Year-to-date performance of the chips sector and AI infrastructure

Backdrop

This is a narrow market. The S&P 500 is +13.1% and the Nasdaq 100 (QQQ) +21.2% this year, but small caps (IWM) fell -5.7% over the last month, high-yield credit (HYG) sits below its 200-day average, and the 10-year Treasury yield is 5.18% (+0.54 pt in a month, FRED print of Sep 24). Chips have carried the index anyway: SOXX is +90.2% and SMH +68.4% year to date. The volatility regime reads Risk-On (VIX 14.87), and on the latest 1-day read the semis lane (+1.27%) led both the Mag 7 (+0.22%) and the safe havens (+0.30%).

The money has gone where the earnings are growing: memory makers raising prices, CPU makers running out of server supply, and the equipment makers that build both. The buyers of that hardware (the AI hosts) have lagged it by a wide margin. The rest of this report quantifies that gap.

Chips — year to date

32 of 32 chip names are up this year, the median is +85.8%, and 14 have more than doubled (SNDK, AEHR, MU, INTC, STX, MRVL, SIMO, AMD, ARM, WDC, SMTC, FORM, ALAB, KLIC). The leader is SNDK (+648.9%) and the laggard AVGO (+1.9%). By cohort, memory & storage (+215.4%) and compute & logic (+116.1%) lead, while analog & embedded (+45.1%) and packaging (AMKR, +35.8%) lag. 23 of 32 names set their year high in June or July, the median name then fell -39.5% peak-to-trough by mid-August, and 28 of 32 are up over the past month. The median chip name is still 24.5% below its 12-month peak close.

SKHY (SK hynix ADR) listed on Jul 10, so it has no YTD base; its 1M is shown. vs peak = last close against the highest close of the trailing 12 months.

Chips — valuation & earnings

Forward P/E is price ÷ next-fiscal-year consensus EPS. The cheapest chips on it are memory: SKHY 5.7, SNDK 6.8, MU 6.8, NXPI 13.2. Those multiples look low because the market prices peak-cycle earnings, not because the stocks are ignored. The richest are ARM 102, AEHR 75, INTC 60. 13 chip names report by the end of October: MU, ASML, TSM, LRCX, INTC, NXPI, AMKR, RMBS, TXN, SKHY, STX, KLAC, FORM. Of those with an options-implied event move and a 5-year profile, 5 (MU, LRCX, INTC, TXN, SKHY) are priced below their own average reaction.

Chips — signals

26 of 32 chip names trade above their 200-day average. Only AMKR, AVGO, CAMT, NVMI, NXPI, RMBS are below it. In the last five sessions 22 names fired at least one upward cross and 3 at least one downward one. 9 reclaimed their 50-day average (ALAB, AMKR, ASML, KLIC, LRCX, MTSI, NXPI, RMBS, TXN), a broad September repair after July's damage. Overbought on RSI (≥ 70): AMD.

Breakout screen: 2,029 US listings ≥ $2B swept; candle patterns read for 1,240 of 1,272 candidates. RSI is shown from the breakout row where a name fired a signal, otherwise from RSI Watch, which only lists tails (≤ 35 / ≥ 65); a name in neither reads '35–65'. 50/150-day distances only exist on breakout rows.

Deep dives

Lam Research (LRCX) — the etch & deposition toolmaker

Background. Lam builds the machines that etch patterns into wafers and lay down atom-thin films. It is strongest in memory, where 3D NAND needs deep, narrow holes etched through hundreds of layers, and it earns a steadier second stream from spares and service on its installed base. It is a capital-spending cyclical, so it moves with memory makers' budgets.

This year. +84.1% YTD, peak close $433.33 on Jun 30, now 27.3% below it. Revenue over the last four quarters: $5.32B → $5.34B → $5.84B → $6.72B, with EPS beating estimates each time. Forward P/E 26.9, PEG 1.14. Trend: above the 50/150/200-day averages, with a 9/20 EMA bull cross on the latest close. Price is testing the $314–323 zone (a possible May double top). Support is $266–274, which held Sep 14. It reports Oct 21 after the close: options imply ±5.3% / ±13.8% against a 5-year average of ±5.7% 65% up · n=20.

