Why Did Korean Chipmakers Trigger a Global Tech Rout?

According to Bloomberg Technology, the selloff began in Seoul on June 23, 2026, when SK Hynix shares dropped 12% and Samsung Electronics fell 8% after analysts from Morgan Stanley and Goldman Sachs downgraded the sector, citing potential oversupply of HBM3e chips. The fear is that hyperscaler demand—which drove a 300% surge in HBM prices over the past 18 months—may plateau as AI model training shifts to more efficient architectures. Reuters confirmed that the selloff spread to U.S. futures within hours, with NVIDIA down 5% pre-market and AMD falling 4.5%. The rout is not a panic—it is a rational repricing of risk after an unprecedented speculative run.

My reading: the catalyst is real but the panic is overblown. HBM oversupply is a near-term inventory correction, not a structural demand collapse. The real story is that investors are finally pricing in the possibility that AI infrastructure spending may not grow at 80% CAGR forever. That is healthy, not apocalyptic.

Which Companies Are Most Exposed to the AI Chip Rout?

To understand the winners and losers, we need a comparison of the key players in the AI chip ecosystem. The table below ranks exposure based on revenue concentration, debt levels, and diversification.

Verdict: SK Hynix is the biggest loser in this rout, while Microsoft emerges as the structural winner due to its diversified revenue and ability to acquire distressed AI assets.

Will the AI Infrastructure Spending Bubble Finally Burst?

Reuters reported that the selloff was exacerbated by a note from Goldman Sachs estimating that hyperscaler capital expenditure on AI infrastructure could reach $1.2 trillion by 2028, but with only a 15% probability of achieving a 20% ROI. This is not a bubble burst—it is a reality check. The AI infrastructure spending cycle is real but lumpy. The key metric to watch is capacity utilization at data centers, which currently sits at 65% for AI-specific clusters, according to a June 2026 report from DCG. If utilization drops below 50%, we will see a genuine oversupply crisis. Until then, this is a correction, not a collapse.

According to DCG, the utilization rate for AI training clusters at major hyperscalers declined from 78% in Q1 2026 to 65% in Q2 2026, driven by a shift from training to inference workloads. That shift explains the HBM oversupply: training requires high-bandwidth memory; inference can use cheaper alternatives. The market is repricing HBM demand downward, but inference demand is rising. The net effect is a rotation, not a rout.

Who Benefits From the AI Chip Selloff?

The beneficiaries of this correction are diversified hyperscalers and AI startups with strong balance sheets. Microsoft, according to Bloomberg, has a $30 billion AI investment fund that it can deploy at lower valuations. Amazon Web Services (AWS) similarly has the cash to acquire distressed AI chip startups like Cerebras or Graphcore, which are now trading at 40% discounts to their 2025 peak valuations. Additionally, companies like Apple, which has low AI revenue exposure (under 10%), are relatively insulated and could use the dip to acquire AI talent and IP at bargain prices.

On the losing side are overleveraged AI startups that raised large rounds at high valuations in 2024–2025. According to PitchBook data cited by Reuters, at least 12 AI chip startups have less than 12 months of runway. If the correction persists for 60 days, expect at least three to file for bankruptcy or be acquired for pennies on the dollar. The winners will be those with diversified revenue, strong cash reserves, or a focus on inference rather than training.

Predictions

1. By September 2026, Microsoft will acquire one distressed AI chip startup (likely Cerebras or Graphcore) for under $2 billion, leveraging the post-rout valuation dip.
2. By December 2026, SK Hynix will announce a 15% reduction in HBM production capacity, citing inventory normalization, which will stabilize prices but not return to 2025 highs.
3. By March 2027, at least three AI chip startups will file for bankruptcy or be acquired for less than their total venture funding, as the correction eliminates overleveraged players.

• March 2024: HBM3e enters mass production, driving SK Hynix stock to record highs.
• December 2025: NVIDIA reports 250% year-over-year data center revenue growth, peak of the AI rally.
• June 23, 2026: Korean chipmaker selloff triggers global tech rout; SK Hynix drops 12%.
• June 24, 2026: Goldman Sachs downgrades AI chip sector, citing oversupply risk.
• Projected September 2026: Microsoft acquires distressed AI chip startup.

Chart: AI Chip Stock Performance, June 2026 (estimated)

• SK Hynix: -12%
• Samsung Electronics: -8%
• NVIDIA: -5%
• AMD: -4.5%
• Microsoft: -1.2%

Source: Bloomberg, Reuters (estimated intraday data as of June 23, 2026).

Article Summary

• The June 2026 tech rout is a rational repricing of HBM oversupply, not a structural collapse of AI demand.
• SK Hynix is the biggest loser; Microsoft is the structural winner due to diversification and cash reserves.
• Hyperscalers will use the correction to acquire distressed AI startups at bargain prices, accelerating consolidation.
• Inference demand is rising even as training demand plateaus, creating a rotation from HBM to cheaper memory solutions.
• Overleveraged AI startups with less than 12 months of runway face imminent failure or acquisition.