To The Moon

On October 1, US chip stocks rallied collectively in the overnight session. The SOXL triple-leveraged semiconductor ETF jumped more than 9%, ARM rose 4%, AMD and Intel gained over 3%, while Micron, SK Hynix, Broadcom, and Nvidia all closed higher. In an instant, the entire computing power-to-storage chain moved upward in sync, and discussions about a new upcycle for AI semiconductors heated up in the market once again.

Last night's US August PCE inflation data cooled more than expected, temporarily easing the Federal Reserve's rate hike pressure and opening a short-term valuation repair window for high-valuation growth stocks. Meanwhile, Micron delivered an earnings report after the bell that beat expectations, with hundred-billion-level long-term agreements locking in orders and further repairing the prosperity expectations for the storage sector. Together, these two factors became the core drivers behind the rally in US chip stocks.

Putting Micron's earnings, the CPU narrative driven by Meta's Muse, and Bank of America's upward revision of the AI data center TAM together, what this round of gains is really trading is a more fundamental shift: the structure of AI computing power itself is changing gears.

Micron's Blockbuster Quarter

Micron reported revenue of $54.23 billion for the fourth quarter of fiscal 2026, a year-over-year surge of 379%, with Non-GAAP earnings per share of $33.42, up more than 11 times year-over-year, comprehensively crushing market expectations. Its data center business unit generated $18 billion in revenue, a year-over-year increase of 1,042%, with a gross margin of 90%.

Objectively speaking, however, these impressive figures are not incremental information, as the market had already fully anticipated strong AI storage demand. The truly noteworthy incremental signal is that customer purchase commitments through long-term supply agreements have surged from $22 billion in June to $32 billion, with most of that prepaid in cash. Correspondingly, the company's remaining performance obligations (RPO) directly reached $150 billion.

The significance of these numbers needs to be understood in the historical context of memory chips. DRAM and NAND have traditionally been the most cyclical categories in the semiconductor sector, with prices fluctuating sharply with supply and demand, and manufacturers often expanding capacity after demand inflection points, ultimately leading to the repeated cycle of "price hikes—capacity expansion—oversupply—price collapse." The market has long assigned a lower valuation center to memory stocks than to logic chips, precisely pricing in this cyclicality.

But now, the $32 billion in customer cash prepayments, combined with the $150 billion RPO scale, is quietly rewriting this inherent valuation premise. As a large number of medium- and long-term procurement contracts lock in downstream demand, the revenue visibility of the memory business has effectively risen from quarterly to annual levels. Micron CEO Sanjay Mehrotra further stated explicitly that supply and demand in fiscal 2027 and 2028 will be even tighter than in 2026. This is not just a judgment supported by contracts, but also a signal of confidence to the market.

The CPU Narrative Behind ARM and Intel

ARM rose as much as 4% in the overnight session, and Intel gained over 3%, outperforming Nvidia's 1.5% gain. If we look only at traditional logic, it is hard to explain why ARM's licensing model has not significantly benefited in the AI training era, and why Intel's process technology and market share remain under pressure—so why are they running ahead?

This may be because the market is still increasing its bets on the imaginative space brought by Meta's Muse. Muse is an AI agent application launched by Meta for consumer users, reaching 642,000 daily active users on US mobile devices on its 12th day, compared with about 231,000 for ChatGPT during the same period. Its key design is to configure an independent cloud-based virtual machine, MuseSecureVM, for each user, equipped with browser, computing, memory, and storage resources.

This design transforms AI's computing power consumption model from "one-time Q&A" to "round-the-clock online operation." After users exit the application, the Agent continues to execute tasks in the background. GPUs handle generation and inference, but completing web browsing, cross-application calls, and task scheduling—these "execution-side" tasks—relies on CPUs. According to estimates by Sinolink Securities, Agent tasks are mainly serial logic workflows, and the CPU-to-GPU ratio is shifting from the previous 1:4 to 1:8 toward 1:1 to 1:2. Data from TrendForce further confirms that CPU task orchestration in Agentic AI can account for 50% to 90% of total latency.

