Why is Wall Street still bullish on AI chips? Analysis of investment opportunities behind the semiconductor pullback.

In July 2026, the semiconductor sector experienced a sharp correction. In the early hours of July 8 Beijing time, all three major U.S. stock indexes closed lower, with the Nasdaq falling 1.16% to 25,818.69 points, the Dow Jones dropping 0.25% to 52,925.15 points, and the S&P 500 declining 0.45% to 7,503.85 points. The Philadelphia Semiconductor Index, which tracks the overall performance of chip stocks, plunged 4.65% to 12,300.52 points, falling below the 50-day moving average and hitting a new closing low since June 10. At the individual stock level, Intel fell over 9%, AMD dropped more than 6%, Micron Technology declined over 4%, while Nvidia bucked the trend, rising 0.71% to $196.93.

This sell-off was not an isolated event. On July 1, the Philadelphia Semiconductor Index tumbled 6.27% in a single day; on July 2, it continued to fall by 5.44%, with a cumulative decline of over 11% in two trading days. The Korea Composite Stock Price Index (KOSPI) plunged about 7.9% on July 2, triggering a sell-side circuit breaker during trading hours. Although Samsung Electronics reported on July 7 that its second-quarter operating profit surged more than 18 times year-over-year to 89.4 trillion won, it only slightly exceeded market expectations of 87.3 trillion won, failing to meet investors' extremely high expectations.

However, just as panic gripped the market, multiple Wall Street institutions—including Goldman Sachs, JPMorgan, Bank of America, UBS, and Morgan Stanley—spoke almost simultaneously, collectively sending the same signal: the semiconductor correction is not the end of the AI rally but provides a new window for positioning.

Nature of the correction: Profit-taking and valuation adjustment, not demand collapse

Understanding the nature of this correction is a prerequisite for judging future trends.

From a fundamental data perspective, the AI demand cycle is far from over. In an industry research report released on July 7, JPMorgan disclosed that global semiconductor sales in May 2026 reached $131.9 billion, up 16.1% month-over-month, significantly above the historical average seasonal growth of 4.5%. Year-over-year, industry sales grew by 118.8%. If the second half of the year only follows historical seasonal patterns, global semiconductor revenue in 2026 could still grow over 90% year-over-year, reaching $1.5 trillion to $1.6 trillion. The latest forecast from the World Semiconductor Trade Statistics (WSTS) suggests that the global semiconductor market size could reach $1.51 trillion in 2026.

Demand is equally strong. JPMorgan strategist Mislav Matejka explicitly stated in a client note on July 6 that the semiconductor upcycle is far from over, noting that "meaningful new supply is unlikely to arrive before 2028." Memory chip makers, including Micron, SK Hynix, and Samsung, have sold out their high-bandwidth memory (HBM) supply through 2026, and new wafer capacity is not expected to materialize substantially until after 2028. AI data centers are expected to consume about 70% of global memory chip production this year.

In a semiconductor industry report released on July 8, Bank of America pointed out that the recent pullback in semiconductor stocks is a normal market correction, not a signal of weakening AI demand. Historical experience shows that semiconductor stocks often consolidate in the summer, and after the market completes profit-taking and valuation adjustments, a new wave of rebounds typically arrives in the fall. The bank maintains its optimistic view of the long-term AI semiconductor cycle, believing the industry is still in the middle of an 8-to-10-year growth cycle.

On July 6, UBS Asset Management stated that despite ongoing market volatility, semiconductor-related stocks are not in a bubble. UBS cited strong demand signals—weekly AI token consumption has grown 8-fold year-to-date—as a key basis for its positive outlook.

Overall, the core drivers of the current adjustment come from three levels: excessive gains earlier (the Philadelphia Semiconductor Index rose over 100% in the first half of the year), profit-taking (crowded trade unwinding and deleveraging), and market revaluation of valuations (the PHLX Semiconductor Index is up over 80% year-to-date, raising earnings expectations). These factors are at play, not a structural deterioration in AI demand itself.

From "Buying the sector" to "Stock picking": A qualitative shift in AI investment logic

This is the most central change in current institutional views.

