Semiconductor Struggles: Why is the Market Turning to Apple?

07/28 2026 464

Recently, Apple's market value briefly surpassed NVIDIA's during trading, reclaiming its position as the world's most valuable company for the first time in over a year. Since mid-2025, when NVIDIA held the top spot for more than 260 consecutive trading days, this marks the first time it has truly relinquished its crown. This is not a simple ranking change: On the same day, the Philadelphia Semiconductor Index fell more than 20% from its late-June peak, officially entering a technical bear market. The inverse relationship between the chip sector's collapse and Apple's ascent holds the key to understanding this shift. Behind the market cap reshuffling lies a systemic revaluation of the entire AI semiconductor industry chain by capital markets, with Apple positioned to benefit.

To understand Apple's ascent, one must first grasp why NVIDIA is declining. In mid-July, Chinese AI company Moonshot AI released its next-generation large model, Kimi K3, claiming to achieve near-global-first-tier performance at a training cost significantly lower than its U.S. counterparts. Multiple institutions labeled this a "DeepSeek moment for domestic models." Markets immediately began re-examining a core question: If high-performance models can be trained with less computing power, how will Silicon Valley's nearly $1 trillion in AI infrastructure investments deliver returns?

This doubt quickly translated into sell-offs. AI-related core beneficiaries like NVIDIA, AMD, Broadcom, and Micron saw consecutive declines, with the memory sector hit particularly hard: SK Hynix plummeted 15.37% in a single day, triggering a circuit breaker on South Korea's benchmark index; Japan's Kioxia saw its market value nearly halve in less than a month. Public statistics show that global AI-related tech stocks have lost market value on the scale of $3 trillion in this round.

More notably, market behavior patterns have shifted. On July 16, TSMC delivered better-than-expected earnings and raised its capital expenditure guidance—a scenario that would have typically boosted chip stocks over the past two years. Instead, global tech stocks continued to fall that day. When positive catalysts fail to work, it signals a changing trading logic. Bank of America's July fund manager survey revealed that while hyperscale cloud providers' capital expenditures are expected to grow 76% this year to $673 billion, growth will sharply decelerate to 25% next year. In other words, investors are not denying AI's long-term value but are beginning to price in "peak capital expenditure growth."

Amid this sell-off, capital needed a destination that preserved AI exposure while avoiding capital expenditure risks—and Apple emerged as nearly the only option. HSBC estimates that in Apple's projected 2026 sales, only about 2.5% will be allocated to capital expenditures like AI data centers, compared to 39% for Meta, Google, Amazon, and Microsoft. This divergence directly impacts cash flows: Apple's 2026 free cash flow is expected to hit a record $140 billion, up over 40% from 2025. In contrast, Google's parent company is projected to see a significant decline in free cash flow over the same period. This forms the core of HSBC's rationale for upgrading Apple: The company avoids getting entangled in the industry's capital expenditure arms race while sitting on a global installed base of approximately 2.5 billion devices, "well-positioned to leverage the upcoming upgraded Apple Intelligence to tap this massive installed base." Wall Street consensus can be summed up as: If you believe AI capital spending will keep expanding, choose NVIDIA; if you think it will slow, Apple is the better bet.

While capital rotation serves as an external factor, Apple's intensive semiconductor layout (strategic deployments) form the industrial fundamentals driving this shift. Over the past two months, Apple has advanced on four parallel fronts.

The first front is process technology positioning. Supply chain information indicates that the iPhone 18 Pro series, launching in September, will debut the A20 Pro chip using TSMC's 2-nanometer (N2) process with WMCM packaging—the world's first mass-produced 2nm mobile SoC. The cost is staggering: 2nm wafer pricing approaches $30,000 per piece, and with initial yield challenges during mass production, the amortized cost per A20 Pro chip is estimated at around $280. Apple has historically secured generational advantages by securing TSMC's most advanced process capacities, a strategy unchanged from N3 to N2. For TSMC, Apple's 2nm orders provide crucial cash flow during its advanced process ramp-up.

The second front is finalizing baseband self-development. Also on the iPhone 18 Pro series, Apple's second-generation self-developed 5G baseband, C2, will fully replace Qualcomm's solution, marking Qualcomm's exit from iPhone's premium lineup. While the C-series basebands still lack 5G millimeter-wave support (meaning U.S. models may retain Qualcomm solutions), the certainty of Qualcomm losing iPhone's premium segment has solidified—an underestimated change in the 2026 semiconductor landscape.