This means that once AI inference services penetrate from tens of millions of developers to billions of ordinary consumers, and each user's AI Agent needs to continuously occupy server CPU resources, the incremental demand for server CPUs will no longer be simple linear growth, but will usher in structural expansion. At the same time, CPU delivery cycles have lengthened from the usual 16 to 20 weeks to 25 to 30 weeks, directly attributed to demand pulled by AI agents like Muse.

Bank of America raised its target price for AMD from $620 to $720 on September 28, with the core reason being that Agentic AI makes CPUs and GPUs complementary rather than substitutes. Bank of America also raised its server CPU TAM forecast from $61 billion in 2026 to $211 billion in 2030, with about $180 billion directly related to AI. The market share of ARM-architecture server CPUs is expected to approach 40% by the end of the decade, and could reach 50% under an optimistic scenario.

Previously, the market largely ignored the pull effect of the AI wave on server CPUs. Now, this significant expectation gap is undergoing systematic repricing by institutional funds.

Macro Level: A Half-Open Window

Returning to the macro level. Last night's August PCE data was indeed a surprising signal for the market: core PCE came in at 3.0% year-over-year and 0.2% month-over-month, both below expectations; headline PCE was 3.4% year-over-year, also below the expected 3.7%. After the data release, the probability of a rate hike in October fell from 45% before the release to 39%, while the probability of a December hike rose to 90%, and the two-year Treasury yield briefly plunged.

Previously, the market had been on edge, worried about extremely sticky inflation and the Fed continuing aggressive rate hikes. But this PCE report directly poured a bucket of "cooling water" on the market, and overall inflation pressure was at least temporarily eased at this moment. The urgency of rate hikes has decreased, and the chip sector, which had been suppressed by high rates for a long time, finally has room to breathe.

But it should be noted that the market is not pricing in "the end of tightening," but rather "a delay in tightening." Goldman Sachs pushed its forecast for the second rate hike from October to December, while stating that "it does not rule out the possibility that no further hikes will ultimately be needed"—the cautious wording is worth noting.

More critical is the signal from the US Treasury market. The 2-year yield briefly fell after the PCE release but then fully recovered the decline; the 10-year yield continued to rise. The short end is pricing in "delay," while the long end is pricing in "resilience." For high-duration assets like chip stocks, if long-end rates do not fall back, the space for valuation expansion remains constrained.

At the same time, second-quarter GDP was sharply revised up from 1.5% to 2.2%, and the upward revision in the investment category clearly points to the pull effect of AI infrastructure construction. August consumer spending rose 0.9% month-over-month, and real personal spending grew 0.6% month-over-month, the largest single-month increase since March 2025. The signal from these data is: the economy is not weak, demand is not bad, and the Fed has no sufficient reason to shift to easing while inflation is still 1 percentage point above target.

Therefore, the effect of PCE on chip stocks is more like a "half-open window" rather than an open door. It eased the urgency of an October rate hike and provided a short-term time window for the sector, but did not change the policy backdrop that tightening has not yet ended. The rise in chip stocks relies more on industry-side narrative drivers rather than a directional shift on the macro side.

Conclusion

At the current juncture, the bull-bear game in the US stock market is actually quite clear. On the positive side, Micron has solidified its performance foundation with hundred-billion-level long-term agreements, the memory industry's prosperity cycle expectations have been extended, and physical AI plus AI Agents have brought entirely new incremental space—the growth logic of the semiconductor sector has indeed been substantially strengthened. But at the same time, risk concerns objectively exist: US domestic demand remains resilient, inflation stickiness has not dissipated, rate hikes are only delayed rather than completely canceled, and the high-rate environment still hangs over the market.

Simply put, it is a continuous tug-of-war between industry positives and macro pressures. Going forward, the market should continue to track four core validation indicators: nonfarm payroll data, September inflation data, Middle East oil price trends, and the pace of global AI capital expenditure implementation. Only if subsequent data can continuously validate the continuation of industry prosperity while US inflation steadily declines can the chip sector's rally be expected to move from the current short-term repair toward a trend reversal.

(Full text ends)