1. Micron has reopened the AI-infrastructure trade, but the debate has shifted from scarcity to sovereign-scale capital intensity.
Micron raised its planned US investment to more than $250bn through 2035, including an additional $3bn for the domestic semiconductor supply chain and $500m of strategic financing for GlobalWafers alongside a 10-year wafer-supply agreement. The stock rose 4.5% on Thursday and helped drive the SOX 3.1% higher. The bull case is that contracted AI-memory demand and localisation materially improve supply visibility and reduce geopolitical risk; the bear case is that Micron is adding fixed-cost exposure after a 200%+ 2026 share-price rally and at a point when Samsung and SK Hynix are also accelerating capacity. Second-order winners include GlobalWafers, ASML, AMAT, LRCX and KLAC; the longer-term risk is that today’s strategic shortage becomes a late-decade DRAM/HBM overbuild.
2. Meta’s custom chip is broadening the AI capex cycle from merchant GPUs into hyperscaler-owned silicon — positive for equipment, more complicated for Nvidia.
Reports that Meta intends to start producing its in-house “Iris” AI chip in September lifted Applied Materials, Lam Research and KLA by roughly 10–11%, while Lumentum and Vertiv also rallied. The key investor debate is whether custom silicon merely expands the overall compute market or progressively caps Nvidia’s share of inference economics. The likely near-term answer is both: Meta still requires enormous infrastructure capacity, but greater ASIC adoption shifts value toward foundries, wafer-fabrication equipment, networking, optics, memory and power while increasing pressure on merchant accelerator pricing over time. Most exposed: META, NVDA, AVGO, MRVL, TSMC, AMAT, LRCX, KLAC, LITE and VRT.
3. Salesforce has become the clearest test of whether “cheap software” is actually cheap or simply ex-growth.
Salesforce fell 2.5% after KeyBanc downgraded the shares, arguing the low historical multiple is misleading once adjusted for weaker growth and that Agentforce has yet to demonstrate sufficiently broad customer activity or a convincing production use case. This is the central SaaS debate: valuation compression alone does not create upside if AI weakens seat economics, data quality constrains agents and CIO budgets shift toward infrastructure, data and security. The positive read-through is that the market is increasingly separating usage-priced infrastructure software and workflow control planes from traditional seat-heavy applications. Better positioned: DDOG, NOW, PLTR, SNOW and cyber platforms; more exposed to the bear case: CRM, WDAY, HUBS, TEAM and ADBE.
4. Datadog is becoming the software market’s preferred proof that AI can expand, rather than cannibalise, the revenue model.
Datadog rose 3% on Thursday and has gained roughly 95% in 2026, supported by its role in monitoring cloud applications and AI infrastructure, usage-based billing and recent passage through $1bn of quarterly revenue. The investor debate is whether observability becomes an unavoidable tax on increasingly distributed AI systems, or whether hyperscaler-native tooling and open-source alternatives eventually compress pricing. Near term, AI increases model, agent, API, cloud, latency and cost complexity, which should expand telemetry volumes faster than traditional application monitoring. The read-across is positive for DDOG, DT, ESTC and Chronosphere/PANW, and strategically important for cloud platforms that want to own the operational control plane.
5. Rubrik’s
$500m UK commitment reinforces cyber resilience as a separate budget pool from preventive security. Rubrik plans to invest more than $500m in Britain over five years and establish London as its European headquarters, signalling confidence in sustained demand for data security and recovery. The equity debate is whether resilience remains a durable high-growth category as ransomware and AI-generated attacks increase recovery requirements, or whether backup incumbents and security platforms eventually bundle the functionality. The second-order implication is that cyber budgets are fragmenting into prevention, detection, identity, cloud/runtime security and recoverability; that favours vendors with proprietary data, rapid recovery and policy enforcement rather than undifferentiated point products. Most exposed: RBRK, PANW, CRWD, VEEAM’s ecosystem, CVLT and data-security platforms.