Over the past two years, market funds have generally adopted a "basket-buying" strategy for the semiconductor sector—buying GPU companies, chip makers, and semiconductor equipment firms, almost all of which generated significant excess returns. But Goldman Sachs explicitly stated in a July 7 report that AI chip trading has entered a more selective phase, and investors should no longer simply buy the entire sector.

The logic behind this shift: The PHLX Semiconductor Index has surged over 80% year-to-date, significantly outperforming the S&P 500 and Nasdaq. This strong performance has raised the bar for future earnings, making the risk-reward profile ahead of the second-quarter earnings season more divergent. In other words, the market has moved from a "concept trading" phase to an "earnings trading" phase.

Goldman Sachs put it most directly: A blanket "buy-the-dip" strategy carries risks. The firm favors CPU, ASIC, memory, and equipment stocks aligned with AI growth, naming AMD and Applied Materials as top picks. However, it is cautious on semiconductor companies tied to the mobile phone supply chain, those with high valuations, or weaker demand.

JPMorgan's perspective is slightly different. The bank maintains an "overweight" view on the semiconductor sector, believing that AI accelerated computing, memory, and networking equipment supply chains will continue to be the most direct beneficiaries of the cycle. But JPMorgan also warns that the valuation gap between AI semiconductor makers and major cloud service providers has reached unsustainable levels. Semiconductor stocks are up 87% this year, while the "Magnificent Seven" have fallen 7% from their year-to-date highs—whether this divergence can persist is a question the market must answer.

Bank of America focuses more on valuation opportunities. Analyst Vivek Arya noted that memory chips trade at a forward P/E of only 10 times, which is severely undervalued given their growing share of AI infrastructure spending. The bank recommends focusing on industry leaders such as Nvidia, Broadcom, Lam Research, and KLA.

UBS offers a warning from another dimension: Increased capital expenditures could pressure the cash flow of hyperscale cloud providers in the second half of 2026. As investors demand greater capital discipline, this could put downward pressure on semiconductor and AI hardware valuations.

Snapshot of views from four Wall Street institutions

| Institution | Core View | Preferred Directions | | --- | --- | --- | | Goldman Sachs | AI chips enter a selective phase, not recommending "basket buying" | CPU, ASIC, memory, equipment; names AMD, Applied Materials | | JPMorgan | Correction offers positioning opportunity; upcycle lasts at least until 2028 | Maintain "overweight" on semiconductors; AI accelerated computing, memory, networking equipment | | Bank of America | Industry still in middle of 8-10 year growth cycle; correction is healthy | Nvidia, Broadcom, Lam Research, KLA; reiterates "buy" on Micron | | UBS | Short-term volatility offers long-term entry; sector not in bubble | Positive on semiconductors; suggests selective AI investment | | Morgan Stanley | AI long-term trend unchanged, but funds may rotate from chips to cloud | Hyperscale cloud providers (Microsoft, Amazon, Meta) |

Source: Compiled from various institutional research reports from July 2026

Where are the next-stage AI investment opportunities?

If the market shifts from "sector-wide gains" to "selective leadership," what are the screening criteria? Synthesizing views from various institutions, we can break it down from three levels of the industry chain.

Chip side: Computing power core remains the most certain direction

Goldman Sachs' clear preference for CPU, ASIC, and memory essentially targets the deepest layer of AI computing infrastructure. GPUs generate intelligence, HBM and DRAM handle high-speed data transfer, and enterprise NAND and SSDs manage hot data and caching. Institutions like Goldman Sachs believe that the AI computing arms race led by cloud giants is pushing memory chips from cyclical commodities to scarce strategic assets. The 2026 DRAM and NAND price increases are not the end but possibly the early stage of a super cycle.

TrendForce predicts that in 2026, Nvidia's global AI chip market share will be about 64%, and AMD's about 8.6%. On July 5, Goldman Sachs sharply raised its 12-month target price for AMD from $450 to $640, maintaining a "buy" rating—a move that itself endorses the long-term value of the AI chip track.

