OpenAI與甲骨文(ORCL)事件，令市場錯誤理解人工智能目前情況，事實上整體AI產業仍非常健康，幾乎不存在泡沫。
從OpenAI宣布1.4萬億美元資本開支，已成為人工智能泡沫擔憂最大源頭；隨後甲骨文進取地為AI數據中心發展進行債務融資，導致近期出現100億美元負現金流。過去七周，睇淡AI的投資者借此「所謂壞消息」支持AI泡沫論，令人工智能股票、納指和標普500指數等，造成巨大下跌壓力。
上述「因素」不僅未能真實反映人工智能技術領域目前最新情況，更未對相關企業構成風險。相反地，近期英偉達(NVDA)、台積電(TSM)、博通(AVGO)、谷歌(GOOG)等主要人工智能企業，發布更多積極利好消息，在目前市場恐慌情緒下，明顯地被市場忽視，更被AI懷疑論者曲解為負面新聞。
儘管OpenAI的1.4萬億美元資本開支計畫或令市場不安，但此計劃只是人工智能龍頭企業的共同草擬方案，絕非OpenAI或合作夥伴（包括甲骨文、AMD、英偉達、博通等）簽訂的合約承諾，故不會影響ChatGPT持有者及其合作企業。
根據OpenAI行政總裁阿特曼預計，OpenAI可能在未來五年或更長時間，與基礎設施供應商共同投入巨額資本開支，但最終方案是靈活調配，僅取決於客戶實際的業務需求。若業務規模較細、收入與利潤較低，投入資本有限下，OpenAI也定必將縮減資本開支。該公司將按照業務動態調整開支，絕不會因為要守住此1.4萬億美元的數字而導致破產。
另一方面，包括英偉達、AMD、甲骨文在內的所有供應商，都理解此計畫彈性，不會盲目擴充從而承擔過份風險，上述頂尖人工智能基礎設施供應商，將根據客戶真實合約金額作出審慎評估，逐步調整投資。簡言之，不會被1.4萬億美元數字誤導而墮入過度擴張的風險。
甲骨文進取的人工智能數據中心建設，以至近期發行180億美元債券，以及100億美元負現金流，成為支持泡沫論的另一組重要「因素」，其債券利差擴大被引述為類似1999年科技泡沫債務融資案例，但投資者必須注意甲骨文仍是主要評級機構鑑定為屬投資級別的債券，並透過審慎管理資產負債表，力守此評級。與1999年非投資級電信與科技企業過度借貸相比，有明顯不同。
況且人工智能行業債務融資資本支出極有限，多數人工智能雲端供應商都擁有強勁資產負債表與現金流，足以支撐擴張步伐。
此外，人工智能領域近期湧現大量利好消息，反映人工智能革命潛力巨大。首先，英偉達上季紮實的「優於預期並調升財測」業績，顯示銷售增長重新加速，成為人工智能穩健增長的有力例證，但睇淡者將其歸咎於主要客戶過度支出，並斷言當需求無法跟上時，最終將崩潰。
上述懷疑論者認為亞馬遜貝佐斯、微軟納德拉、谷歌皮查伊、Meta朱克伯格與甲骨文艾里森，齊齊嚴重錯判市場需求，造成過度投資。但我相信全球頂尖商業領袖集體犯下巨大錯誤機會極微。相反，科技行政總裁應是基於穩固底層需求，從而作出審慎擴張的資本開支。
博通業績同樣好過市場預期，兼調升本季業績。至於未有提及2026年收入指引以及18個月內730億美元人工智能銷售額，只是管理層謹慎保守預測。與多家投資銀行與管理層深入交流後，均見大行都上調目標價。此間 ASIC晶片龍頭企業，正邁向人工智能增長道路。
最後，從谷歌、Anthropic等AI企業均見到大量利好消息，也表明人工智能增長動力仍充沛，兼且持續擴張，近乎所有華爾街機構都對人工智能領域持積極態度，認為AI技術革命仍處早中期階段，仍遠未達成熟或飽和點。多數人工智能基建與能源供應商（如輝達、台積電、博通、AMD、美光、SK海力士、應用材料、ASML、GE Vernova等），都將持續受惠AI增長浪潮，部分企業表現突出，部分則或遭淘汰。不過，整體產業可預見未來前途一片光明。
投資者應審慎評估人工智能泡沫憂慮的真實性，從而在目前做出「賣出」、「持有」或「增持」相關股票。
AI bubble fear is overdone and now is a great bargain hunt opportunity for long term gain
AI bubble has been vastly exaggerated by misunderstanding led by certain aggressive AI firms OpenAI & Oracle but broader AI industry remains very healthy with very little or no bubble at all. The ‘announcements’ of US$1.4 trillion ‘capex plan’ of OpenAI has become the biggest reason for the AI bubble fears and followed by Oracle’s debt financing of its aggressive AI datacentre expansion resulting in a US$10 billion negative cash flow recently. The AI sceptics exploited these ‘examples’ to support their AI bubble theory that has been pressuring the AI theme, NASDAQ & S&P500 for the last 7 weeks.
