Semiconductor earnings strengthen as valuations contract: Evercore’s view of the cycle

Why was the share-price response weak despite earnings beats?

Texas Instruments reported earnings per share of 2.62 dollars for the third quarter of 2026, 7% above market consensus. Fourth-quarter guidance also exceeded consensus, but the share-price reaction after the announcement was limited. Following the results, its earnings-per-share estimate for the next 12 months rose 11%, while its price-to-earnings ratio fell 8%.
Lam Research’s earnings per share for the quarter ending in September beat expectations by about 1%, while its outlook for the quarter ending in December met expectations. After the results, its earnings-per-share estimate for the next 12 months rose 2%, while its price-to-earnings ratio fell 14%.
Evercore sees this as a late-cycle pattern: when earnings forecasts exceed the previous cycle’s peak, investors may be willing to pay a lower price-to-earnings ratio. The two companies’ guidance also suggests that year-on-year revenue growth will peak in the fourth quarter of 2026; a peak in the growth rate does not mean revenue will immediately decline.

Evercore distinguishes between two time horizons: the semiconductor cycle is primarily supply-driven and typically lasts three to four years, while the evolution of demand for computing platforms can last longer. Long-term AI expansion and a semiconductor cycle entering its later stages can therefore occur at the same time.
Demand remains supported. The firm expects cloud infrastructure capital spending to exceed 1.4 trillion dollars in 2027, up 43% from 2026. Faster growth in cloud providers’ new annualized recurring revenue and higher incremental operating margins also support its expectation of continued investment.
But funding and inventories are sending opposing signals: Evercore expects large cloud providers’ free cash flow to turn negative in 2027. As of the second quarter of 2026, electronics manufacturing services providers’ inventories, adjusted for prices, were 25% above their long-term trend, and inventory days had also begun to rise. These changes will put pressure on industry valuations.
The firm concludes that further earnings upgrades can partly offset price-to-earnings compression, but differences in companies’ growth and starting valuations will have a greater influence on outcomes.

After speaking with more than 20 AI engineers, Evercore concluded that the demand driver has shifted from post-training reinforcement learning to agents: one task can spawn hundreds to thousands of sub-agents, with server CPUs handling coordination, scheduling and tool calls.
The firm’s latest research shows that CPU-to-GPU configuration ratios are moving from the previous range of 1∶2 to 1∶4 toward 1∶1, while some agent workloads are also moving to standalone CPU racks. Server CPU lead times have lengthened from 8 to 12 weeks in the first-quarter survey to 12 to 16 weeks; experts expect prices to rise 15% to 20% in the third and fourth quarters of 2026.
On that basis, Evercore expects CPU growth to outpace GPU growth for the foreseeable future and is positive on Arm, AMD, Intel and Rambus, naming Arm a principal long-term beneficiary.

Evercore raised its baseline forecast for wafer fabrication equipment spending in 2027 from 215 billion dollars to 235 billion dollars, and its 2026 forecast from 155 billion dollars to 159 billion dollars. The upgrades mainly reflect investment in AI memory and advanced logic chips, while mature-node spending is expected to fall as Chinese and analog chipmakers cut spending.
Lam Research management expects industry spending on wafer fabrication equipment to reach more than 1500 hundred million dollars in 2026 and exceed 200 billion dollars in 2027; cleanroom space limits capacity expansion. The company’s revenue outlook for the quarter ending in December is 8.6 billion dollars, up 5% sequentially, but component supply constraints limited revenue.
Growth in equipment spending does not necessarily bring valuation expansion: while equipment spending doubled from 2013 to 2018, equipment stocks’ price-to-earnings ratio relative to the S&P 500 fell from 1.2 times to 0.7 times. Evercore believes investment in AI memory and advanced processes can drive earnings upgrades, while it still expects equipment stocks’ price-to-earnings ratios to compress.