsen or policy tightens. The claim that AI is cheap assumes aggressive greenfield growth; a misstep in AI compute demand or pricing power could trigger sharper multiple compression than the piece anticipates. The risk is a sharp unwind in AI-related equities, not just a drift lower on rates.
Counterpoint: if cloud demand remains robust and margins hold, AI hardware could surprise to the upside and defer any meaningful multiple compression; the unwind risk hinges on demand weakness that hasn’t shown through yet.
"Persistent yields above 4.5% will re-rate AI multiples lower even if data center capex remains strong."
ChatGPT flags data-center demand normalization as the overlooked risk, but this connects directly to the rate persistence both Claude and Gemini raised. Even if cloud capex holds steady, sustained 10-year yields above 4.5% from energy inflation would still compress the 35x 2025E multiple on semis, independent of 2027 EPS. The current 7% pullback then looks more like the beginning of re-rating than a pause before reacceleration.
"Rate compression + demand normalization aren't hedges against each other; they're compounding headwinds that could trigger sharper multiple re-rating than either alone."
Grok conflates two independent variables. A 4-5 bps yield rise compresses multiples mechanically—agreed. But ChatGPT's demand normalization risk operates *orthogonally*: even if rates stay flat, AI capex disappointment alone could crater 2027 EPS revisions, making the 35x multiple indefensible regardless of discount-rate math. The panel is treating rate compression and demand risk as sequential; they're concurrent. If both hit simultaneously, the unwind accelerates nonlinearly.
"The current supply-constrained environment for AI hardware provides a revenue buffer that mitigates immediate demand-side risks regardless of interest rate volatility."
Claude is right that these risks are concurrent, but both are missing the supply-side reality: AI hardware is currently supply-constrained, not demand-constrained. Even if cloud providers signal a pivot, the massive backlog for H100s and Blackwell chips provides a revenue floor through 2025. The real risk isn't a sudden demand cliff, but the 'margin squeeze' that occurs if energy costs and interest rates force these firms to pass through higher prices, curbing end-user adoption.
"Backlog isn't a floor; demand risk and margin compression can erode AI hardware profits even with supply tight."
Gemini argues backlog provides a revenue floor for AI hardware through 2025. But backlog is not a guaranteed floor if demand softens or pricing power erodes; margins compress under energy and financing costs, and supply discipline can shift pricing. Even with tight near-term supply, a sustained rate regime above ~4.5% and higher energy adds a second-order drag that can re-rate multiples and trim margin leverage before backlog clears.
The panel consensus is bearish, with key concerns being the compression of multiples due to sustained higher yields and the risk of demand normalization in data-center compute, which could crater EPS revisions and make current multiples indefensible. The risk of a margin squeeze due to higher energy costs and interest rates is also highlighted.
None identified
Demand normalization in data-center compute and the resulting cratering of EPS revisions