Broadcom expects $230 billion in AI chip sales in 2028 – but warns of capital shortage
On Monday, October 12, 2026, Broadcom CEO Hock Tan pointed to capital — not a shortage of chips or power — as the AI industry's most likely constraint. The following Tuesday, Samsung fell 8 percent and SK Hynix 4 percent in Seoul, and the KOSPI was briefly halted. At the same time, fresh IDC figures show Nvidia's share of China's AI chip market falling as Huawei closes the gap — two events the sources do not connect.
What Happened in Seoul
The Korean stock market reacted sharply on Tuesday, October 13. Samsung Electronics fell 8 percent, SK Hynix fell 4 percent, and South Korea's main index, the KOSPI, was briefly halted as selling intensified, according to a roundup from Stocktwits via MSN (ef827938). In overnight US trading, the tripled-leveraged South Korea ETF Direxion Daily South Korea Bull 3X Shares (KORU) fell 11.4 percent.
The trigger, according to the same source, was investors reassessing the sustainability of AI infrastructure spending after warnings about financing constraints from the top executives of Broadcom and Intel the previous Monday. The Korean companies are among the world's most important suppliers of memory chips used in AI data centers, and are therefore directly exposed to how long the flow of capital into the sector lasts.
Several details remain uncertain: the market data come from a single roundup, and it is not known how long the KOSPI halt lasted, where the indices ultimately closed, or whether the selling continued beyond Tuesday.
What Hock Tan Actually Said
Tan spoke on Monday, October 12, at the DBS Asia Future Forward Conference in Singapore. According to CNBC, relayed via Quartz and Yahoo Finance, he said that financing could overtake supply chain bottlenecks as the main brake on AI growth: "What I think might be the biggest constraint is capital" (4b9fd3a8).
According to Reuters, he likened parts of the investment flow into the AI sector to "casino gambling" (8452cb3e), and said capital would be the biggest constraint the company faces as AI development continues. Via Bloomberg, as quoted by the New York Post, he also used an arcade analogy: "When you are a kid, you go to an arcade and play. Your parents give you $10 in coins, you spend them in ten minutes, and you have learned nothing. That is what is happening in corporations today" (a7d0522a).
The same source quotes Tan as saying that AI is not yet profitable: "For now, it's still losing money. So honestly it says nothing at this point, except that there is a huge potential for revenue growth in delivering frontier models for generative AI."
An important caveat applies to all these quotations: no transcript or recording from the conference exists among the sources, and the wording comes via CNBC, Reuters and Bloomberg. The renderings may vary somewhat in phrasing and emphasis.
Intel CEO Lip-Bu Tan, according to CNBC, expressed similar concerns about capital cost and returns. Both CEOs, according to the source, pointed out that financing constraints could prove harder to overcome than the much-discussed shortages of semiconductors, memory and power.
The Warning With an "But" — Tan Was Not Simply a Bear
The message was not uniformly negative. According to CNBC, Tan said demand for computing power exceeds what the industry can deliver, and that the imbalance could persist through 2029 and potentially into 2030. Constraints on wafer production, by contrast, could ease around 2028 or 2029 (4b9fd3a8).
Broadcom also has its own, highly ambitious projections: in September, the company forecast roughly $58 billion in AI chip sales for fiscal 2026, around $115 billion in 2027 and $230 billion in 2028 (fd88d047). These are the company's own estimates, not established facts.
The tension between the warning and the forecasts has already generated debate. Tae Kim, author of the technology newsletter Key Context, argued on X that the market misread Tan: "Broadcom CEO Hock Tan was actually incredibly bullish on AI demand and OpenAI. He was only criticizing those SELLING AI stocks" (ef827938). Whether Tan issued a bearish warning about the sector or was instead criticizing investors and share sellers is genuinely contested — and the sources do not allow that question to be settled.
According to Reuters, Tan said at the same time that Anthropic and OpenAI have shown they can generate significant revenue from their latest models (8452cb3e) — but significant revenue does not prove profitability, as the New York Post quotation also conveys.
How AI Is Financed Now — With Debt
Tan's point about capital has a concrete mechanism behind it. According to Yahoo Finance and Quartz, Broadcom is in the process of raising $60 billion in debt financing to help Anthropic and other companies procure AI chips and infrastructure. Banks are arranging a senior loan secured on $42 billion, while Blackstone is leading a junior tranche of $18 billion (4b9fd3a8).
The fact that some of the chips are to be paid for with newly raised debt illustrates precisely the dependency Tan points to: the buildout is constrained not by factory capacity alone, but by how much capital credit markets are willing to provide — and at what price.
The Parallel Story: The China Numbers
In the same news cycle, Reuters reported IDC data showing a geopolitical shift in China. Nvidia's share of the Chinese data center market for AI chips fell to 34 percent in the first half of 2026, while Huawei reached 28 percent. Nvidia delivered 678,000 AI accelerators in China between January and June, down 44 percent from the year before, while Huawei delivered 572,000 units, up from more than 300,000 the year earlier. US export restrictions limit Nvidia's availability, while Chinese competitors gain ground (ef827938).
In the non-GPU segment, which Chinese state agencies, internet companies and telecom operators are adopting, Huawei, T-Head, Kunlunxin and Cambricon together account for 95 percent, according to the Reuters report.
Two things are worth noting. First, these figures rest on IDC data as relayed by Reuters via Stocktwits; the underlying IDC report is not among the sources. Second: the sources establish no link between the China figures and the Korean stock decline. These are two parallel stories — one about the role of capital in the global AI buildout, one about how export restrictions are reshaping the Chinese market.
What We Don't Know
Several questions remain open: the duration of the KOSPI halt and the closing levels are not documented, and it is unclear whether selling continued beyond Tuesday. No primary source from the Singapore conference confirms Tan's exact wording. The underlying IDC report is not available to verify the China figures. And whether Tan's warning was a well-founded signal about the sector's financing needs or a misread comment aimed at share sellers — as critics claim — remains unresolved. What is striking, however, is that the man behind one of the industry's most optimistic sales forecasts, up to $230 billion in 2028, is himself pointing to money as the narrow pass.