1. Alphabet is the first major test of whether hyperscalers can convert AI capex into cash returns rather than merely revenue growth.
Alphabet reports after today’s close with expectations centred on roughly $117bn of revenue, c.$45bn of quarterly capex and Google Cloud growth of around 60%+; the harder issue is that Big Tech’s collective infrastructure spending is beginning to outrun internally generated cash. Reuters estimates that by 2027 the five largest US hyperscalers could require $534bn of incremental capex for only $340bn of additional operating cash flow, with Oracle already producing negative free cash flow and Microsoft recently spending more on capex than it generated in operating cash. The bull case is that cloud AI, inference and advertising productivity eventually create a much larger revenue pool and justify a temporary cash-flow trough. The bear case is that businesses once valued as asset-light software platforms are becoming utilities with rising depreciation, financing and power costs. A strong Alphabet print would support GOOGL, NVDA, AVGO, ANET, VRT, MU and the broader infrastructure chain; weak monetisation or another capex increase without proportional cloud acceleration would pressure the hyperscalers first and then feed backwards into semiconductor estimates.
2. The semiconductor rebound looks like tactical FOMO ahead of earnings, not yet a resolution of the duration debate.
The Philadelphia Semiconductor Index rose 5.2% on Tuesday, led by Sandisk, Western Digital and Micron gains of roughly 12–14%, after entering a technical bear market last week. What changed is positioning rather than fundamentals: investors appear reluctant to remain underweight into Alphabet, Intel and Texas Instruments results after the sharp correction, but the core debate is still whether AI demand can compound long enough to justify today’s capacity build. Bulls see contracted memory demand, leading-edge scarcity and strong cloud growth as evidence that the sell-off overshot; bears argue that a two-day rebound does not resolve synchronised investment across foundries, HBM, storage and equipment. The cleanest relative exposures remain TSMC and Broadcom, which benefit across GPU and custom-silicon architectures, while Micron, Western Digital, Sandisk and the equipment names retain more cycle and pricing sensitivity.
3. ServiceNow’s result tonight is the most important software referendum of the quarter: can an incumbent disrupt its own seat-based model before AI does it externally?
Consensus expects revenue of about $3.9bn, up roughly 22% yoy, but investors will focus on current backlog, renewal rates and whether AI-consumption revenue is becoming material enough to offset pressure on user-based pricing. ServiceNow is deliberately repositioning itself as an orchestration layer for enterprise agents and has moved towards a hybrid model combining seats with AI usage, but that transition may lower gross margin and complicate revenue visibility. The bull case is that NOW owns the workflow, permissions and data context required to coordinate agents across large enterprises; the bear case is that customers use agents to reduce human licences before ServiceNow captures equivalent consumption revenue. Its $7.8bn Armis acquisition also turns the print into an early test of whether cyber and workflow integration deepens the platform or simply raises leverage and execution risk. Read-across will be broad: positive for NOW, PANW, CRWD, DDOG and SNOW if AI orchestration drives incremental platform spend; negative for CRM, WDAY, ADBE, HUBS and TEAM if the market concludes that workflow automation accelerates seat compression.
4. The software debate is becoming “moat plus migration path”, rather than simply cheap versus expensive.
Morgan Stanley argues that sentiment has become excessively negative after the software index fell more than 25% from its 2025 highs and IGV declined around 13% in 2026, but the emerging framework is highly selective. Investors need both an embedded control point today and a credible path into AI-native pricing tomorrow. That favours Microsoft, Palo Alto, CrowdStrike, ServiceNow, Snowflake and Datadog, where identity, security, data, observability or workflow ownership gives the vendor a reason to remain central as agents proliferate. Adobe and Workday are more exposed because AI can automate content creation or administrative workflows before those vendors demonstrate equivalent incremental monetisation. The second-order implication is that the historic SaaS valuation framework—growth plus margin under a predictable per-seat model—may not return uniformly; usage-based infrastructure and control-plane software could command a structurally higher multiple, while application SaaS remains valued on proof rather than installed-base quality alone.
5. Cybersecurity remains the strongest relative software category, but consolidation is now as important as demand growth.
AI is increasing attack speed, non-human identities, tool access and the number of autonomous applications that require policy and monitoring, while CrowdStrike’s threat research reports an 89% rise in attacks by AI-enabled adversaries. The demand argument is therefore becoming less controversial; the investor debate is whether incremental spending accrues to broad platforms or is diluted across another generation of point products. The likely winners are vendors controlling endpoint, identity, network, cloud and SOC telemetry—PANW, CRWD, ZS, OKTA, CYBR, FTNT and MSFT—because they can bundle AI-agent governance into existing enforcement layers. The bear case is valuation: PANW and CRWD already discount substantial platform consolidation, meaning the next leg requires explicit ARR attach, renewal uplift and AI-security revenue rather than threat commentary alone. Exposure-management and recovery names such as TENB, QLYS, RBRK and CVLT remain second-order beneficiaries, but with less certainty that they capture the primary control-plane budget.