Not only are those ‘factors’ not truly a real reflection of the broader AI tech world, neither are them real imminent risk on related firms just yet. Instead, there are much more positive & favourable news announced recently from lots of major AI players such as Nvidia, TSMC, Broadcom, Google, etc. but are either ignored or mis-interpreted as bad news by the AI sceptics in a fearful market.
While OpenAI’s US$1.4 trillion 5-year capex plan sounds very bubble and scary, this is only a plan it has been drafting with its business partners including most AI leaders, this dizzy US$1.4 trillion is only a long-term planning soft number but not at all a committed contracts for OpenAI or its partners that could jeopardize ChatGPT owner and its announced partners such as Oracle, AMD, Nvidia, Broadcom, & others.
Based on CEO Altman’s projection, Open AI may spend this huge capex over the next 5 years or longer with these infrastructure providers but the ultimate spending amount could vary freely and likely would rely on how much business demand OpenAI receives eventually from its customers. Should it have lower business, less revenue & profit, & therefore less capital available, the AI giant would and could spend much less. The firm will adjust spending as per business momentum and will not bankrupt itself by sticking to this US$1.4 trillion figure.
On the other side, all its suppliers including Nvidia, AMD, or Oracle understand well the flexibility or non-committed nature of such plans and are not blindly expanding their capacity and supply now and risk over capacity. These top AI infrastructure providers will tune their own investment gradually on careful estimations of all customers genuine & committed contractual amount. In short, they will not be misled by this US$1.4 trillion number and risk major overexpansion.
Oracle’s aggressive buildout of AI datacentre, recent US$18 billion bond issuance, & a US$10 billion negative cash flow have been another set of major ‘factor’ supporting the bubble theory. Its credit spread widening is used as example of debt-financing like 1999 tech bubble. However, investors must note that Oracle maintains its investment grade by major rating firms and keen to keep it that way by carefully managing its balance sheet. This is totally different from the excessive borrowing by non-investment grade telecom & tech firms back in 1999. More importantly, debt financing AI capex is extremely limited in the AI sector as most hyperscalers have very strong balance sheet & cash flow to fund their expansion.
On the other hand, there have been abundant favourable news for the AI sector showing the robust strength of the AI revolution. Firstly, Nvidia’s solid ‘beat and raise’ result last quarter showing a phenomenal re-acceleration of sales growth is compelling evidence of solid AI growth. However, the bears dismiss that as overspending by its major clients which will collapse later when their underlying demand fails to catch up. These sceptics think that Amazon’s Bezos, Microsoft’s Nadella, Google’s Pichai, Meta’s Zuckerberg, & Oracle’s Ellison collectively make a gigantic mistake of over estimating demand and over-investing. I believe this collective colossal mistake is extremely unlikely by these top global business leaders. Instead, these top tech CEOs should be expanding capex carefully on solid underlying demand.
