Author: Ada, Deep Tide TechFlow
Over the past week, the US stock AI sector presented a strange picture, with records being broken one after another and stocks being sold off one after another.
On June 1, Alphabet, sitting on hundreds of billions in cash, announced one of the largest equity financings in history; On June 3 after market close, Broadcom delivered its best-ever quarterly report, only for its stock to crash the next day; On June 5, the Nasdaq plunged 4% in a single day, with the semiconductor sector shedding roughly $1 trillion in market value in one session; On June 10 after market close, Oracle reported record revenue and backlog, yet its stock still fell; On June 11, the pricing moment arrived for what is set to be the largest SpaceX IPO in history. The financial numbers themselves are not the issue; the problem is the way these numbers are being achieved: more and more money is being borrowed into this AI infrastructure race through increasingly complex means. When the market began tallying up this bill, even record-breaking numbers couldn't save stock prices.
The Same Script: Record First, Then Get Sold Off
Broadcom went first. According to its earnings report and multiple media outlets, for its fiscal Q2 ended May 3, Broadcom's revenue reached $22.219 billion, up 48% year-over-year; AI chip revenue was $10.8 billion, surging 143% year-over-year; and earnings per share exceeded Wall Street expectations. But the market focused solely on one gap: the company's AI chip revenue guidance for the next quarter is $16 billion, below analyst expectations, and CEO Hock Tan did not raise the full-year AI revenue target, mentioning that Google might diversify its chip supply chain. Nonetheless, Broadcom's stock plummeted about 15% the next day, erasing nearly $280 billion in market value in a single day, making it one of the largest single-day market cap losses in Wall Street history.
Oracle's turn came a week later. According to its earnings report and CNBC, for its fiscal Q4 ended May 31, Oracle's revenue was $19.2 billion, up 21% year-over-year; cloud infrastructure revenue was $5.8 billion, up 93% year-over-year; adjusted EPS was $2.11, beating analyst expectations of $1.95. The backlog was even more staggering: Remaining Performance Obligations (RPO) reached $638 billion, skyrocketing 363% year-over-year, far exceeding analyst expectations of $595.7 billion. Yet the stock also didn't find favor, dropping about 9% in after-hours trading.
Sandwiched between these two earnings reports was a broad sell-off on June 5. According to TheStreet and CNBC, the Nasdaq Composite Index fell 4% that day, its worst single-day performance since the April 2025 tariff turmoil, triggered by Broadcom's cautious AI chip outlook, with AMD and Intel leading the decline across the entire semiconductor sector.
It is worth noting that the June 5 plunge wasn't solely about "AI skepticism." That day, US non-farm payrolls added 172,000 jobs, far exceeding expectations, pushing up rate hike expectations and triggering a rotation from high-valuation growth stocks into defensive sectors like healthcare and consumer staples. AI stocks, having the highest valuations, fell the hardest. In other words, macroeconomic interest rates and sector rotation were one driver, while concerns over AI capital expenditures were another. The two factors compounded, rather than having a single cause.
What's Being Sold Off Isn't the Income Statement, It's the Cash Flow Statement
Putting these three market segments together reveals a commonality: the income statement is still recording "record-breaking" numbers, but the market has already switched to reading the cash flow statement and the balance sheet. The pricing focus has shifted from "how much was earned" to "how much more will be burned and borrowed to earn this."
Oracle is the most straightforward case study. According to its earnings report, for the full FY2026, operating cash flow reached a record $32 billion, up 54% year-over-year, but free cash flow was negative $23.7 billion. For the year, the company has already raised $43 billion in debt and $5 billion in equity. What truly broke investor sentiment was its forward-looking language. According to CNBC, Oracle plans to raise approximately $40 billion more in FY2027 through a combination of debt and equity. A company that just raised nearly $50 billion, with negative free cash flow, and now forecasts another $40 billion financing round—when this appears alongside "record-breaking" numbers, the market chose to price the former.
Broadcom's logic is similar, but manifests elsewhere. According to Barron's, Broadcom lowered its Q3 gross margin guidance from 77% to 74%, attributing it to a higher revenue mix from lower-margin AI chips. Coupled with a retreat from "selling complete systems" to "selling only chips," and customers demanding chip leases to pass on financing pressure, the market sees a business with explosive growth, but deteriorating margins and capital intensity.
