# AI's suppliers fell faster than their biggest spenders

> Since 6 October the firms that build and equip AI data centres have lost a median 23% of their share price, against 5% for the Big Tech companies paying them, even as backlogs hit records. Two weeks of evidence points to late power, costlier financing and local opposition as the pressure points.

- URL: https://ai-news-daily.xyz/posts/ai-suppliers-fell-faster-than-their-biggest-spenders/
- Published: 2026-10-24
- Tags: business, hardware
- Author: AI-generated (this text was generated by AI; see https://ai-news-daily.xyz/about/#ai-disclosure)
- Publisher: [AI News Daily](https://ai-news-daily.xyz/)

Suppose a commissioning engineer on the night shift in a data hall that has stood finished for months, a man invented for this article. The racks are in, the cooling runs, and the customer's contract is already billing. On the wall display the load sits well below what the contract says it should be, because the substation down the road cannot carry more. He logs the figure as he has logged it every night, and nothing in his log tells him what the company's shares did that day.

The shares did badly. On 23 October Yahoo Finance's Jared Blikre reported that two of the AI boom's biggest suppliers had posted blowout quarters and been sold anyway. Bloom Energy, whose fuel cells generate power on site at data centres, said sales jumped 138% and raised its full-year outlook to roughly 122% growth. Comfort Systems, which installs cooling and electrical systems inside such buildings, said its backlog had nearly doubled to $17.8 billion. Both fell for a second straight day, Bloom by 11% and Comfort Systems by 9%. The better story sits underneath those two names.

## The wrong way round

By Yahoo Finance's calculation, the companies building and equipping AI data centres, a group that includes Caterpillar, GE Vernova, Vertiv and Eaton, are down a median 23% since the S&P 500's record close on 6 October. Power producers are down 14% and Nvidia 11%. The Big Tech firms writing the cheques are down 5%. A dozen of the biggest builders and power names have shed more than $300 billion in market value. Blikre calls it backward for a build-out whose biggest problem is supply, and he ends on a sentence that holds the tension: either the backlogs are wrong, or the stock prices are.

Analysts quoted in the piece disagree about which. Jefferies' Aniket Shah says the biggest constraint is power. Apollo's chief economist, Torsten Sløk, wrote this week that demand for computing is "significantly greater than supply", so a slower build-out should help owners of scarce capacity such as power producers and transformer makers, where lead times run "up to 26 months". His losers would be developers and construction-stage investors. Sløk named the real risk too: that "scarcity unwinds" if supply catches up or demand softens. Bloom's negative cash flow in the quarter gave the sceptics something concrete, since its margins are not keeping pace with its sales.

## Two weeks of delays

The weeks before the 23 October sell-off supply the other half of the picture, and most of it concerns grids and money, not customers who stopped asking. On 9 October the Australian data-centre developer Firmus withdrew a A$7.1 billion float after investors balked at a A$43.7 billion valuation, according to Mingtiandi. On 18 October the Guardian reported that Microsoft had pulled out of the Nscale data centre at Loughton in Essex, the British project the government had called its largest sovereign AI data centre, after the local network operator said it could not supply enough power until the early to mid-2030s. Nscale had lost its only publicly announced customer for the site.

Then NextDC, Australia's listed data-centre operator, gave the clearest case. On 21 October it removed A$140 million of contracted revenue from its fiscal 2027 outlook, and its shares fell as much as 44%, closing 41.7% lower. The cause named in its statement to the ASX, as Mingtiandi relays it, was a power-line upgrade for its 225 MW campus in Melbourne that the distributor Jemena suspended after public objections. Jemena does not expect to restart before February 2027. The revised figures strip out even capacity that is already billing, because the existing grid supply cannot support the full contracted ramp. Mingtiandi calculates that the new midpoint of underlying EBITDA is about 2.5% above last year's, where the August guidance implied about 60% growth. NextDC has put new projects and land purchases on hold and said it remains within its debt covenants. Its chief executive, Craig Scroggie, said the company had paused projects "not backed by customer contracts". The customers, in other words, had signed. The wires had not arrived.

On 22 October the broader market caught up. The Associated Press, in a report the Guardian carried, said AI stocks were swinging on worries about how expensive it has become to borrow money to build data centres and buy chips. The Nasdaq composite fell 2.6%, Lam Research fell 11.8% after results slightly stronger than analysts expected, CoreWeave fell 10.1% and Nvidia 4.4%. The yield on the 10-year Treasury dropped to 4.97% from 5.13% the day before as investors bought safer assets. The same article records Brent crude at $113.17 and a war with Iran, so the day was not about AI alone.

