1. AI memory is now correcting despite record profits — the market is pricing peak scarcity risk.
Samsung and SK Hynix sold off again in Seoul, with Samsung down as much as 7.6% and SK Hynix down 5.2%, even after Samsung’s Q2 profit estimate showed a 19x jump. The debate is no longer demand; it is whether memory pricing, hyperscaler appetite and AI capex ROI can survive higher component costs and rising 2027–30 capacity. Exposed: Samsung, SK Hynix, MU, ASML, AMAT, LRCX, KLAC and NVDA.
2. SK Hynix’s planned
$28bn Nasdaq ADR is becoming a referendum on AI-chip equity appetite. The listing gives global investors cleaner access to the HBM winner, but it also arrives exactly as investors question whether memory stocks have moved from shortage to overbuild. Strong demand would reopen the AI scarcity trade; weak demand would reinforce the “peak FOMO” bear case.
3. Hyperscaler funding is now the second-order AI risk.
Amazon is raising $25bn of bonds, with demand reportedly peaking at $62bn, while BofA now sees Alphabet capex at $195bn in 2026 and $290bn in 2027, Meta at $145bn and $185bn, and Amazon 2027 capex at $230bn. This is the key investor debate: AI capex is moving from cash-funded optionality to debt-funded infrastructure buildout. Positive for NVDA, AVGO, MU, power and networking suppliers; more scrutiny for AMZN, GOOGL, META, MSFT and ORCL.
4. Software is stabilising, but this is still “AI Armageddon is overdone”, not proof of AI monetisation.
ServiceNow rose 2.6% on 7 July, its fourth straight gain, despite a weak market, while Salesforce also gained. The debate is whether workflow software has been de-rated too far, or whether investors are simply covering shorts before evidence of seat compression, usage pricing and agent attach becomes clearer. Better positioned: NOW, PLTR, DDOG, SNOW, PANW and CRWD; more debated: CRM, ADBE, TEAM, HUBS and WDAY.
5. Cybersecurity remains the cleanest software AI story, but valuation risk is rising.
The strongest debate is identity and agentic AI: autonomous agents create non-human users with permissions, lateral-movement risk and data access, which supports identity, endpoint, cloud and platform security. That favours PANW, CRWD, OKTA, ZS, CYBR, SAIL and FTNT, but the next earnings season needs hard proof: ARR attach, renewal expansion, AI-security monetisation and platformisation KPIs, not just “AI tailwind” language.
