KOSPI Plunges 10% in Global Tech Rout, Strong Dollar Batters EM Currencies, and VCs Pour Millions Into AI Safety

The week of June 23–24, 2026 delivered three simultaneous signals from different layers of the global capital stack. The KOSPI's 10% single-session crash — with circuit breakers halted trading for the first time in years — is the most severe single-day expression of "AI monetisation anxiety" yet: Samsung and SK Hynix (together >50% of KOSPI market cap) fell >12% each on rumours that US hyperscaler capex schedules are slowing, threatening HBM demand. The US Dollar Index (DXY) hitting a 13-month high (driven by the Fed's hawkish June dot plot with 9/18 members projecting another 25bp hike) has simultaneously battered the Korean Won, Japanese Yen, and Indian Rupee — forcing the RBI to intervene directly through state-run banks, and raising the spectre of 2013-style EM currency crises for economies with high energy import bills. And three Series A rounds in AI safety (Coval, $28M), developer infrastructure (Runlayer, $30M), and AI tooling (JustAI, $17M) demonstrate that early-stage VC remains open for high-conviction infrastructure bets even as late-stage mega-rounds dominate by dollar volume.

---

📉 KOSPI Circuit Breaker — The Anatomy of a 10% Crash

Why South Korea Was the Epicentre

KOSPI concentration in semiconductors:

When a stock with 40% of the index falls 12%, the mathematical minimum impact is −4.8% on the index — the remaining 60% selling off another 5.2% from contagion completed the 10% move.

What triggered the sell-off — the hyperscaler capex rumour: The catalyst was an unverified report (attributed to a Morgan Stanley analyst note, later walked back) that:

• Microsoft was delaying $3–5B in data centre equipment orders due to slower-than-expected Azure AI revenue
• Amazon AWS was revising capex timelines for Trainium 2 chip deployments in Bedrock
• These delays would reduce HBM orders from Samsung and SK Hynix by 15–20% in H2 2026

HBM demand sensitivity — why the rumour hit so hard:

HBM (High Bandwidth Memory) is the critical component in AI accelerators: NVIDIA H100 uses 6 HBM3 stacks; H200 uses 8 HBM3E stacks; B200 uses 8 HBM3E stacks. Samsung and SK Hynix are the only two volume HBM suppliers globally (Micron is entering the market). Any reduction in hyperscaler GPU orders → proportional HBM demand drop → hits Samsung and SK Hynix revenue directly.

Circuit breaker mechanics — the halt: South Korea's circuit breaker triggered at −8% in the KOSPI on the morning of June 23:

• Level 1: Index −8% → trading halted for 20 minutes (first trigger)
• Level 2: Index −15% at reopening → trading halted for 20 minutes again
• Level 3: Index −20% → market closed for the day
• The KOSPI hit Level 1 but recovered to −10% by close without triggering Level 2

The rebound (June 24) — why +3.3%:

• Bargain hunters and index-tracking funds (ETFs required to hold KOSPI weights) stepped in
• Morgan Stanley walked back the capex delay report as "unverified"
• South Korean retail investors (known as "KOSPI retail warriors") bought the dip aggressively
• Net foreign buying turned positive by mid-session (from heavy selling at open)

What to watch — HBM demand confirmation: The KOSPI recovery is fragile until HBM demand is confirmed by:

• Micron Q3 FY2026 earnings (July 2026): HBM guidance
• Samsung Q2 2026 results: HBM3E volume and pricing
• NVIDIA Q2 FY2027 (August): B200 Blackwell shipment volume update

---

💵 DXY at 13-Month High — The EM Currency Storm

Dollar Strength: Mechanics and Regional Impact

DXY composition and June 24 level:

Asian EM currencies — the contagion:

India's specific challenge — the INR at ₹95.8: India's rupee is under pressure from three simultaneous forces:

1. FII equity outflows: Foreign Institutional Investors have sold ₹42,000 crore (~$5B) in Indian equities YTD 2026, as dollar strength makes USD-denominated assets more attractive
2. Energy import bill: India imports ~85% of oil. With Brent at ~$85/bbl (up from $72 at year-start), India's oil import bill has risen by ~$30B annualised
3. DXY strength: As global reserve currency, USD strength mechanically weakens all EM currencies simultaneously

RBI's FX intervention mechanics: The RBI does not directly buy rupees in the market (it doesn't hold rupee foreign reserves). Instead:

• The RBI instructs state-owned banks (SBI, BOB, Canara Bank) to sell USD from their own foreign exchange holdings
• This increases USD supply in the forex market → reduces USD/INR rate (strengthens rupee)
• The RBI simultaneously provides USD liquidity to state banks via swap lines so they can rebuild their foreign exchange positions
• Net effect: Uses India's $671B forex reserves as backstop

