1. The key debate has moved from AI demand to AI profitability.
The latest tape is no longer rewarding “AI exposure” indiscriminately: the S&P 500 and Nasdaq reportedly fell c.2% and 4.6% over the week despite strong Micron news, as investors questioned AI returns, private credit risk, supply chains and rates. The buy-side debate is now whether AI capex is a self-funding productivity cycle or a margin/ROIC drag for hyperscalers. Most exposed: NVDA, AVGO, AMD, MU, MSFT, AMZN, GOOGL, META and ORCL.
2. Qualcomm is becoming the cleanest “new AI infrastructure challenger” story.
Qualcomm used its 24 June Investor Day to target $40bn of non-handset revenue by 2029, including >$15bn from data centres, helped by Meta, Modular and Alphawave. The debate is not whether it displaces Nvidia; it is whether it can become a credible inference/custom-silicon alternative alongside Broadcom and Marvell. The second-order read-across is more competition in AI silicon, more pressure on software ecosystems, and potentially lower long-run AI compute costs.
3. Memory remains the strongest part of the AI semi stack, but scarcity cuts both ways.
Micron’s Anthropic supply agreement reinforces that AI model companies are locking up memory/storage as strategic infrastructure, not commodity input. Bulls see HBM as sold-out, contracted and structurally higher-margin; bears argue that rising memory cost worsens hyperscaler AI economics and eventually invites overcapacity. Exposed: MU, SK Hynix, Samsung, WDC, STX, NVDA, AMD and AVGO.
4. Software is trying to re-enter the debate through rotation, not fundamentals yet.
The “mega rotation” out of crowded mega-cap tech into broader value/cyclicals is important because software may benefit if investors reduce semi concentration. But SaaS still needs proof that AI drives attach, workflow ownership or usage pricing rather than seat compression. Better-positioned software remains security, observability and data platforms; exposed names include PANW, CRWD, ZS, FTNT, DDOG, SNOW, PLTR, NOW, CRM and MDB.
5. Cybersecurity’s relative set-up remains strong, but the market wants platform evidence.
The AI debate is more favourable for cyber than generic SaaS because AI expands attack surface, data movement, identity sprawl and runtime-governance needs. The key debate is who captures this spend: platform vendors with enforcement points and telemetry, or point tools selling AI features. Best exposed: PANW, CRWD, ZS and FTNT; second-order beneficiaries include NET, DDOG and identity/security-data platforms.
