Capital & Industrial Strategy

The $500 billion-plus Nvidia-SK Group initiative announced this week is best understood not as a single deal but as a multi-layered supply chain strategy. At its core, Nvidia is securing preferential access to SK Hynix's high-bandwidth memory — the critical bottleneck for GPU performance — at a scale that will structurally disadvantage competitors dependent on the same supplier. As CNBC reported, Nvidia is aggressively locking down HBM supply as an essential component for its GPU systems. Separately, Bloomberg confirmed Nvidia will invest $1 billion directly into Naver to co-finance an AI data centre under construction in South Korea, deepening its Korean ecosystem play beyond the chip supply relationship.

The strategic intent here is twofold: first, constrain memory availability for AMD, which is attempting to scale its GPU business; second, anchor South Korea's AI infrastructure buildout around Nvidia's ecosystem, making it harder for domestic or Chinese alternatives to gain a foothold in a strategically important market. The Reuters announcement frames this as a mutual AI data centre initiative, but from Nvidia's perspective it is as much a defensive supply chain move as a growth investment.

SK Group Chairman Chey Tae Won disclosed to Bloomberg that Anthropic has approached SK Hynix seeking supplies to manufacture its own semiconductors. This is a significant strategic signal: Anthropic, which has positioned itself primarily as a frontier model developer, is now exploring vertical integration into custom silicon — a path previously taken by Google (TPUs), Apple (Neural Engine), and Amazon (Trainium/Inferentia). The timing is not coincidental; Anthropic's simultaneous release of Claude Opus 5, described by Reuters as both its best-performing and most cost-effective model, underscores a growing commercial pressure to reduce inference costs — something custom silicon directly addresses.

If confirmed and pursued at scale, this represents a material shift in Anthropic's capital requirements and strategic profile. Custom chip development demands multi-billion dollar commitments and multi-year timelines, which would reshape Anthropic's funding needs well beyond its current AI development roadmap. It also raises questions about competitive dynamics: Anthropic building custom silicon would reduce its dependence on Nvidia and could allow it to compete on cost in a market where CNBC reports enterprise buyers are increasingly cost-sensitive.

A coalition of more than 20 technology companies — including Nvidia, Microsoft, Meta, a16z, Dell, and Palantir — submitted a letter to US policymakers calling for the promotion of open-weight AI models and against broad restrictions, as reported by Bloomberg and Politico. The political context is Washington's ongoing debate over how to respond to Chinese open-weight models, particularly concerns about model distillation enabling Chinese labs to extract capability from American frontier models. The signatories argue that restricting open-weight models would undermine US technological leadership rather than protect it.

The strategic interests behind the letter are not uniform. For Nvidia, open-weight models drive GPU demand regardless of which lab produces them — restrictions would reduce the total addressable compute market. For Meta, which releases Llama as open-weight, restrictions would directly threaten its model distribution strategy. For Microsoft, open-weight models underpin Azure's AI platform offerings. The conspicuous absence of OpenAI and Anthropic — the two companies most commercially threatened by capable open-weight alternatives — is itself a data point. Both companies operate closed, proprietary model businesses where open-weight proliferation is a competitive threat, creating a direct conflict of interest with the coalition's position, as CNBC noted.

With Alphabet having raised its 2026 capex guidance by up to $15 billion, fixed-income markets are responding with widening credit spreads on Google, Amazon, and Meta, as CNBC reported. This is a qualitatively different form of pushback than equity analyst scepticism: bond investors are pricing in higher credit risk, meaning the cost of debt financing for AI infrastructure is rising at the precise moment hyperscalers are most reliant on it. Moody's has explicitly warned that AI spending is forcing even the world's most cash-generative corporations to lean heavily on debt, equity issuance, and off-balance-sheet structures, and has flagged threats to credit quality at Amazon, Meta, Alphabet, and others, per CNBC.

This dynamic matters for the investment thesis underpinning AI infrastructure: if the cost of capital rises for the largest buyers of AI compute, the implied returns required to justify continued capex escalation increase commensurately. Microsoft, Meta, and Amazon reporting earnings in the coming week face what Bloomberg describes as renewed investor anxiety over debt-fuelled capex plans. The practical question for each company is whether they can demonstrate revenue attribution to AI spending that satisfies both equity and debt investors simultaneously.

The Wall Street Journal reports that corporate AI buyers — both large enterprises and SMEs — are actively mixing frontier and lower-cost commodity models to reduce expenditure, fundamentally changing the economics of the model provider layer. This 'model mixing' behaviour, accelerated by the availability of capable open-weight and Chinese alternatives, means that frontier model providers can no longer assume enterprise customers will default to the most capable offering. The price premium for frontier capability is being arbitraged away by procurement teams that now have genuine alternatives.

This dynamic aligns with Anthropic's positioning of Claude Opus 5 as both highest-performing and most cost-effective — a deliberate dual message to enterprise buyers under cost pressure. It also contextualises the open-weight lobbying push: if enterprises can cost-effectively mix open-weight models for routine tasks with frontier models for complex ones, the market for pure-play frontier access contracts, redistributing value toward infrastructure, orchestration, and application layers. SLB, flagged by Fortune as positioning itself at the intersection of AI data centre power infrastructure and energy sector AI deployment, is a concrete example of the vertical integration and sector-specific deployment trend that benefits from cheaper AI access rather than being threatened by it.