Investors started questioning whether data centres can be built on schedule

The $NASDAQ(.IXIC)$ fell 5.54 per cent, $Invesco QQQ(QQQ)$ declined 6.49 per cent and the $S&P 500(.SPX)$ lost 4.36 per cent on Thursday. AI hardware and capital-intensive infrastructure fell further: $SanDisk Corp.(SNDK)$ dropped 22.26 per cent, $Micron Technology(MU)$ 17.48 per cent, $Advanced Micro Devices(AMD)$ 17.28 per cent, $Marvell Technology(MRVL)$ 16.48 per cent, $NEBIUS(NBIS)$ 14.89 per cent, $SK hynix(SKHY)$ 13.40 per cent and $Taiwan Semiconductor Manufacturing(TSM)$ 10.75 per cent.

AI demand did not suddenly disappear. The change was that power use, environmental impact, local permits and regulatory costs returned to the centre of the data-centre debate after the election. Senator Sanders's S.4214, introduced in March, remains before a Senate committee. It is a moratorium proposal, not a federal ban in force. Markets nevertheless price the risk early: slower approvals can delay revenue while capital expenditure continues.

Asset-light software separated from capital-intensive infrastructure

$Alphabet(GOOG)$ fell only 1.97 per cent, while Nebius lost 14.89 per cent. Large platforms have advertising, cloud revenue and operating cash flow, allowing them to adjust where and when they build. Neoclouds depend more heavily on financing, timely facility delivery and rising utilisation. A one-quarter delay can bring interest and depreciation forward while pushing revenue back.

$Cloudflare, Inc.(NET)$ closed down 8.28 per cent and rebounded about 15 per cent after results. Third-quarter revenue rose 39 per cent to $782.5 million and adjusted earnings reached $0.38 a share. The company raised full-year revenue guidance and expects GAAP profitability in the fourth quarter. AI inference traffic was already becoming revenue and profit, giving it a shorter return path than building capacity first and filling it later.

$IREN Ltd(IREN)$ closed down 16.95 per cent and rebounded about 10.5 per cent after results. First-quarter financial-year 2027 revenue was $228 million, including $200 million from AI Cloud. Adjusted EBITDA was $66 million and the net loss was $263 million. Its target of more than $4 billion in annualised revenue is contracted and funded, but the timing moved from the end of 2026 to January 2027 and could extend into February. Demand remained; delivery timing determined when revenue could be recognised.

Memory fundamentals did not reverse, but valuations deleveraged first

SanDisk's previous results and next-quarter guidance both exceeded market expectations, with enterprise SSD demand still strong, yet the shares fell 22.26 per cent. The move looked more like compression in a high-expectation asset than a sudden deterioration in reported shipments.

$Micron Technology(MU)$ fell 17.48 per cent. Tight HBM supply and multi-year customer agreements continued to support demand, but delayed data-centre projects could change when server-memory orders convert into revenue. SKHY fell 13.40 per cent and also faced uncertainty around equipment-export licences for its Dalian plant. The issue complicates the funding and timing of Solidigm's planned roughly $45 billion NAND project in New York, but public reporting did not say that the project had been cancelled.

Strong results could not offset a higher policy discount

$Advanced Micro Devices(AMD)$ fell 17.28 per cent only two days after reporting 97 per cent growth in Data Centre revenue and guiding fourth-quarter revenue to about $17 billion. The results did not deteriorate in one day; the valuation investors were willing to pay for future expansion did. Demand for MI450 and Helios still has to pass through customer capital budgets, facility construction and power approvals before becoming final revenue.

$Taiwan Semiconductor Manufacturing(TSM)$ fell 10.75 per cent as the same risk moved upstream. Advanced-node orders depend on customers' expansion plans, so delayed data-centre projects can change expectations for future utilisation. Marvell dropped 16.48 per cent and also carried dilution, a long closing period and integration costs from its all-stock Lumentum acquisition. A lower buyer share price further reduces the value of a fixed exchange consideration.

Intel and Tesla carried company-specific risks as well

$Intel(INTC)$ fell 10.35 per cent. Third-quarter adjusted earnings exceeded expectations, but its fourth-quarter revenue and profit outlook was weak, while PC demand and a factory accident also weighed on the forecast. Sector de-risking amplified the move, but Intel still needs server revenue and foundry investment to convert into profit more quickly.

$Tesla Motors(TSLA)$ fell 10.80 per cent. Robotaxi, AI and businesses associated with Elon Musk make its valuation more sensitive to the political environment. When political relationships and regulatory direction are repriced after an election, companies with more value tied to distant businesses can move more sharply. That does not mean the automotive business deteriorated by the same amount in one session; it means the policy discount increased.

Oklo and Vistra report pre-market before payrolls at 21:30

Oklo and Vistra report in the US pre-market on 6 November. Consensus expectations for Oklo are a loss of about $0.27 a share and approximately $1 million in revenue. Regulatory approvals, fuel supply, the timetable for its first project and data-centre power agreements matter more than current revenue. Vistra is expected to report about $6.03 billion in revenue and $2.98 in earnings per share. Power prices, fleet cash flow and Cogentrix integration will test whether power producers can capture AI electricity demand. These figures are consensus expectations, not company guidance or signed orders.

The October US employment report is due at 21:30 Beijing time. Economists expect roughly 60,000 new jobs after 29,000 in September, with unemployment expected to remain at 4.2 per cent. A strong report could lift Treasury yields and pressure valuations; a weak report could improve rate-cut expectations while increasing earnings concerns. Markets need a release that does not raise rate and recession risks at the same time.

