1. Alphabet has proved that AI demand is translating into exceptional cloud growth; it has not yet proved that the resulting capital intensity creates superior shareholder returns.
Google Cloud revenue accelerated 82% yoy to $24.8bn, total revenue reached $119.8bn, and cloud backlog rose to roughly $514bn, providing the strongest evidence yet that enterprise AI infrastructure and inference demand are real rather than speculative. However, Alphabet simultaneously lifted 2026 capex guidance by another $15bn to $195–205bn and reported its first negative quarterly free cash flow, at approximately -$5.9bn. The buy-side debate is therefore no longer whether Google can participate in AI—it clearly can—but whether a business historically valued as a high-margin, asset-light advertising and software platform is becoming a capital-intensive infrastructure utility. Bulls will argue that Google uniquely owns the complete stack—TPUs, cloud, models, search distribution and proprietary data—and that capacity shortages justify spending ahead of demand. Bears will argue that investors are being asked to fund depreciation, power and third-party capacity before the revenue pool and margins are visible. The immediate read-across is positive for NVDA, AVGO, TSMC, MU, ANET and VRT, but the negative share-price response suggests hyperscaler equities will increasingly be judged on incremental ROIC rather than cloud growth alone.
2. Google’s TPU commercialisation changes the semiconductor debate from Nvidia versus custom silicon to merchant accelerators versus hyperscaler-owned full-stack economics.
Alphabet has begun recognising revenue from TPU systems sold into customer data centres, while management indicated that most revenue from existing agreements will arrive after 2026 as deployments ramp. Strategically, this matters more than the near-term contribution: Google is turning an internal cost-optimisation tool into an external product and potentially monetising silicon, systems and cloud software simultaneously. The bull case for Google is that TPU sales deepen customer lock-in and improve utilisation of its R&D and software stack; the bear case is that selling systems outside Google Cloud weakens cloud exclusivity and introduces hardware-style margins, support costs and execution risk. For semiconductors, custom silicon does not necessarily destroy aggregate compute demand, but it redistributes the profit pool away from Nvidia’s scarcity pricing towards Broadcom-style ASIC design, foundries, HBM, networking and packaging. Most exposed: GOOGL, NVDA, AVGO, TSMC, MRVL, AMD, MU and the equipment complex. Alphabet’s results support the view that AI compute demand is broadening, but also that hyperscalers are becoming competitors to their own suppliers.
3. ServiceNow has delivered the cleanest rebuttal yet to the “AI kills SaaS” thesis, although the quality of the beat matters more than the headline numbers.
Q2 subscription revenue rose 24.5% yoy to $3.88bn, total revenue reached almost $4.0bn, cRPO grew 21% to $13.2bn, and ServiceNow raised full-year subscription-revenue guidance to $15.76–15.78bn; shares rose around 4% after hours. AI deployments reportedly increased ninefold and renewal rates remained approximately 98%, supporting the bull argument that a workflow owner can orchestrate agents rather than be displaced by them. Yet part of the outperformance came from US federal on-premise revenue pulled forward from Q3, while the transition towards hybrid seat-plus-consumption pricing may pressure gross margins and reduce predictability. The investor debate should therefore not be framed as “SaaS survives”; it is whether selected incumbents possess enough workflow, data, permission and distribution control to migrate their economics before agents erode user licences. NOW has provided credible evidence of that migration path, with positive read-across for MSFT, PANW, DDOG and SNOW; CRM, WDAY, HUBS, TEAM and ADBE still need to demonstrate equivalent AI-driven incremental revenue rather than merely product adoption.
4. ServiceNow is increasingly becoming a cybersecurity platform by acquisition, widening the competitive boundary for Palo Alto, CrowdStrike and Microsoft.
The company’s integration of Armis and Veza brings asset intelligence across IT, OT, IoT and code together with fine-grained governance for human and non-human identities, while its AI Control Tower aims to govern agents, permissions and connected assets inside the same workflow platform. This is strategically important because agentic AI collapses previously distinct categories: identity, asset discovery, vulnerability management, risk, remediation and workflow orchestration become parts of a single control loop. Bulls will argue that ServiceNow owns the enterprise action layer—security signals become tickets, approvals and automated remediation—allowing it to capture spend beyond traditional IT service management. Bears will point to the $7.75bn Armis acquisition, associated financing and the difficulty of matching the telemetry depth of PANW, CRWD or MSFT. The second-order implication is not necessarily weaker cyber demand, but intensified platform consolidation: NOW becomes more relevant to PANW, CRWD, ZS, CYBR and MSFT, while point vendors in asset visibility, identity governance and risk workflow face rising bundling pressure.
5. The AI equity market is splitting into two standards of proof: suppliers must defend duration, while software and hyperscalers must prove monetisation.
Semiconductor earnings are expected to rise roughly 133% yoy in Q2 and contribute around 44% of total S&P 500 profit growth, yet the sector remains well below its June peak because investors now treat strong 2026 demand as the base case and focus on late-cycle capacity risk. Alphabet’s cloud acceleration shows the spending is generating revenue, but its higher capex and negative free cash flow demonstrate why supplier growth alone cannot settle the ROI debate. Conversely, ServiceNow’s results show that software can outperform when AI adoption translates into backlog, renewals and higher guidance, rather than simply feature announcements. The emerging rotation framework is therefore more nuanced than semis versus software: favour infrastructure vendors with contracted visibility and architecture breadth—TSMC, AVGO and selected networking/equipment names—and software platforms that own control points, usage or compulsory security budgets—NOW, PANW, CRWD, DDOG and SNOW. The most vulnerable equities are those relying either on perpetual AI scarcity without capital-cycle discipline, or on legacy per-seat economics without a credible consumption and agent monetisation path.
