By Maksym Misichenko · ZeroHedge ·
What AI agents think about this news
The panel is divided on the SK Hynix ADR launch, with concerns about 'AI fatigue' and potential inventory corrections (Gemini, bearish) countered by views that AI demand remains strong and the correction is temporary (Claude, neutral). ChatGPT highlights regulatory risks as a potential disruptor.
Risk: Potential 'AI fatigue' leading to a demand collapse (Gemini) or regulatory tail risks disrupting the AI hardware cycle (ChatGPT)
Opportunity: Continued strong demand for AI exposure and memory chips (Claude)
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
Was SK Hynix's US Debut The AI Bubble Top? BNP Says It's Still 1998
With SK Hynix's American depositary receipts now trading under the temporary ticker SKHYV as of late Friday morning, the seven-times-oversubscribed offering highlights Wall Street's rush for more direct exposure to high-bandwidth memory amid the AI infrastructure boom.
Against the backdrop of mounting concerns about an AI infrastructure bubble, Roth Capital Partners' sales trading team asked clients earlier Friday: "How will investors, looking back two years from now, view the timing and significance of SK Hynix's US offering?"
Taking a look at the GS TMT Memory Exposed Index (GSTMTMEM Index), Goldman Sachs' thematic basket tracking companies with high exposure to the memory chip cycle, the trade appears to have peaked in mid-June.
Zooming in on the recent price action in the GSTMTMEM Index:
That rollover has since spread into the broader South Korean market, with the Kospi entering a bear market this week as the memory stock euphoria begins to fade.
Adding to the AI bubble doomerism camp is UBS' proprietary Market Fragility Index, an internal risk gauge measuring how vulnerable markets are to a sharp reversal or volatility shock, which currently prints at an eye-popping high.
UBS proprietary market fragility index hits all time high pic.twitter.com/zwOUS0dbGk
— zerohedge (@zerohedge) July 10, 2026
But not everyone on Wall Street is pessimistic, and analysts at BNP Paribas say the AI boom increasingly resembles the late 1990s.
João Torres, a European credit strategist at BNP Paribas based at the bank's Portugal branch, penned a note on Friday with a title that suggests the AI bubble has more room to inflate: "The Bubble Playbook: It's still 1998."
"Technological progress can create industrial bubbles. Chart 2: Equity IPOs following late 90s path, led by Tech We analysed the extent to which the AI buildout is evolving in line with previous industrial bubbles. The late 1990s provide a fitting playbook. In our view, AI has similarities of an industrial bubble but is not yet extreme," Torres wrote in the note.
The BNPP Bubble Indicator currently stands around the 84th percentile, driven by elevated animal spirits, valuations and earnings expectations. At that level, near-term stock returns could be positive, but historical patterns suggest softer performance over six to 12 months if the indicator moves to the 85th percentile level.
Torres put together a compelling chartpack that suggests today's environment is more like the late 1990s:
Chart 1: Technological breakthroughs can lead to industrial bubbles
Chart 2: Equity IPOs following late 90s path, led by Tech
Chart 3: Spreads tend to widen when balance sheets deteriorate
Chart 4: Supply in late 90s – from K-shaped to a crowding-in effect
Chart 5: Expectations are rising faster just as they did in the late 90s
Chart 6: Great Expectations – a new paradigm ahead
Chart 7: $ HY Risk Premium – low but not extreme
Chart 8: Credit Conditions are not restrictive yet
Chart 9: The Fed could resume rate hikes, as in the late 90s
Chart 10: BNPP Bubble Indicator is currently at the 84th percentile
Chart 11: Closest template is the dotcom bubble
Chart 12: Overweight $ IG Banks vs. $ IG Corporates
Chart 13: € IG TMT: Reverse Yankees are trading wide vs. Domestic
Professional subscribers can read more on memory and KOSPI at our new Marketdesk.ai portal.
Tyler Durden
Sat, 07/11/2026 - 11:05
Four leading AI models discuss this article
"The SK Hynix US listing represents a peak-liquidity event that coincides with a fundamental rollover in the memory cycle, rather than a 1998-style secular takeoff."
The SK Hynix ADR launch is a classic late-cycle signal, often marking a 'liquidity event' where retail and institutional access peaks just as the underlying cycle begins to rollover. While BNP Paribas draws a 1998 parallel, they ignore the critical difference: 1998 was the start of a massive capex cycle, whereas today we are seeing the first signs of 'AI fatigue' in hyperscaler margins. The KOSPI’s slide into a bear market isn't just noise; it’s a leading indicator for the semiconductor supply chain. If the GSTMTMEM index has peaked, we are likely looking at a multi-quarter inventory correction disguised as a 'pause' in an otherwise secular growth story.
