July’s correction reveals a two-speed sector — TSMC as the barometer of real demand July 2026 has revealed a reality that Q2’s enthusiasm had obscured : semiconductors are not a monolithic asset class. After the VanEck Semiconductor ETF (SMH) posted an exceptional +71% gain in Q2 2026, driven by explosive demand for AI-dedicated chips, the sector suffered a sharp correction at the start of the summer quarter. Micron fell 13%, Intel 9%, AMD 7% — while the Nasdaq itself rebounded 0.9% on July 14, supported by a softer-than-expected June CPI reading (3.5% vs. 3.8% consensus). This intra-sector divergence is precisely the signal we monitor. It does not reflect irrational panic : it represents a methodical reassessment of the AI value chain’s secondary players against the central question : who actually captures value in the artificial intelligence ecosystem, and at what price? Markets are beginning to distinguish the direct, structural beneficiaries of the AI revolution — foundries, equipment manufacturers, specialist chip designers — from peripheral players whose growth depends on a less predictable second-order demand. IBM’s profit warning, which sent its shares down 25% after the company signalled that enterprise clients are redirecting IT budgets toward AI infrastructure at the expense of legacy software, illustrates this creative destruction perfectly. The aggregate growth of the semiconductor market does not benefit all sector participants equally. Taiwan Semiconductor Manufacturing Company’s quarterly results, due July 16, 2026, represent the most reliable test of genuine advanced chip demand. As the world’s leading foundry — manufacturing chips designed by Nvidia, Apple, AMD and Qualcomm — TSMC has unique visibility across the entire technology value chain. Market consensus stands at $40.04 billion in revenue, up 33% year-on-year (from $30.07B in Q2 2025), and $3.81 in ADR earnings per share, +54% YoY. Gross margins are expected between 65.5% and 67.5%, an exceptional level reflecting TSMC’s pricing power on its 3nm and 2nm processes — the only processes capable of manufacturing the latest-generation AI accelerator chips. Management guidance, particularly on CoWoS advanced packaging capacity (used for Nvidia’s H200 and B100 GPUs), will be scrutinised closely. A signal of capacity strain would confirm the structural solidity of AI demand; an easing would suggest inventory build-up at hyperscalers, which would weigh on 2027 orders. Beyond the headline number, the geographic and customer breakdown will matter. North America’s share of TSMC revenues (primarily Nvidia and Apple) exceeded 70% in Q1 2026, reflecting the concentration of AI demand in the United States. Diversification toward Asia (smartphones) and Europe (automotive, industrial) will be a resilience indicator for subsequent quarters. July’s correction draws a clear dividing line between two categories of sector participants. On one side, players structurally tied to AI demand — TSMC, ASML, Nvidia — show contained declines of 2% to 4%, as their order books are filled 12-18 months out. On the other, players whose AI exposure remains marginal or indirect — Micron (consumer DRAM memory), Intel (PC and legacy servers) — are suffering drops of 10% to 13%, revealing their cyclical vulnerability. Micron is emblematic of this tension. The company benefits from HBM (High Bandwidth Memory) demand, the memory used in AI training GPUs, but its exposure remains limited compared to SK Hynix, which dominates this segment alongside Nvidia. Its conventional memory revenues (DDR5 for PCs) remain subject to classic inventory cycles — precisely what the market is penalising. Intel, for its part, is navigating a perilous industrial transition. Its plan to return to foundry leadership (Intel Foundry Services) faces delays; its datacenter server market share continues to erode against AMD; and its AI chip (Gaudi) is slow to find meaningful adoption. The market anticipates a recovery by 2028, but patience premiums have their limits. The valuation question is central for the wealth management investor. The table below compares the P/E (Price-to-Earnings) multiples of the sector’s main players against the S&P 500 median (21x as of July 14, 2026). The differential is striking, and justifies increased selectivity in sector allocations. Nvidia’s P/E of 40x is often presented as excessive — but