1. Intel has delivered the first genuinely credible turnaround print, but the equity debate now shifts from revenue recovery to whether foundry economics can ever justify the capital burden.
Q2 revenue rose 25% yoy to $16.1bn, materially ahead of expectations, while Data Center and AI revenue increased 59% to $6.3bn and adjusted EPS of $0.42 was almost twice consensus. Intel also guided Q3 revenue to $15.8–16.8bn and raised 2026 capex above $20bn. What changed is that Intel is no longer relying solely on restructuring or sovereign support: CPU and AI-compute demand are visibly reaccelerating. The bear case, however, remains concentrated in foundry, which generated a roughly $2.1bn operating loss and still lacks sufficient external scale to validate the TSMC challenger thesis. Bulls will argue that stronger internal volumes, edge AI and ASIC opportunities provide the utilisation bridge towards external customers; bears will argue that higher capex merely deepens the cash commitment before process leadership and customer wins are proven. Positive read-across extends to ASML, AMAT, LRCX and KLAC, while the competitive implications are more mixed for AMD and TSMC.
2. Alphabet’s post-results sell-off has crystallised the hyperscaler problem: spectacular AI revenue growth is no longer sufficient when free cash flow and capital efficiency deteriorate simultaneously.
Google Cloud grew 82% yoy to $24.8bn, yet Alphabet raised 2026 capex guidance to $195–205bn, reported its first negative quarterly free cash flow and fell sharply as investors focused on the widening gap between infrastructure investment and realised cash returns. Thursday’s broader Nasdaq decline of more than 2%, accompanied by rising bond yields and Brent crude above $100/bbl, compounds the issue by increasing both data-centre operating costs and the discount rate applied to long-duration AI investments. The bull case is that current capacity remains constrained and Alphabet owns the complete stack—TPUs, models, cloud, search distribution and proprietary data—so near-term cash compression represents investment ahead of contracted demand. The bear case is that hyperscalers are becoming capital-intensive utilities whose returns increasingly accrue to chip, memory, networking and power suppliers. Most exposed: GOOGL, MSFT, AMZN, META and ORCL on ROIC; NVDA, AVGO, MU, ANET and VRT on the durability of the spending cycle.
3. SAP and ServiceNow have both rebutted the simplistic “AI kills enterprise software” thesis, but they also show that defending the franchise requires materially higher investment and a different economic model.
SAP’s Q2 cloud revenue rose 24% at constant currencies to €6.28bn, current cloud backlog increased 26% to €22.93bn, and Cloud ERP Suite revenue grew 27%, demonstrating that mission-critical systems of record retain strong migration and renewal momentum. ServiceNow similarly delivered 24% revenue growth, 21% cRPO growth to $13.2bn, a 98% renewal rate and more than $1bn of AI annual contract value. Yet SAP trimmed its operating-profit outlook by roughly €100m after AI-data acquisitions, while ServiceNow is absorbing the cost and leverage associated with the $7.8bn Armis acquisition and moving towards hybrid seat-plus-consumption pricing. The investor debate is therefore not survival, but migration economics: workflow and ERP incumbents can remain central if they own enterprise data, permissions and execution, although AI may reduce gross-margin visibility and weaken the simplicity of the old per-seat SaaS model. Positive read-across: NOW, SAP, MSFT and selected workflow platforms; unresolved battlegrounds: CRM, WDAY, ADBE, HUBS and TEAM.
4. OpenAI’s reported “Presence” launch has reopened the software-disintermediation debate just as ServiceNow appeared to have settled it.
The new platform is reportedly designed to deploy agents across enterprise functions such as customer service and IT requests, directly overlapping with workflows historically controlled by ServiceNow and other application vendors. ServiceNow initially rallied following its earnings beat but subsequently reversed as investors considered whether frontier-model providers could move above the model layer and become the primary enterprise interface. The bull case for incumbents is that agents still require trusted systems of record, permissioning, auditability, integration and deterministic workflow execution; OpenAI may own the conversational layer without replacing the underlying transaction and governance platforms. The bear case is that the agent becomes the user interface, reducing application engagement and eventually weakening seat-based pricing and vendor-specific workflow differentiation. The second-order winners may be identity, cyber, observability and integration platforms that govern multi-agent environments regardless of which interface wins. Most exposed: NOW, CRM, MSFT, SAP and private OpenAI, with PANW, CRWD, ZS, OKTA and DDOG positioned as enabling control layers.
5. China’s memory expansion introduces a more credible structural threat to the AI-memory scarcity thesis than the recent semiconductor correction alone.
Chinese memory producers are benefiting from the AI-driven shortage, preparing potential public listings and expanding their strategic relevance, attracting greater US scrutiny while increasing the prospect of future price competition. The near-term bull case for Micron, Samsung and SK Hynix remains strong because HBM and advanced DRAM demand are constrained and China still faces technological and equipment limitations. The longer-term bear case is that government-backed Chinese capacity gradually commoditises conventional DRAM and NAND, freeing incumbent capital to move more aggressively into HBM and thereby accelerating the eventual supply response across the whole memory stack. The second-order impact is bifurcated: near-term localisation and capacity expansion support ASML, AMAT, LRCX and KLAC where export rules permit, but intensifying restrictions may fragment equipment markets and strengthen domestic Chinese alternatives. Most exposed: MU, Samsung, SK Hynix, WDC, Sandisk, ASML, AMAT and LRCX.
