1. IBM’s profit warning is the clearest evidence yet that AI infrastructure is not merely disrupting software economics; it is actively crowding software out of finite enterprise IT budgets.
IBM expects Q2 revenue of roughly $17.2bn, below c.$17.9bn consensus, after customers diverted spending towards servers, storage and increasingly expensive memory rather than closing planned software and mainframe transactions; the shares fell c.25%, dragging ServiceNow, Workday, Salesforce, Palantir, Microsoft and the wider software complex lower. The bull interpretation is that this is a temporary procurement distortion caused by component shortages and front-loaded hardware purchases; the bear interpretation is much more consequential — AI adoption may structurally redirect budget from high-margin licences and services towards chips, data centres, security and power, leaving “cheap” SaaS ex-growth rather than undervalued. The second-order implication is a more durable valuation bifurcation between infrastructure beneficiaries — MU, NVDA, AVGO, servers, networking and cyber — and application vendors that cannot prove incremental AI revenue exceeds seat compression and budget displacement.
2. TSMC’s Thursday result has become a referendum on whether hardware’s budget capture can sustain both semiconductor earnings and current valuations.
Q2 revenue has already reached a record T$1.27tn / c.$39.6bn, up 36% yoy, and consensus expects net profit to rise c.59% yoy to T$632.6bn / $19.7bn, a fifth consecutive quarterly record. The investor debate is no longer whether AI demand is strong — IBM’s warning arguably confirms that it is — but whether TSMC can raise revenue, pricing or capex guidance enough to clear expectations after a substantial share-price rerating. A strong outlook would validate continued tightness across leading-edge wafers and advanced packaging and support NVDA, AVGO, AMD, MRVL, ASML, AMAT, LRCX and KLAC; merely in-line guidance would strengthen the bear case that exceptional semiconductor fundamentals are already capitalised while hyperscaler returns and late-decade capacity risk remain unresolved.
3. Cybersecurity is now visibly competing for the same budget pool as software, but unlike generic SaaS it is becoming a compulsory cost of AI deployment.
The White House is launching an AI and cybersecurity coordination group as frontier models increasingly identify software and infrastructure vulnerabilities, while Reuters Breakingviews argues that AI-generated phishing, deepfakes and malware will force enterprises to spend more on defence and cyber insurance, reducing the net productivity benefit from AI. This sharpens the cyber bull case: AI may lower development and labour costs, but a portion of those savings is likely to be recycled into endpoint, identity, cloud, data, runtime and recovery controls. The bear debate is valuation and consolidation — greater spending does not guarantee every vendor wins — but the budget logic increasingly favours platforms with proprietary telemetry and enforcement points, particularly PANW, CRWD, FTNT, ZS, CYBR and OKTA, with RBRK, CVLT, TENB and QLYS benefiting from resilience and remediation requirements.
4. Fortinet’s FortiEndpoint expansion shows how the cyber battleground is moving from endpoint detection towards a single agent controlling AI usage, data and access.
Fortinet has added AI application visibility and governance, integrated data-loss prevention, secure access and an AI security assistant within one endpoint product, console and licence. Strategically, this is less about another feature release and more about platform compression: endpoint telemetry is becoming the enforcement layer for which AI tools employees and agents can use, what information they can access and whether sensitive data can leave the organisation. Bulls will see Fortinet’s networking footprint and bundling economics as an advantage against point products; bears will question whether it has the endpoint depth and identity context to displace CRWD, Microsoft or PANW. The wider read-through is that AI security is converging previously separate budgets — endpoint, DLP, secure access, browser and AI governance — increasing pricing pressure on standalone vendors while favouring FTNT, CRWD, PANW, ZS and MSFT.
5. US export policy is re-emerging as the largest non-fundamental variable in AI semiconductor forecasts.
A Commerce Department official said further regulatory action on AI and chips is coming and indicated that the current administration does not intend to replace the existing AI diffusion framework, which governs access to advanced accelerators across different country tiers. The bull case is that controls preserve US technological leadership and direct allied sovereign demand towards approved US chips, clouds and infrastructure providers; the bear case is that licensing uncertainty delays deployments, encourages domestic alternatives and makes Nvidia, AMD and hyperscaler forecasts dependent on geopolitics rather than end-demand alone. Second-order beneficiaries include compliant cloud providers, networking and sovereign-infrastructure partners, while NVDA, AMD and other advanced-chip suppliers retain the greatest direct policy sensitivity; persistent restrictions also support China’s incentive to accelerate local accelerators, foundry capacity and software ecosystems.
