Who Repriced the Chip Complex If No Chip Company Said Anything?

The $NASDAQ(.IXIC)$ Composite closed 2.3 per cent higher on Monday at a record, its first since June; the $S&P 500(.SPX)$ closed 1.49 per cent higher at 7,764.70, which leaves it 0.4 per cent short of its own record; the $Dow Jones(.DJI)$ closed 0.7 per cent higher. What brought risk appetite back was falling oil prices and lower Treasury yields, and no company put out an announcement that day big enough to move the whole market. The $Philadelphia Semiconductor Index(SOX)$ rose 4.3 per cent, a fifth straight session of gains. The distance that opened up between the three indices says that a handful of sectors did the lifting.

$ARM Holdings(ARM)$ led, closing 17.16 per cent higher at US$322.90, the biggest gain in the Nasdaq 100 that day; Intel closed 12.14 per cent higher at US$121.78 and $Advanced Micro Devices(AMD)$ 9.95 per cent higher at US$615.52, taking its market value above US$1 trillion at the close for the first time. None of the three put out any news of its own. What the market took as the reason was Muse, Meta's AI assistant, reaching number one among free apps in the US App Store.

From there the market shifted its reading of what the demand looks like: if people really do start using assistants that carry out tasks on their own at scale, what gets consumed is not only the GPUs used in training but the general-purpose CPUs in servers and the silicon at the edge, and that part of the bill has spent two years in the shadow of GPUs. Arm, Intel and AMD were bought on that reading. Nvidia and Broadcom were not — $NVIDIA(NVDA)$ closed 2.30 per cent higher at US$227.38 and $Broadcom(AVGO)$ 1.60 per cent higher at US$362.66. On the same AI theme, one side gained double digits and the other gained two points.

Two footnotes came from the supply side that day. Intel's chief executive, Lip-Bu Tan, said the company can meet only about half of the demand it is seeing. Piper Sandler started coverage of Arm the same day at overweight with a target price of US$320; Arm closed the session at US$322.90, above that target. The sell-side number was written against the previous session's price, and the market cleared it inside a single day.

$Meta Platforms, Inc.(META)$ itself closed 11.43 per cent higher at US$741.25, its best day since April 2025. Muse launched on 8 September, and Sensor Tower puts downloads at around 2.5 million over its first two weeks; Wells Fargo lifted its target price to US$796 the same day, and the company's annual Connect conference falls in that week. What the market has doubted about Meta for a year is whether it can still build a consumer product that people actually use, and Muse answers that directly. What it does not answer is the half that comes after: there is a long stretch between downloads and revenue, the top spot among free apps can change hands within days, and a Meta executive has conceded the product is “heavily inspired” by OpenClaw.

Memory did not join in. $SanDisk Corp.(SNDK)$ closed 1.41 per cent lower at US$1,766.64, the only faller in that group, having closed 10.99 per cent higher in the previous session — the S&P 100 reshuffle took effect on Monday, the funds tracking the index had already switched their holdings at the last close before that, and no fresh bid arrived on the day it took effect. $Micron Technology(MU)$ closed 2.77 per cent higher at US$1,043.96, back above US$1,000, and SKHY 0.73 per cent higher at US$188.86; both were still up, just nothing like the chip names.

Two opposite readings turned up on the infrastructure side. $Marvell Technology(MRVL)$ closed 5.38 per cent higher at US$257.38, having shown 38 technology demonstrations at the European optical communications conference, among them what it calls the industry's first 2nm optical interconnect, pushing AI data centre networking from 1.6T towards 3.2T; that points at the wall beyond compute, where the silicon keeps getting faster while the bandwidth and power needed to connect it sit close to their limits. $NEBIUS(NBIS)$ closed 4.14 per cent higher at US$232.80, and on the same day Rothschild Redburn put both it and CoreWeave at sell, citing falling GPU rental prices, hyperscalers building their own capacity and expensive financing. The company selling the connection is paid out of its customers' capital budgets; the company renting out compute is paid in rent, and the rent is coming down.

Taken apart, what was repriced on Monday was an entire processor chain, and the basis given for it was where one free app sits in an app store. That position can change hands within days, while the contracts for capacity, foundry work and packaging are signed by the year. What the market bought that day was a change in the shape of demand, not anything a company had newly said out loud.

The above is personal analysis, not investment advice.

💬 【Talking Point】

$Intel(INTC)$ chief executive said on Monday that the company can meet only about half of the demand it is seeing, and Piper Sandler put a US$320 target on $ARM Holdings(ARM)$ that same day, which the shares had already passed by the close at US$322.90. What would have to show up in an order book before that CPU-side demand counts as real?

💰 【Bounty】

Drop your view in the comments and there are Tiger Coins in it for you! 🎁

🔔 Better shared than saved — tag a friend and split the coins!