The "2013 taper tantrum" comparison: The current dollar surge (DXY +5.3% YTD) is less severe than the 2013 taper tantrum (DXY +8.2%), but:

• India's current account deficit is smaller now (~1.8% GDP vs 4.8% in 2013)
• Forex reserves are much higher ($671B now vs $270B in 2013)
• This means RBI can absorb the shock better — but persistent dollar strength remains a challenge

---

🛡️ AI Safety and Developer Infrastructure — The Early-Stage VC Resilience

Three Series A Rounds in Critical AI Infrastructure

Overview of the three rounds:

Coval — AI voice safety (the regulatory nexus): Enterprise AI deployments of voice agents (call centres, customer service, financial advice bots) face:

• CFPB (Consumer Financial Protection Bureau) regulations on AI-generated financial advice
• HIPAA requirements for AI in healthcare communication
• EU AI Act requirements for real-time human oversight of "high-risk" AI systems
• FTC guidelines on disclosure requirements for AI agents

Coval's platform monitors AI voice agent conversations in real time, flagging compliance violations, hallucinations, and safety failures. As AI call centres scale (Klarna disclosed its AI agent replaced 700 human agents in 2024), the compliance monitoring market grows proportionally.

Why Norwest led at $28M Series A: Norwest Venture Partners (a Wells Fargo-affiliated VC managing $10B+ AUM) focuses on enterprise software with strong regulatory tailwinds. The rationale:

• Regulatory moat: Companies that embed into enterprise compliance workflows create switching costs comparable to payroll providers
• TAM: AI monitoring/observability market estimated $8.4B by 2028 (Gartner)
• Network effects: As Coval processes more voice agent interactions, its safety models improve → better product → more enterprise clients → more data

Runlayer — AI deployment infrastructure: Runlayer abstracts the complexity of deploying AI models in production environments:

• Problem it solves: Deploying an AI model from prototype (Jupyter notebook) to production-grade API requires DevOps expertise most ML teams lack (versioning, rollback, A/B testing, observability, autoscaling)
• Competitors: Replicate, Modal, Banana (niche), AWS SageMaker (complex), Azure ML (complex)
• Differentiation: One-click production deployment for any Hugging Face model with built-in monitoring and cost tracking

The Series A market context — why these rounds closed:

Series A is the one funding stage that is growing (by deal count) — because:

1. Seed companies from 2023–2025 AI boom are now proving revenue and seeking Series A
2. Series A investors see opportunity to buy proven teams at compressed valuations before late-stage mega-rounds reprice the sector upward

---

📌 The Bottom Line

• kospi-10pct-circuit-breaker-samsung-hynix-hbm-ai-overcapacity: Samsung + SK Hynix = 40% KOSPI weight → fell 12%+ each = minimum −4.8pp mathematical contribution; catalyst: unverified Morgan Stanley note (MS later walked back) alleging Microsoft/Amazon delaying HBM orders 15-20%; HBM uniquely sensitive: only 2 volume suppliers (Samsung + SK Hynix); circuit breaker triggered at −8% (Level 1); recovered next day +3.3% on Morgan Stanley retraction + retail "KOSPI warrior" buying; HBM demand confirmation watch: Micron/Samsung Q2/NVIDIA Blackwell shipments; HBM +120% YoY vs standard DRAM +15%.
• dxy-13month-high-krw-inr-jpy-em-rbi-fx-intervention: DXY 105.2 (+5.3% YTD); Fed hawkish dot plot (9/18 projecting another hike) → USD magnetic; KRW −5.8%, JPY −6.2%, INR −4.1%, IDR −6.1% simultaneous EM pressure; India triple pressure: FII outflows ₹42,000 crore + oil import bill +$30B annualised + DXY mechanical; RBI intervention: state banks (SBI/BOB/Canara) sell USD directly; reserves $671B backstop (vs $270B 2013 taper tantrum); India current account deficit 1.8% now (vs 4.8% in 2013) = much better positioned to absorb.
• coval-28m-runlayer-30m-justai-17m-ai-safety-developer-infra: Series A = only funding stage growing by deal count (+3% YoY); Coval: AI voice agent real-time compliance monitoring, CFPB/HIPAA/EU AI Act tailwinds, regulatory moat = switching cost comparable to payroll providers, $8.4B TAM by 2028; Runlayer: one-click production deployment for HuggingFace models (prototype→production DevOps abstraction), competing with Replicate/Modal/AWS SageMaker on simplicity; JustAI: AI developer tooling; all three targeting B2B enterprise infrastructure vs consumer AI (which is compressed).

---

📬 Stay Updated

Get global market analysis delivered to your inbox every week. Subscribe to our free newsletter →

---

Disclosure: This post contains affiliate links. If you purchase through our links, we earn a small commission at no extra cost to you. We only recommend products we believe in.