Capital & Industrial Strategy

SoftBank has announced plans to invest up to €75 billion — approximately $87 billion — to build 5 gigawatts of AI data centre capacity in France, according to reporting by Bloomberg and the Financial Times. Masayoshi Son is framing France as the centrepiece of his European AI infrastructure ambitions, a positioning that is inseparable from the French government's aggressive AI industrial strategy under President Macron, which has included regulatory accommodations, energy access commitments, and direct sovereign engagement to attract hyperscale investment.

This announcement is a confirmed intention, not a closed deal — the scale and phasing of capital deployment will depend on permitting, energy grid capacity, and SoftBank's own fundraising cycle, which remains heavily reliant on Vision Fund II performance and the anticipated monetisation of portfolio positions including Arm. The strategic logic is clear: by anchoring in France, SoftBank secures a beachhead in the EU single market, benefits from French nuclear baseload power — a critical differentiator for AI compute economics — and aligns with a government that has explicitly positioned itself as Europe's most capital-friendly AI host. Whether SoftBank can actually deploy this capital at the pace implied by the announcement is the operative question.

Dell, Nokia, Lenovo, Cisco, and Intel have collectively staged a dramatic market re-rating — adding approximately $1.7 trillion in combined market capitalisation — driven by the realisation that AI infrastructure buildout is a massive procurement event for exactly the hardware categories these companies dominate: servers, networking, storage, and enterprise compute, according to Bloomberg. Bloomberg Intelligence's Mandeep Singh characterises this as a structural pivot rather than a speculative overhang, noting that enterprise AI deployment cycles are now pulling through server refresh and network upgrade budgets that had been deferred for years.

The investment thesis here is distinct from frontier AI model plays. These companies are not winning because of AI R&D capability — they are winning because AI workloads require physical infrastructure at scale, and the hyperscaler capex cycle flows directly into their order books. Dell's AI-optimised server revenue has been the clearest signal; Nokia's network infrastructure business benefits from the data centre interconnect and enterprise edge networking demands that AI inference at scale generates. For capital allocators, this represents a lower-volatility, higher-visibility earnings growth profile compared to pure-play AI software names, which explains the institutional rotation.

With OpenAI and SpaceX IPOs generating significant liquidity events for early-stage investors, a portion of that capital is being actively redirected toward Asian hardware and component manufacturers positioned in the AI supply chain, according to Bloomberg. Investors are targeting companies across semiconductor packaging, cooling systems, power management, and AI-specific memory — the physical substrate of the inference and training infrastructure buildout. This rotation reflects a thesis that the application and model layers are increasingly contested and margin-compressed, while the supply chain layer benefits from durable demand with less competitive disruption risk.

The geographic concentration of this supply chain — Taiwan, South Korea, Japan, and increasingly Malaysia and Thailand for assembly — creates a distinct risk profile around geopolitical exposure, particularly US-China technology controls and Taiwan Strait stability. Sophisticated allocators are differentiating between supply chain nodes with genuine pricing power and those that are commoditising rapidly; advanced packaging and HBM memory currently sit in the former category, while standard PCB and enclosure manufacturing increasingly sits in the latter.

Meta is reported to be developing an AI-powered pendant device, according to TechCrunch, adding to a hardware portfolio that includes Ray-Ban smart glasses. Simultaneously, CNBC has published a substantive analysis of Meta's structural challenge: the company generates effectively all of its revenue from advertising, and every previous attempt to build a second revenue line — payments, commerce, VR/AR hardware — has failed to achieve material scale. The AI push is being framed internally and externally as the most credible attempt yet, given Meta's distribution advantages through WhatsApp, Messenger, Instagram, and Facebook.

The strategic logic for AI hardware is to create a persistent, ambient AI interface that captures user engagement outside the screen-based ad inventory Meta currently monetises. However, the pendant specifically is at the early speculation stage — this is reported product development, not an announced product. The more near-term AI monetisation question is whether Meta AI's integration into its messaging and social platforms can generate subscription or API revenue at scale, and whether enterprise AI services through its Llama ecosystem can build a durable B2B revenue line. Capital markets are pricing in AI optionality in Meta's multiple, but the actual revenue conversion timeline remains speculative.

Microsoft's decision to move GitHub Copilot to token-based billing has generated sharp developer backlash, with users characterising the pricing model as opaque and potentially expensive relative to the previous per-seat subscription, according to TechCrunch. The strategic context is that Microsoft is attempting to align Copilot pricing with actual compute consumption as inference costs become a more significant line item — a rational move from a unit economics perspective, but one that creates immediate friction with developers who valued predictable flat-rate pricing.

The competitive implication is significant: developer tooling loyalty is fragile, and any pricing friction creates an evaluation window for alternatives including Cursor, Windsurf, and JetBrains AI Assistant. Enterprise procurement teams are also likely to scrutinise the total cost of ownership implications of token-based billing at scale, which could slow renewal cycles. This is a direct signal about the tension between AI provider economics — where inference costs are real and variable — and enterprise buyer expectations of SaaS-style predictable pricing. How Microsoft resolves this tension will set a precedent for how the broader AI software market structures enterprise pricing.