The sell-off repriced the return timeline rather than disproving AI demand

Cloudflare's rebound showed that AI software traffic could still be rewarded. IREN's rebound alongside a delivery delay showed that a contract is not current revenue. The decline in memory, chips and neoclouds showed what happens when a longer construction period pushes cash returns further out.

The next test is not another statement that AI demand is strong. Projects must receive approvals, equipment must arrive on time, utilisation must rise and cash must remain after the investment. Demand is still present, but the market is now pricing companies by how soon their returns can arrive.

A note on sources: percentage moves for the Nasdaq Composite, QQQ, the S&P 500, AMD, Micron, Marvell, TSMC, SanDisk, SKHY, Intel, Nebius, Alphabet and Tesla use the 5 November US regular-session close. Earnings-reaction exception: Cloudflare and IREN use extended trading after their results. The status of Senator Sanders's S.4214 comes from congressional records and remains a proposal before committee. SanDisk, Intel, Cloudflare and IREN figures come from their current disclosures. The SK hynix / Solidigm project and Dalian export-licence details come from Reuters reporting citing people familiar with the matter, not a company announcement suspending the New York project. Oklo and Vistra timing comes from company investor-relations information; preview figures are market consensus expectations. The payrolls time comes from the Bureau of Labour Statistics schedule. Interpretations of valuation, financing and policy effects are analysis. This is personal analysis and not investment advice.

💬 【Talking Point】

Do you agree that this sell-off repriced construction timelines rather than signalling a reversal in AI demand?

💰 【Bounty】

Share your view and reasoning in the comments for a chance to earn Tiger Coins! 🎁

🔔 Better shared than saved — tag a friend and join the discussion!

# Robotaxi Promise Meets Congress — Which Risk Wins?

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

Comment(3)

  • Top
  • Latest
  • Jerry Lam
    ·16:54
    🐯 **我认同这轮抛售主要是在重新定价AI数据中心的建设周期和投资回报,而不是AI需求突然逆转。**

    过去市场看的是订单有多大、资本开支增长有多快;现在更关注**数据中心能否按时通电、设备能否交付,以及收入何时真正转化为自由现金流。**

    GOOG仅跌约2%,NBIS却跌近15%,正好体现了两种商业模式的差异。大型科技平台有成熟业务支撑现金流,而新云企业高度依赖融资、建设进度和设备利用率。

    **我认为接下来AI板块会进一步分化:**

    ① **GOOG、MSFT:** 现金流和业务多元化提供一定缓冲。

    ② **NVDA、AMD、MU:** 需求仍有支撑,但需要警惕客户延迟采购及估值压缩。

    ③ **NBIS、IREN:** 增长潜力大,但项目延期、融资成本和折旧压力可能放大风险。

    尤其值得注意的是,**合同金额不等于已确认收入,EBITDA增长也不等于自由现金流改善。**

    我的策略是暂不急于重仓抄底,优先观察科技巨头资本开支指引、数据中心交付进度及信用利差变化。

    **AI产业长期前景可能依然向好,但未来股价表现将更多取决于谁能最快把AI需求变成真实利润和现金流。** 🐯

    Reply
    Report
  • 吉3186
    ·15:52
    这次抛售更像是市场重新评估AI基础设施的建设时间和投资回报,而不是AI需求突然消失。

    主要有三个原因:

    1. 建设可能延迟: 电力、环保、土地和政府审批,可能推迟数据中心项目,影响AMD、MU、SNDK、TSM等公司的订单和收入。

    2. 资金压力增加: 数据中心尚未投入运营,利息、折旧和建设成本却已经发生,对IREN、NBIS等需要大量融资的公司尤其不利。

    3. 高估值开始修正: 即使AMD、美光、闪迪业绩不错,只要未来增长可能延迟,投资者就会降低估值。路透社也指出,市场正在担忧AI资本开支的回报及融资压力。

    我的投资看法

    股票 重点观察

    GOOG 现金流及AI商业化
    AMD AI芯片订单及利润率
    MU HBM需求及内存价格
    SNDK 企业SSD需求及库存
    TSM 先进制程订单
    IREN、NBIS 融资、交付及现金流

    最终结论

    我认为AI长期趋势仍然存在,但短期风险明显增加。

    过去市场关注AI能带来多少收入,现在市场更关注投入这么多钱,究竟什么时候才能赚回来。

    我的策略是:

    不因暴跌就急着抄底。

    不把所有AI股票当成同一种投资。

    等订单、现金流和盈利能力稳定后,再考虑分批买入。

    记住:股价跌了不代表便宜,必须确认公司的未来盈利没有同步恶化。

    Reply
    Report
  • 靖润
    ·15:45




    这波史诗级暴跌,根本不是在交易“AI需求消失”,而是在极其冷酷地重新定价“物理建设时间表”和“监管摩擦成本”。


    剥开表象,就三层逻辑:


    1. 需求没丢,卡在物理层:大模型还在迭代,算力订单也没砍。真正卡脖子的是电力不够、电网批不下来、环保监管(民主党上台后更严)。
    2. 高息环境下的“时间惩罚”:原计划2026年建好的机房,现在要拖到2028年。在5%无风险利率下,晚两年的现金流折现到今天,估值直接腰斩。
    3. 重资产的死穴被点中:台积电、美光扩产的机器已经折旧了,机房建不起来,收入收不回来,现金流直接断层。
    夜盘SNDK、MU、TSM的反弹纯粹是空头回补,别当反转去接飞刀。
    坚决避开靠发债买卡的纯算力租赁和纯硬件厂。既然基建卡在“缺电”和“审批”上,资金就该去“卖水人”——电网设备、液冷温控、微电网(比如ETN、VRT),这才是确定性的物理瓶颈。
    现在,现金为王,看戏,等杀出真正的深坑再捡。
    Reply
    Report