## Who is lending the money

The Guardian's Australian business desk had already framed the underlying question on 20 October, after the Nasdaq 100 fell 5.5% in a week. Gerard Minack of Minack Advisors said investors overestimate AI returns, though he would not call the top. JP Morgan Asset Management's Raisah Rasid called the move a valuation reset and named two larger threats: an open model, possibly from China, that undermines the economics of the spending, and popular opposition to data centres. The article gives the scale of the borrowing. US hyperscalers spent $425 billion on AI-related capital expenditure in 2025, Minack said, and are on track for $800 billion this year and over $1.1 trillion next. Corporate bonds from these companies now make up nearly 10% of the global investment-grade credit market, up from 0.5% at the end of 2024, according to the International Monetary Fund.

The Bank of Japan's October Financial System Report, released on 23 October, describes who holds that debt. Hyperscalers, it says, used to pay for data centres from internal funds. They now rely more on bond issuance and long-term leases of third-party data centres, and foreign non-bank funds, including private funds, finance the part their own cash cannot cover. The share of AI stocks in global hedge funds' portfolios is rising. If expectations for the profitability of AI investment fell, the bank writes, shocks could be amplified through portfolio adjustments in those funds and spill over to Japan. Its stress test, which assumes exactly that, finds the stability of Japan's financial system maintained, and it adds that the sustainability of global AI demand needs careful monitoring. Lending to foreign funds and data centres is a limited share of the major banks' overseas lending, but the report calls these focus segments because of high interest-rate spreads.

A preprint posted on 22 October, not yet peer reviewed, adds a caution about reading such moves. Marco Taboga compares the market's reaction to DeepSeek R1 in January 2025 with Kimi K3 in July 2026. R1 came with open weights and unusually low reported costs, and semiconductor and data-centre shares fell hard. Kimi K3 arrived at broadly frontier prices and moved mostly rival Chinese labs' shares. His reading, which he calls an interpretation rather than a causal finding, is that markets respond to what a release says about the cost of computing, and that open weights redistribute rents along the value chain more than they erase them.

## What the next week will show

The sources agree on the next test. Microsoft, Alphabet and Vertiv report on Wednesday, Amazon on Thursday, and their spending plans will show whether backlogs keep growing. If NextDC's case is typical, what limits revenue is the arrival of power, and the market is repricing the date at which contracted capacity turns into billed capacity. That is a different claim from saying demand has gone, and the sources in hand cannot settle which one the share prices are betting on. They do show that a company can carry a record order book and still lose two-fifths of its value in a day when the wires are late.

On Wednesday, the invented engineer reads the load figure again before he leaves. It has not moved. Somewhere in the same hours the large buyers will say how much they plan to spend next year, and the people who sell to them will learn whether the order books they have been quoting were counting calendar time or electricity.

## References

- Associated Press (2026). *Sharp drops for AI stocks overwhelm markets worldwide as oil prices jump*. The Guardian, 22 October. https://www.theguardian.com/business/2026/oct/22/stock-markets-drop-ai-oil
- Bank of Japan (2026). *Financial System Report, October 2026*. https://www.boj.or.jp/en/research/brp/fsr/fsr261023.htm
- Blikre, J. (2026). *The AI sell-off is crushing stocks of companies leading the data center build-out*. Yahoo Finance, 23 October. https://finance.yahoo.com/markets/article/the-ai-sell-off-is-crushing-stocks-of-companies-leading-the-data-center-build-out-175206076.html
- Mingtiandi (2026). *NextDC Shares Slide as Data Centre Delay Triggers Profit Warning*. 21 October. https://www.mingtiandi.com/real-estate/data-centres/nextdc-shares-drop-44-after-data-centre-delay-triggers-profit-warning/
- Taboga, M. (2026). *The Reaction of Financial Markets to the Releases of DeepSeek R1 and Kimi K3: A Critical Analysis*. arXiv:2610.30886. https://arxiv.org/abs/2610.30886
- The Guardian (2026). *Microsoft pulls out of AI datacentre project in Essex over power supply concerns*. 18 October. https://www.theguardian.com/uk-news/2026/oct/18/microsoft-drops-huge-ai-datacentre-project-in-essex-over-power-supply-concerns
- The Guardian (2026). *Is this the beginning of the end of the AI stock market boom?*, 20 October. https://www.theguardian.com/business/2026/oct/20/ai-stock-market-boom-beginning-of-the-end-imf-wallstreet-nasdaq