Infrastructure side: Value chain extension from chips to cloud

Morgan Stanley provides another important perspective. Chief equity strategist Michael Wilson noted in a July 6 report that momentum in semiconductor stocks is fading, and investors are starting to turn to this year's laggards—AI hyperscale cloud providers, including Microsoft, Amazon, and Meta. Wilson's team believes that semiconductor companies' growth ultimately depends on the capital expenditures of cloud giants, and the current divergence between chip stocks and cloud giants may be unsustainable.

JPMorgan predicts that the global semiconductor equipment market will be about $159 billion in 2026, up 28% year-over-year, growing further to $205 billion in 2027, and reaching $237 billion in 2028. Long-term equipment procurement plans from global leading companies lock in long-term demand, making the two-year upcycle for the semiconductor equipment sector from 2026 to 2027 highly certain.

AI application side: Commercialization closed loop is forming

Morgan Stanley Fund's equity investment department pointed out that the industry has formed a positive cycle of "continuous iteration of large models → steady growth in corporate revenue → increased AI capital expenditure," and the business model has gradually proven viable. With explosive growth in token demand driven by coding and AI agents, revenue from large models and cloud providers is entering the next accelerated growth phase.

Bank of America expects global cloud and AI infrastructure capital expenditure to reach $1.5 trillion by 2027. This level of spending means that future AI winners may not be limited to chip companies but the entire infrastructure ecosystem—from computing chips to storage devices, from network facilities to data center operations, from underlying hardware to upper-layer software applications—all benefiting from this long-term trend.

Conclusion

In summary, Wall Street institutions' collective bullishness on AI semiconductors is built on a clear logical chain: the AI demand cycle is far from over (the global semiconductor market is expected to exceed $1.5 trillion); supply constraints persist (new capacity won't materialize substantially until 2028); and capital expenditure continues to expand (cloud and AI infrastructure spending is expected to reach $1.5 trillion by 2027).

But this does not mean the semiconductor sector will replicate the broad-based gains of the past two years. The consensus among Goldman Sachs, JPMorgan, Bank of America, UBS, and Morgan Stanley is that the AI theme still holds, but the market has moved from concept trading to earnings trading. Future divergence will depend on companies' ability to deliver earnings, the proportion of AI revenue, and their strategic position in the broader infrastructure ecosystem.

For investors, the current correction may offer an opportunity to reassess portfolio structure—moving from "buying the sector" to "stock picking," from chasing beta to uncovering alpha. This is both a challenge and a starting point for a new round of positioning.

FAQ

Q1: What are the main reasons for this semiconductor correction?

This correction is mainly driven by three factors: profit-taking from excessive earlier gains, unwinding of crowded trades and deleveraging, and market reassessment of high valuations. JPMorgan and Bank of America both note that this is a normal market correction, not a structural change in AI demand. New capacity is not expected to come online substantially until 2028, and the supply-demand dynamics remain healthy.

Q2: How do Wall Street institutions view the long-term outlook for AI chips?

Institutions such as Goldman Sachs, JPMorgan, Bank of America, and UBS generally believe the long-term trend for AI chips remains unchanged. JPMorgan expects the semiconductor upcycle to last at least until 2028; Bank of America believes the industry is still in the middle of an 8-to-10-year growth cycle; UBS states the sector is not in a bubble; Morgan Stanley sees the long-term AI trend unchanged but notes funds may rotate from chip stocks to areas like cloud computing.

Q3: What does the shift from "buying the sector" to "stock picking" in AI investment logic mean?

This means the market is no longer simply bullish on the entire semiconductor sector but is beginning to distinguish true AI beneficiaries from followers. Goldman Sachs explicitly advises against continuing "basket-buying" of the semiconductor sector. Future investment opportunities will concentrate on companies with solid profitability, a high proportion of AI revenue, and the ability to consistently benefit from AI infrastructure expansion.

Q4: What are the key directions for next-stage AI semiconductor investment?

Institutions generally favor directions including: CPU, ASIC, GPU, and HBM memory on the chip side; semiconductor equipment, data centers, and cloud computing on the infrastructure side; and enterprise AI software and AI agents on the application side. Bank of America expects global cloud and AI infrastructure capital expenditure to reach $1.5 trillion by 2027.