Broadcom CEO Hock Tan: "Capital is the biggest constraint" as cash flow tightens for big tech infrastructure spending.
Samsung Electronics and SK hynix see shares tumble despite record operating profits and bullish outlooks.
Even as the global AI semiconductor shortage persists and U.S. "Big Tech" firms report massive earnings, heads of global semiconductor companies are publicly raising concerns over the financial burden of scaling AI investments. As capital requirements for data centers and advanced chips accelerate, the focus is shifting toward whether investment resources and profitability—rather than just semiconductor production capacity—will determine the growth trajectory of the AI industry.
Speaking at the "DBS Asia Future Forward Conference" in Singapore on the 12th, Broadcom CEO Hock Tan stated, "The biggest constraint could be capital." He argued that merely increasing the supply of chips and electricity cannot resolve the colossal investment costs required to expand AI infrastructure.
Lip-Bu Tan, a member of Intel’s board and former executive, also identified capital and the cost of funding as immediate challenges. "In the long run, people will want to see the return on investment (ROI)," he noted, emphasizing that the massive funds poured into AI must eventually translate into tangible business performance.
Surging AI Capex Pressures Cash Flow: $1.5 Trillion in External Funding Required
U.S. tech giants such as Microsoft, Alphabet, Amazon, and Meta are generating immense cash flows from their core businesses in advertising, cloud computing, and software. AI is contributing to revenue growth through cloud service expansion and improved ad efficiency.
However, capital expenditure (CapEx) for AI infrastructure is growing at an alarming rate. According to Barclays Private Bank, the combined cash-based capital expenditure of these four companies in the second quarter of this year reached $165 billion, an 87% surge compared to the same period last year. During the same period, their combined free cash flow—cash generated from operations minus capital expenditure—was a mere $6.7 billion, as most of their earnings were reinvested back into infrastructure.
The scope of investment extends far beyond chips. It includes data center sites, power grids, cooling facilities, and network equipment. As the race for AI model performance intensifies, the demand for computing power and electricity increases, driving up the initial capital requirements.
Morgan Stanley estimates that a total of $2.9 trillion in AI infrastructure investment will be needed by 2028, with $1.5 trillion of that dependent on external financing. While major tech companies have largely funded projects through internal cash flows thus far, the role of bond markets and financial institutions may become significantly more critical in the future.
Broadcom CEO Hock Tan also questioned the profitability of companies adopting AI. He pointed to the reality that many firms are spending heavily on AI integration without clear outcomes. "Most companies are not finding a return on investment," Tan noted. He likened the situation to a childhood experience: "You take $10 from your parents to an arcade, spend it all in 10 minutes, and learn absolutely nothing."
That said, neither executive is predicting a sharp slowdown in AI demand. Hock Tan suggested that the expansion of AI infrastructure could continue beyond 2029 into 2030, and Lip-Bu Tan echoed expectations for strong demand over the next two to three years. Their remarks focus less on AI demand itself and more on the ballooning costs and subsequent profitability challenges.
Samsung and SK hynix: Record Profits Met by Foreign Selling
What of the suppliers? Samsung Electronics and SK hynix, the primary beneficiaries of the AI investment race, are seeing a divergence between their financial performance and stock prices.
On the 8th, Samsung Electronics announced preliminary Q3 results with sales of 195 trillion won and an operating profit of 107.4 trillion won. The operating profit represented a 782.5% increase year-on-year, marking a record quarterly high.
SK hynix is also expected to post record-breaking figures. Meritz Securities estimated on the 12th that SK hynix would report a Q3 operating profit of 80.1 trillion won—a 32% increase from the previous quarter. Analyst Kim Sun-woo attributed this improvement to increased shipment volumes and rising average selling prices (ASP) for both DRAM and NAND flash.
The strength of both companies is underpinned by the supply shortage of High Bandwidth Memory (HBM) and server DRAM, alongside rising memory prices. As AI data center investments have surged, supply has struggled to keep pace, significantly boosting the profitability of memory manufacturers.
Yet, stock prices have moved in the opposite direction. On the 12th, Samsung Electronics closed down 4.39% at 50,500 won, while SK hynix fell 4.52% to 160,500 won. Samsung had already seen a 2.42% drop on the 8th, the day it announced its preliminary results.
Foreign investors led the sell-off. On the 12th, foreign investors net sold approximately 2 trillion won in KOSPI stocks, pushing the KOSPI index down 3.65% to 2,638.42. The market sentiment was further dampened by rising U.S. Treasury yields, instability in international oil prices, and profit-taking after the rapid rally in semiconductor stocks. U.S. chip stocks also weakened, with the Philadelphia Semiconductor Index falling roughly 2.7% on the 12th, while Broadcom and Intel shares declined 1.9% and 3.2%, respectively.
Despite record earnings, the share prices have plummeted, reflecting the market’s focus on the sustainability of the current semiconductor boom and what lies beyond the current cycle.
Accelerating Expansion in Korea: Long-Term AI Demand remains Key
Samsung Electronics and SK hynix are expanding production facilities to meet AI memory demand. Samsung is reinforcing its capabilities in advanced memory, foundry, and packaging, while SK hynix is focusing on expanding production capacity centered on its Yongin semiconductor cluster.
SK hynix is currently pushing to shorten the construction schedule for its Yongin facility. There is also discussion of establishing additional production bases in the Honam region, though the scale and timeline for such investments remain unconfirmed.
Currently, the memory market remains in a supply deficit, particularly for HBM and server DRAM. With major clients continuing to scale up AI infrastructure investments, there is limited evidence for a sudden contraction in demand in the near term.
However, there is a mismatch between the time required to build a semiconductor fab and the investment cycles of AI companies. While a memory fab takes years to move from groundbreaking to mass production, a client's investment plans can be adjusted at any time based on funding costs and business performance.
Broadcom’s Hock Tan predicted that wafer supply constraints might ease around 2028-2029. While the current market is one where suppliers struggle to meet order volume due to shortages, the point at which supply capacity increases will make client purchasing power and future investment plans even more critical.
The expansion race between Samsung and SK hynix is an investment looking beyond the current supply shortfall toward the growth of the AI market years from now. While current competitiveness is defined by how quickly and how many chips one can supply, the situation may shift once those factories are completed. The pace of the AI industry’s growth is not determined by production capacity alone; it depends on whether customers remain profitable enough to continue purchasing those chips and investing further.
