Samsung and SK Hynix shares fell by more than 10% on July 28, 2026, as the AI chip sell-off accelerated across markets, according to The Guardian. That single-day slide in two of the world’s key memory suppliers forced a blunt question back onto trading desks: how durable is the AI capex story when growth rates cool and rivals multiply?
The pullback was not confined to Seoul. In late July 2026, the BBC reported chip stocks sliding in both US and Asian sessions as investors reassessed the pace and payback of AI spending. The reset followed days of choppy trading for big tech, and it arrived just as markets were preparing for another round of earnings guidance from hyperscalers.
Why the AI chip sell-off intensified
The Guardian tied the rout to two forces: renewed fears over AI spending and rising Chinese competition. Those drivers cut to the core of how today’s AI stack makes money. Memory suppliers such as Samsung and SK Hynix have ridden a surge in demand for high-bandwidth memory, the ultra-fast DRAM that feeds modern accelerators. As production scales, pricing power can waver, and that’s where investors get nervous.
HBM has been the quiet star of the AI boom, shuttling data at rates earlier designs couldn’t touch. It’s also where margins can swing the most when supply catches up. Background on the technology helps explain the sensitivity: High Bandwidth Memory stacks multiple DRAM dies to widen the data path, which boosts throughput but complicates manufacturing. When yields improve and more capacity comes online, average selling prices face pressure. Traders price that risk fast.
On the demand side, the market is recalibrating to slower year-on-year capex growth from cloud providers after a blistering 2024–2025. The BBC’s late-July coverage of broad tech weakness underscored the same theme: investors want clearer proof that AI workloads are scaling from headline demos to durable, revenue-driving products. When that proof lags, multiples compress.
How the semiconductor rout meets Chinese competition
The Guardian flagged Chinese competition as a second spark, which matters for both pricing and share. It’s not only about rival accelerators. Memory, packaging, and supporting components see the same substitution push. One day earlier, the BBC highlighted a Chinese chipmaker whose shares surged nearly 470% on debut, a sign of local investor appetite and policy tailwinds. Domestic capital chasing domestic fabs tightens the vise on incumbents abroad.
Trade policy sets the stage. US export controls have tried to throttle the highest-end compute headed to China, yet supply chains reroute and design workarounds appear. For context on the rules that shape those choices, the US Bureau of Industry and Security maintains guidance on semiconductor export controls. When controls move, order books and discounting can move with them. That feeds back into valuation swings for listed chip firms.
This is why July’s moves felt sharper than a standard growth scare. Margins at the heart of AI systems—HBM and advanced packaging—are uniquely exposed to both capacity ramps and geopolitical friction. When the two hit at once, the tape can look brutal.
What the AI chip sell-off signals for earnings season
Volatile days often fade, but the questions stick. The next set of earnings updates from hyperscalers and chip suppliers will carry outsized weight. Three checkpoints stand out.
- Cloud capex guidance: Are 2026 outlays tracking up, flat, or down versus prior plans? Investors will look for not just totals but mix—how much goes to inference versus training, and how fast on-prem customers are pulling forward orders.
- HBM memory pricing: Any shift in average selling prices or contract duration will be parsed line by line. Even small downticks can ripple through earnings models for Samsung and SK Hynix.
- Inventory and order visibility: Watch for language on cancellations, pushouts, or tighter allocation. A move from waitlists to spot availability would signal a new phase of the cycle.
The BBC’s late-July programming asked a blunt question—whether AI is burning cash or building moats. That debate shows up in these checkpoints. If cloud providers can tie model rollouts to measurable user growth or lower unit costs, the market may forgive a slower capex slope. If not, the AI chip sell-off can linger.
Pricing power, product mix, and who’s most exposed
Memory makers carry the most direct HBM risk today, but they are not alone. Advanced packaging houses sit one layer up the stack. If back-end constraints ease, GPUs ship faster, which helps integrators but can soften the scarcity premium that lifted component prices in 2025. Foundries further upstream balance leading-edge wafer supply against customer roadmaps with more moving parts than usual.
Downstream, device makers tied to consumer cycles face a different squeeze. If AI features fail to command price premiums in phones or PCs, the expected trickle-down boost to unit demand can disappoint. That leaves the heaviest spenders—large cloud platforms—carrying more of the narrative burden for near-term growth. A clear map from model training to user adoption would calm nerves fast.
One underappreciated wild card is how quickly enterprises outside big tech convert pilots into production. If deployment timelines slip, orders that looked locked for late 2026 could drift into 2027. To gauge that risk, listen for commentary on migration from proof-of-concept to paid workloads, and on toolchains that shrink the gap. A broad move to simpler, on-prem deployments would shift the mix of winners again.
Policy and supply chains: what could steady the market
Policy clarity helps. Transparent timelines for export rules remove one source of guesswork for suppliers and buyers. So does forward guidance from hyperscalers about capacity expansion. For readers tracking the plumbing behind those disclosures, background on hyperscale buildouts explains why even small shifts in data center design can swing multi-billion-dollar orders.
Supply chain diversity matters as much as capacity. When a few vendors dominate HBM, every yield hiccup and every price negotiation looms large. New entrants can soften that edge, but they also compress margins. That’s the trade-off investors weighed in late July 2026, as headlines from The Guardian and the BBC landed within hours of each other.
There’s a path to calmer trading. Solid beats from cloud platforms, firmer HBM pricing, and signs of enterprise adoption would flip the mood. If those prints arrive over the next two quarters, the AI chip sell-off will read like a stress test rather than a turning point. If they don’t, expect more days where one line about spending plans wipes billions off semiconductor market caps.
For now, one takeaway stands up. The AI buildout is still real, but the market is done paying up for stories alone. It wants receipts. Whether Samsung, SK Hynix, and their peers provide them will decide where this tape goes next. For more on this, see nytimes.com.
Related reading: AI in Education • Data Privacy • AI in Society