Intel & AMD — the CPU shortage trade

Why they rose. Agentic AI (software that plans, calls tools and runs many tasks at once) leans on ordinary server CPUs, not just GPUs. Server CPU supply tightened through 2026, and the Q1 reports in late April turned that into numbers. The latest leg (Sep 21) came on the strong launch of Meta's Muse AI agent and a supply-chain report of an AMD Q4 price increase (CNBC, Bloomberg, Fusion Worldwide).

Background. Neither is CPU-only. Intel also runs a foundry business that makes chips for outside customers (the long-term bet), owns most of Mobileye and part of Altera, and has the US government, Nvidia and SoftBank as shareholders since 2025. AMD sells Instinct AI GPUs (its challenge to Nvidia, including a large OpenAI deployment deal), Ryzen PC chips, console chips for PlayStation and Xbox, and Xilinx FPGAs.

This year. INTC +233.3% and AMD +194.5% YTD. INTC is still 12.7% below its June peak after September's recovery; AMD is at a record. Revenue: INTC $13.65B → $13.67B → $13.58B → $16.13B; AMD $9.25B → $10.27B → $10.25B → $11.54B. Both are stretched: INTC's RSI turned down from above 70 on the latest close and AMD's RSI is 73. INTC's forward P/E is 59.6 on negative trailing GAAP EPS; AMD's is 40.5 (PEG 0.38, on +106% expected EPS growth). INTC reports Oct 22 (implied ±6.6% / ±16.0%; 5-year ±9.7% 35% up · n=20); AMD reports Nov 3.

Memory — the biggest winners

SNDK, MU, STX, SIMO, WDC have all more than doubled, led by SNDK (+648.9%). The driver is a memory shortage: AI servers need far more DRAM, HBM and flash, contract prices rose sharply through the year, and suppliers signed multiyear deals (CNBC, IG). MacroView's own numbers show it: MU revenue went $11.31B → $13.64B → $23.86B → $41.46B over four quarters. MU's forward P/E is 6.8, the market pricing peak-cycle earnings. MU reports Sep 30 after the close (implied ±6.6% / ±8.6%; 5-year ±7.0% 40% up · n=20), the first read for the whole cohort and for LRCX.

Nvidia — fundamentals up, stock lagging

NVDA is only +20.7% YTD against SOXX +90.2%, even though its revenue went $57.01B → $68.13B → $81.61B → $96.22B over its last four quarters. At a forward P/E of 14.3 it is among the cheapest compute names. Coverage attributes the gap to positioning (it was priced for the best case first) and to worries about how much higher AI capex can go (CNBC). It reports Nov 17.

AI infrastructure — year to date

The AI hosts (who run the compute) have lagged the chipmakers (who build it). The median AI-infrastructure name is +10.5% YTD against +85.8% for chips; 19 of 28 are up. The data-center ETF DTCR is +31.9% and the cloud ETF SKYY +26.6%. Neoclouds lead (+183.5% median; DOCN +190.4%), and the US hyperscalers are a +7.4% median, with Oracle at the bottom (-29.7%, 56.2% below its peak). Asia-Pacific platforms are the weakest cohort (-14.8%).

Returns are in each listing's own currency. Dropped as stale (last close not Sep 25): 018260.KS (2026-09-23), 300442.SZ (2026-09-24). Converted miners are ex-bitcoin miners rebuilding their sites for AI/HPC; several still earn most of their revenue from mining.

AI infrastructure — valuation, earnings & signals

11 of 28 AI-infrastructure names trade above their 200-day average, against 26 of 32 chips. In the last five sessions 8 fired an upward cross and 13 a downward one. Hyperscaler forward P/Es: MSFT 21.8, GOOGL 23.1, AMZN 23.8, META 21.5, ORCL 12.5. The four US platforms all report Oct 28–29, and their capex guidance is what the chip names upstream are waiting for.

Many hosts are loss-making or carry no consensus (miners, neoclouds), so forward P/E reads n/a there. That describes the earnings, not a data gap. Non-US lines have no RSI or crosses (the screens cover US listings only).

Earnings ahead

20 of this report's names report before the end of October. The sequence matters more than any single date: MU (Sep 30) sets the memory read, then TSM and ASML (Oct 14–15) on foundry and lithography demand, the equipment makers and Intel (Oct 21–28), and finally the hyperscalers' capex guidance (Oct 28–29). Seasonally, October has been a good month for the S&P 500: median +2.2%, up in 64% of 33 years.

Scenarios