Broadcom’s AI chip business is still growing fast, but its latest guidance sent a different signal to South Korea’s memory giants.
The company’s latest results showed strong demand for custom AI accelerators and networking chips, but its outlook fell short of Wall Street’s expectations. The reaction spread quickly to South Korea, where Samsung Electronics and SK Hynix fell as investors questioned how much AI upside was already priced into Korean memory suppliers.
Why strong AI revenue wasn’t enough
Broadcom reported $22.19 billion in fiscal second-quarter revenue, while AI semiconductor revenue rose 143% year over year to $10.8 billion, according to Investor’s Business Daily. But its $16 billion fiscal third-quarter AI chip sales forecast missed Wall Street’s $17.2 billion estimate, shifting attention from raw demand to the capacity and margin questions facing AI chip suppliers.
MarketWatch reported that Broadcom fell 12.6% on June 4, wiping $286 billion from its market value.
For Samsung and SK Hynix, the supply-chain question starts with high-bandwidth memory, a key component in AI accelerator systems. Broadcom’s custom chips, or ASICs, are designed for workloads such as inference and still require advanced memory. That hardware pressure is also reaching PCs, as RTX Spark systems show local AI workloads moving onto developer machines.
Investors treated Broadcom’s guidance as a signal about expectations, not demand alone.
The impact was clearest in South Korea, where The Wall Street Journal reported that the KOSPI closed 5.5% lower on June 5 after Broadcom’s guidance hit AI-linked chip sentiment. Samsung and SK Hynix make up about half of the KOSPI’s market capitalization, turning a global AI-chip reset into a Korea-market story.
Broadcom did not disprove HBM demand. It showed that AI suppliers now need clearer proof of margins and capacity.
What the Samsung and SK Hynix selloff signals
Samsung and SK Hynix still need to balance HBM supply across Nvidia-linked GPU systems, Broadcom-linked custom accelerators, and other AI customers. Tight capacity can still create shipment delays, customer allocation disputes, or pressure to spend heavily on new production.
SK Group Chairman Chey Tae-won told reporters at Computex in Taipei that SK Hynix plans to double memory wafer capacity within five years and expects AI-driven memory tightness to last until at least 2030, Tom’s Hardware reported. That keeps capacity decisions central to AI infrastructure planning.
Broadcom’s AI chip growth also depends on cloud and technology customers continuing to build data center infrastructure. Enterprise buyers are testing cheaper models such as DeepSeek as AI costs rise, putting more scrutiny on infrastructure budgets. If major AI buyers slow spending or push suppliers harder on pricing, Korean memory suppliers could feel pressure before the long-term demand story changes.
Enterprise IT teams planning AI infrastructure still face volatile supply chains. Similar issues are showing up in AI agent devices for the enterprise, where hardware, cloud services, identity controls, and endpoint management increasingly overlap. Accelerator availability, memory pricing, and data center timelines can shift as chipmakers adjust forecasts.
AI infrastructure demand has not cooled, but Broadcom’s selloff showed how quickly questions about capacity, margins, and delivery timelines can move through Korea’s memory supply chain. For teams planning AI infrastructure, that can affect hardware availability, memory pricing, and data center timelines long before demand actually slows.
Also read: Microsoft’s AI-powered Windows developer tools show how local models, agent controls, and AI-ready hardware are moving deeper into enterprise workflows.