# Two-Thirds of Next Year's Capacity Already Sold — Who's Still Shorting Memory?

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Report

Comment4

  • Top
  • Latest
  • 吉3186
    ·09-22 19:39
    The order book needs to show real customer commitments, not just strong demand claims.
    I would look for these 5 things:
    Large server CPU orders from companies like Meta, Microsoft, Amazon, or Google.
    Multi-year contracts, showing customers plan to keep buying.
    Higher shipment volumes quarter after quarter.
    Backlog growth — orders waiting to be delivered keep increasing.
    Revenue growth matching the orders — orders must eventually become real sales and cash flow.
    For Intel, “we can only supply half of demand” sounds positive, but the important question is: Are customers actually placing large, long-term orders?
    For AMD and ARM, the same rule applies.
    Bottom line:
    Demand talk = interesting.
    Signed orders + rising shipments + rising revenue = real proof.
    I would trust the second one much more.
    Reply
    Report
  • 靖润
    ·13:21
    要让CPU的需求算“真的”,光靠Metaverse的App下载量没用,得看到真金白银的采购合同。


    现在Intel说“只能满足一半需求”,这更多是卖方惯用的预期管理。App Store排行榜几天就能翻篇,但晶圆代工和封装的合同是按年签的。要让市场信服AI Agent真的引爆了CPU,订单簿上必须出现两个硬指标:


    第一,云厂商的服务器采购订单。得看到微软、亚马逊这些大客户在通用计算(CPU)上的资本开支预算实质性上修,并且明确是为了部署AI Agent推理,而不是传统的Web服务。


    第二,Intel和AMD的数据中心CPU营收指引大幅上调。现在光靠“供需紧张”喊口号没用,得在财报里看到长期的、锁定价格的订单(LTAs)。Intel自己在财报没出之前喊缺货,市场信了,股价涨了,但这只是情绪层面的抢跑。如果下个季度财报拿不出暴涨的服务器订单数据,现在冲进去的资金马上就会作鸟兽散。


    截图里那位作者说得很对:周一市场买的只是“需求形态的变化”,而不是任何公司刚宣布了什么。ARM市值直接冲破机构给出的320美元目标价,这本身就是极度情绪化的逼空,跟基本面彻底脱钩了。


    在这种时候,去赌财报指引超预期,性价比极低。等订单簿上的白纸黑字真出来了,再重新定价也不迟。
    Reply
    Report
  • Mkoh
    ·09:13
    Macroeconomic data and monetary policy shifts: Stronger-than-expected economic reports (e.g., jobs data), rising Treasury yields, inflation readings, or changing expectations around interest rates/Fed actions. These affect high-growth, high-valuation tech stocks more sharply by altering discount rates on future cash flows.
    Broader market sentiment and risk appetite: Overall equity market moves, risk-on/risk-off shifts, or rotation out of (or into) growth/tech sectors into value, defensive, or other areas. Chip stocks often amplify broader moves due to their cyclical and high-beta nature.

    Market structure and positioning dynamics: Crowded trades, high leverage (including single-stock leveraged ETFs, especially in markets like Korea), retail options activity, forced unwinding of positions, or momentum-driven flows. These amplify moves and create volatility independent of fundamentals.
    Index and fund flows: Rebalancing of semiconductor indices (e.g., SOX), ETF creations/redemptions
    Reply
    Report
  • Jerry Lam
    ·09-22 20:21
    我觉得 CPU 端需求要从“故事”变成“真实”,不能只看 CEO 说供不应求,也不能只看股价一天涨多少,最后必须回到订单簿和盈利兑现。

    我会重点看四个信号:

    第一,Backlog 是否连续几个季度增长。
    而且最好是服务器 CPU、AI Head Node 等高价值产品贡献,而不是普通 PC 补库存。

    第二,客户承诺是不是越来越长。
    如果只是临时抢货,说明可能是短缺;如果 hyperscaler 开始签更长期采购、扩大下一代平台部署,才说明需求具有持续性。

    第三,ASP 和毛利率能不能同时改善。
    真正的供不应求应该带来更强议价能力。如果出货增加但价格、利润率没有改善,那可能只是产能瓶颈,并不一定创造更多股东价值。

    第四,也是我最看重的:自由现金流。
    因为“只能满足一半需求”听起来很强,但如果为了满足需求需要投入更多 CapEx,最后现金流仍然很差,那这个短缺的价值就要打折。

    所以我不会因为 Intel 说只能满足一半需求,就直接认定 CPU 已经进入新超级周期。

    我更想看到的是:

    订单越来越多 → 交付越来越快 → ASP不降 → 毛利率改善 → 自由现金流增长。

    如果这条链真的跑通,那 Muse 带来的就不只是一次题材炒作,而可能是市场第一次提前看到 Agent 时代服务器 CPU 需求结构发生变化。

    一句话:

    缺货证明需求存在,订单证明需求持续,利润和现金流才证明这份需求真的值钱。

    Reply
    Report