The third front is restructuring the storage supply chain, currently the most geopolitically sensitive move. AI data center expansion has triggered a global storage chip shortage, with JPMorgan projecting DRAM and NAND supply-demand tightness to persist until 2028. Cost pressures have materially impacted Apple: On June 25, Apple announced price hikes of 15-25% across Mac, iPad, and home devices. Apple's response is to seek new upstream suppliers: Public reports indicate Apple is lobbying the U.S. government for permission to procure DRAM from ChangXin Memory Technologies and is simultaneously engaging with Yangtze Memory Technologies for Chinese market devices. This move carries dual significance: short-term cost relief and long-term incorporation of Chinese storage firms into its negotiation pool as leverage against Samsung, SK Hynix, and Micron.

The fourth front—and Apple's most vulnerable—is AI server chips. Reports suggest Apple's self-developed M2 Ultra server chips cannot handle ultrascale models like Google's Gemini, forcing some cloud-based Siri workloads to run on Google Cloud via NVIDIA GPUs—a situation deemed "difficult to collaborate" by some Apple executives. More troublingly, the next-gen AI server chip codenamed Baltra has been delayed, while a truly NVIDIA Blackwell-competitive M7 Ultra server version won't arrive until 2029. This leaves Apple facing a roughly three-year technology gap in AI computing autonomy.

Apple's remedial actions are already underway: In recent months, it has proactively approached multiple semiconductor startups for acquisition talks; extended its AI server chip collaboration agreement with Broadcom to 2031; and its CFO announced abandoning the long-held "net cash neutral" financial policy, widely interpreted as stockpiling ammunition for major acquisitions. Interestingly, Apple's self-developed chip empire originated from its $278 million acquisition of PA Semi in 2008. With hardware chief John Ternus assuming the CEO role in September, a "chip-focused" Apple may pursue mergers and acquisitions more aggressively than under the Cook era.

Apple's AI and semiconductor strategies represent two sides of the same coin: For non-differentiating components, it procures externally at the lowest cost; for experience-critical elements, it invests heavily in self-development.

Large models fall into the former category. Public reports indicate Apple pays roughly $1 billion annually for a customized 1.2 trillion-parameter Gemini model from Google—about eight times the scale of Apple's previous cloud-based self-developed models. However, Apple has not simply "rebranded" it: The foundational model for the new Siri is a distilled, efficient small model based on Apple's proprietary data and optimized for Apple Silicon, with most inference completed on-device and through Apple's private cloud nodes. On July 14, iOS 27 beta opened to general users, marking Apple's largest-ever AI public beta, with the official version launching in September. The $1 billion annual licensing fee, compared to the thousands of billions in annual capital expenditures by the four major cloud providers, offers clear cost-effectiveness.

Another key catalyst comes from China. On July 16, Chinese regulators approved Apple's partnerships with Alibaba and Baidu to launch Apple Intelligence in China. As China represents Apple's second-largest market, the long absence of AI features has been a core factor suppressing sales expectations there, making this approval highly significant.

The hardware refresh narrative is carried by foldable iPhones. Supply chain sources indicate Apple has raised its 2026 foldable iPhone production target to around 10 million units, up from previous estimates of 7-8 million. However, Counterpoint analyst Lin Keyu believes the "10 million shipment target is a bit high," and Ming-Chi Kuo cautions that manufacturing yields may pose constraints. Contrary evidence exists: UBS's early-July survey showed Apple's existing AI features have yet to ignite a refresh boom, with user upgrade intentions continuing to decline. Whether AI can truly drive refresh cycles awaits sales data after September. Nevertheless, fundamentals remain solid: Apple's revenue grew over 15% year-on-year in its last two fiscal quarters.

Apple's ascent fundamentally signals a shift in valuation anchors for the semiconductor industry: Over the past two years, the sector priced stocks based on "who sells more GPUs," but this market cap reshuffling indicates a weighting shift toward "who can first turn AI into stable profits." This does not mean the end of the AI narrative—most institutions agree this represents a diffusion of AI trading from infrastructure to applications and endpoints, rather than a retreat. In this process, terminal manufacturers' bargaining power rises, pure computing power suppliers' valuation elasticity converges, and Apple's moves across process, baseband, storage, and AI chips will continue reshaping the positions of TSMC, Qualcomm, Samsung, and Broadcom.

Of course, this throne contest is far from over. Storage price hikes eroding profit margins, the three-year AI server chip gap, and September's CEO transition remain variables for Apple. But regardless of the final outcome, this July 2026 reshuffling already declares: The "faith-driven" phase of AI semiconductor sentiment has ended, and the "value validation" phase has begun. Apple has provided another possible answer for the second half in its most "Apple-like" way—without race-to-market tactics, without burning cash, by transforming 2.5 billion devices into AI distribution channels.

Solemnly declare: the copyright of this article belongs to the original author. The reprinted article is only for the purpose of spreading more information. If the author's information is marked incorrectly, please contact us immediately to modify or delete it. Thank you.