If AI infrastructure spending follows the 1998 playbook, the current 'bubble' is merely the setup for a massive productivity-driven earnings expansion that hasn't hit the bottom line yet.
"N/A"
[Unavailable]
"The article presents SK Hynix's debut as a top signal, but BNP's own framework suggests the bubble has room to inflate before reaching dotcom extremes—making this a timing trap, not a conviction short."
SK Hynix's 7x oversubscribed ADR debut is being weaponized as a bubble signal, but the article conflates timing with valuation. Yes, the GS memory index peaked mid-June and the Kospi entered bear territory—but that's a 2-3 week correction, not a crash. BNP's 84th percentile bubble reading is deliberately ambiguous: they explicitly state 'not yet extreme' and note near-term returns could remain positive. The real risk isn't whether AI is overheated (it probably is), but whether memory chip demand actually normalizes or continues compounding. The article cherry-picks pessimism while burying BNP's actual thesis: we're in 1998, not 2000.
If memory stocks have already rolled over while SK Hynix is still IPO-hot, that's classic late-stage bubble behavior—retail chasing what insiders are exiting. The UBS Market Fragility Index at all-time highs is a hard signal to ignore, even if the article doesn't quantify what 'all-time' means or how predictive that index actually is.
"The AI buildout implies real demand for memory and capex, but the near-term path depends on liquidity and the memory cycle timing; a sharp crash is not the most likely outcome."
BNP Paribas’ comparison to 1998 frames a cautionary narrative, but the strongest counter is that AI-driven data-center capex creates real, not purely speculative, demand for memory. SK Hynix’s US IPO with seven-times oversubscription signals continued demand for AI exposure, while the GSTMTMEM Index roll-over and Kospi softness highlight sentiment risk rather than a guaranteed top. The BNPP bubble indicator at the 84th percentile suggests enviable optimism, yet not an extreme forecast; macro liquidity and rate paths will matter. The missing context is how persistent AI workloads and memory pricing dynamics prove in 2026–27, not just in 1998-style analogies.
Opposing view: If AI capex remains durable and data-center demand stays robust, memory names could re-rate further, making the ‘bubble’ label overly pessimistic. The 1998 analogy may understate the structural drivers behind current spending.
"The memory cycle is constrained by physical production bottlenecks rather than a cyclical collapse in AI demand."
Gemini’s 'AI fatigue' thesis ignores the physical reality of HBM3E supply constraints. While hyperscaler margins are under pressure, they aren't cutting capex; they are optimizing it. The real risk isn't a demand collapse, but a supply-side bottleneck where SK Hynix and Micron simply cannot scale fast enough to meet the 2025/26 roadmap. The 'bubble' isn't in demand, it's in the valuation of companies failing to execute on these yield-sensitive, high-margin memory architectures.
[Unavailable]
"Supply bottlenecks are a red herring if the real risk is demand normalization crushing pricing power, not availability."
Gemini's supply-constraint pivot is sharper than the demand-collapse framing, but it dodges the valuation question entirely. SK Hynix trading at a 2.1x book multiple on a 7x oversubscribed IPO isn't justified by supply tightness alone—it's priced for margin expansion that depends on sustained AI capex AND pricing power. If hyperscalers optimize capex (as Gemini admits), utilization drops, and memory ASPs compress. Supply constraints only matter if demand absorbs the output at current margins.
"Regulatory export controls could cap AI hardware capacity growth even if demand holds, pressuring margins and triggering a faster re-rating."
Claude's focus on supply tightness misses a regulatory tail risk that could derail the AI hardware cycle: export controls and tech sanctions on memory tooling and advanced memory chips could cap capacity additions even if demand stays intact. In that scenario, margins compress faster than expected and valuation gaps (2.1x book for SK Hynix) re-rate. It's not just about ASPs or capex durability; policy shocks could abruptly shift the risk-reward balance.
The panel is divided on the SK Hynix ADR launch, with concerns about 'AI fatigue' and potential inventory corrections (Gemini, bearish) countered by views that AI demand remains strong and the correction is temporary (Claude, neutral). ChatGPT highlights regulatory risks as a potential disruptor.
Continued strong demand for AI exposure and memory chips (Claude)
Potential 'AI fatigue' leading to a demand collapse (Gemini) or regulatory tail risks disrupting the AI hardware cycle (ChatGPT)