Goldman Sachs provided a framework for this shift. According to its research report, investors' tolerance for capital expenditure growth depends on earnings strength and the visibility of AI monetization; the same report noted that Alphabet's stock rose due to raising its profit guidance, while Meta's fell due to flat guidance. The market is no longer indiscriminately rewarding "growth," but differentiating winners from losers based on "the ability to monetize."
The Financing Chain Takes Center Stage: Even the Most Cash-Rich Players Are Borrowing
If the income statement is the surface, the financing chain is the true protagonist of this past week. From the very upstream to the very downstream, almost every link is paying for this same AI infrastructure build-out by leveraging up or diluting equity.
The most convincing case study is Alphabet. According to its SEC filing, on June 1, it announced an $80 billion equity financing; on June 2, it upsized and priced the deal at $84.75 billion, including a $10 billion private investment from Berkshire Hathaway. The anomaly is that this company is not short on cash at all. According to multiple media reports, Alphabet had $126.8 billion in cash as of March 2026, annual operating cash flow of $174 billion, and has issued over $55 billion in new debt since November. Despite this, Melius Research estimates Google's free cash flow will turn negative in the coming years. Investor Dan Niles' commentary on this is that capital is not infinite, and the fact that Google, with "the strongest technology stack in all of AI," still needs to raise massive amounts of financing precisely illustrates the intensity of this round of investment.
Looking downstream, every link on the chain is doing the same. New cloud vendor Oracle has negative free cash flow, relies on dual-track debt and equity financing, and asks customers to prepay for GPU costs or bring their own GPUs to reduce its own construction capital outlay; the chip "shovel sellers" like Broadcom, on June 9, partnered with Apollo and Blackstone to establish the AI XPV Platform, with an initial $35 billion targeting over 20 gigawatts of computing power by 2028, serving frontier labs including Anthropic and OpenAI. And at the very end of the chain, the labs are simultaneously employing even more aggressive tools: there were previous reports of SoftBank arranging margin loans secured by OpenAI equity; now SpaceX is rushing towards a Nasdaq IPO targeting $75 billion, Anthropic has confidentially filed for an IPO, and OpenAI is expected to follow closely.
The total scale of this investment is also rapidly inflating. According to CreditSights, the combined capital expenditure for mega-cap companies in 2026 is estimated at around $750 billion, up about 67% from 2025; Goldman Sachs' alternative metric puts the 2026 capex forecast for mega-caps at $518 billion, revised up from $314 billion at the start of the year. Whichever metric is used, the direction is consistent: expenditures are accelerating, the portion that can be covered by operating cash flow is shrinking, and the gap needs to be filled by capital markets.
The Chain's Pressure Point Rests on a Few Unprofitable Labs
Leverage itself isn't terrifying; what's terrifying is who the leverage ultimately rests upon. Pulling this financing chain to its end reveals that its pressure point is highly concentrated.
Oracle's $638 billion backlog seems impregnable, but according to Bank of America, over 50% comes from OpenAI alone; simultaneously, Oracle also disclosed that most of the RPO growth in the past two quarters came from large AI contracts, with customers either prepaying GPU costs or procuring GPUs themselves before handing them to Oracle. Broadcom's six major custom chip customers are similarly concentrated among a handful like Google, Meta, Anthropic, and OpenAI. In other words, from the financing of the mega-caps, to the orders of the "shovel sellers," to the injection of private credit and insurance funds, the ultimate payer for the entire chain converges onto a small group of unprofitable frontier labs like OpenAI and Anthropic, who are themselves queuing up for funding.
The record-breaking revenue is real, and the $638 billion backlog is real; but the buyers of these orders are highly concentrated, are themselves sustained by financing, and the leverage of the entire chain is being re-examined by the market. This past week, the market did not deny AI's growth; it simply started demanding to see clearly who will pay the bill for this growth, and how. SpaceX is set to price after market close on June 11 and list on the Nasdaq at $135 per share, with a valuation of approximately $1.77 trillion on June 12. Whether this largest-ever IPO can be digested smoothly will be the next stress test for this financing chain.