it must be viewed in relation to growth. The PEG ratio (P/E divided by earnings growth rate) for Nvidia stands at approximately 0.6, historically low for a company in a phase of exponential growth. TSMC, with a P/E of 28x against 54% growth, presents a similar PEG and superior cyclical resilience, making it a more comfortable entry point in a correction environment. July’s correction should be read as a healthy normalisation following Q2 2026’s excess positioning — not as the beginning of a structural trend reversal. Hyperscalers (Microsoft, Google, Amazon, Meta) have announced cumulative capex plans exceeding $250 billion for 2026, the majority of which will go to GPUs and datacenter infrastructure. This demand is contractualised and unlikely to evaporate over a few quarters. In a diversified wealth portfolio, semiconductor exposure highlights several trade-offs. An index allocation via sector ETFs (SMH, SOXX) implies growing concentration on Nvidia, whose weight exceeds 20% in these indices — introducing significant idiosyncratic risk. July’s correction illustrates how this positioning creates feedback effects : ETFs mechanically selling during redemptions amplify price movements. A more granular approach distinguishes three player profiles to consider within a global equity allocation : structural beneficiaries with near-monopolistic competitive positions (TSMC in advanced foundry, ASML in EUV equipment), dominant beneficiaries subject to growing competition (Nvidia facing AMD and hyperscaler custom chips), and restructuring players whose valuation today reflects an option on recovery (Intel). The macroeconomic context also plays a role. June CPI at 3.5% and the S&P 500 at 7,543 points on July 14 signal that the environment remains constructive for growth equities — but the July 28-29 FOMC meeting, with a residual 25% probability of a 25bp rate hike, could trigger additional volatility on long-duration assets, of which high-growth semiconductors are structurally a part. Gold, at $4,080/oz on July 14 (up 2% on the day alone), continues to play its diversification role within tech-heavy portfolios. The historically low correlation between gold and semiconductors makes it a natural hedging tool during phases of sector profit-taking. This document is provided for informational purposes only and does not constitute investment advice, a personalised recommendation, or an offer to buy or sell financial products. Past performance is not indicative of future results. All investments involve risk, including the risk of capital loss. The information contained in this article reflects Riviera Wealth Management’s analysis at the date of publication and is subject to change. Riviera Wealth Management is a registered investment adviser (CIF), registered with ORIAS under number 11060879 and a member of CNCGP.
Semiconductors : After the AI Euphoria, the Time for Selection
July’s Correction : Technical Pullback or Structural Signal?
“The first law of investing in technology cycles : aggregate demand does not prejudge individual profitability. The gold rush enriched the pick sellers, not every prospector.”
Investment principle — Riviera Wealth Management approach
TSMC : Barometer of Real AI Demand
Anatomy of the Pullback : Two Categories of Players
Valuations : Is the AI Premium Justified?
Company
P/E 2026e
EPS Growth 2026e
AI Positioning
Nvidia (NVDA)
40x
+68% YoY
Dominant — H200/B100 GPU, CUDA
AMD
35x
+42% YoY
Challenger — MI300X, EPYC
TSMC (ADR)
28x
+54% YoY
Structural — 2nm monopoly
ASML
32x
+28% YoY
Structural — EUV monopoly
Micron
22x
+38% YoY
Cyclical — HBM + consumer DRAM
Intel
18x
–4% YoY
Restructuring — IFS, Gaudi
S&P 500 (median)
21x
+9% YoY
—
Implications for Wealth Portfolio Allocation
“Never confuse a sector’s growth with the profitability of every investment within it. Distinguishing between aggregate demand and individual value capture is the investor’s primary competence.”
Investment Committee — Riviera Wealth Management, July 2026
Key Takeaways
01
02
July 2026 Correction — Semiconductors
Change in % from end-June 2026 peak (data as of July 14, 2026)
Memory / legacy architecture players
Equipment / AI foundry leaders
03
04
Valuation Multiples — Semiconductors vs. S&P 500
2026e Price-to-Earnings as of July 14, 2026 (FactSet consensus)
Reasonable valuation (≤21x)
Moderate premium (28x)
Elevated premium (≥35x)
05
What to Remember
