The Gist — Capital & Industrial Strategy

Top Line

Nvidia's Seoul blitz produced a cluster of confirmed deals — a multi-year chip co-development pact with SK Hynix, gigawatt-scale data centre agreements with Naver, and expanded robotics and infrastructure partnerships with LG and SK Group — cementing Nvidia's position as the indispensable infrastructure layer across Asia's AI build-out.

Chinese AI startup Moonshot AI is in talks to raise up to $2 billion at a $30 billion valuation, its third fundraise in six months, signalling that China's frontier model race is intensifying and capital requirements are escalating rapidly despite a global tech equity rout.

A broad selloff in AI-linked equities — SoftBank down over 7%, Asian tech names tracking U.S. losses led by Broadcom — has collided with a prospective flood of AI-related equity issuance, raising serious questions on Wall Street about whether demand can absorb the supply pipeline.

UK AI simulation startup PhysicsX closed a confirmed $300 million raise at a $2.4 billion valuation led by Temasek, while the UK government separately signalled a procurement commitment to buy AI chips from domestic tech firms — a rare pairing of private and public capital signals in one market.

Private equity software buyout deal value collapsed to $50 billion in the first five months of 2026, the lowest since the pandemic, as AI-driven valuation uncertainty and rising rates freeze leveraged acquisition logic in legacy software.

Key Developments

Nvidia's Korea Sweep: Supply Chain Lock-In and Ecosystem Expansion

Jensen Huang's Seoul visit produced a set of commercially significant, confirmed agreements that collectively tighten Nvidia's grip on the AI infrastructure stack in Northeast Asia. The most technically consequential is the multi-year co-development pact with SK Hynix on next-generation AI memory — confirmed by both parties — which positions SK Hynix as a preferred architectural partner for HBM and future memory generations, a direct competitive blow to Samsung at a moment when memory design integration with GPU roadmaps is becoming a strategic differentiator. Huang separately flagged a prolonged chip shortage, a signal that has the dual effect of validating supply-side investment and reinforcing customer urgency. Bloomberg

Beyond the memory deal, Nvidia confirmed that Naver will build gigawatt-scale AI factories using Nvidia technology and models, locking Korea's dominant internet platform into Nvidia's compute and software ecosystem as it competes against hyperscaler encroachment. LG Group deals on humanoid robotics and data centres extend Nvidia's footprint into physical AI, a segment where hardware-software integration creates durable switching costs. The Seoul meetings were attended by leaders of SK Group, LG Group, and Naver — a deliberate display of coalition-building that mirrors Nvidia's Gulf state strategy from 2025, suggesting a systematic approach to securing national AI infrastructure commitments market by market. Wall Street Journal Reuters

Why it matters

Nvidia is converting transactional chip sales into long-cycle architectural partnerships and national infrastructure commitments, making displacement by AMD, Intel, or custom silicon structurally harder and commercially costlier for its largest customers.

What to watch

Whether Samsung moves to counter with its own co-development agreements with AMD or custom ASIC buyers — and whether SK Hynix's preferential alignment with Nvidia triggers any South Korean government concern about concentration risk in its own semiconductor export base.

AI Equity Rout Meets IPO Supply Flood: A Structural Demand Problem

The current selloff is not merely a sentiment correction — it is occurring precisely as a large pipeline of AI-related equity issuance is readying to hit public markets. Bloomberg's analysis identifies the prospect of mega AI deals flooding share supply at a moment when institutional demand is wavering, a technical condition that could suppress valuations across the sector regardless of fundamental performance. Bloomberg SoftBank's 7%-plus single-session decline and Broadcom-led losses across Asian tech names reflect how quickly sentiment can reverse when leverage and momentum positioning unwind simultaneously. CNBC

TechCrunch's analysis of the 'Tokenpocalypse' — rising API prices driven by AI labs preparing for public market scrutiny of their unit economics — adds a further layer: companies are repricing access to their models upward partly to demonstrate margin improvement ahead of IPO roadshows. TechCrunch Jensen Huang's public characterisation of the selloff as a buying opportunity is strategically rational from Nvidia's perspective — it reinforces the capital expenditure commitments of its largest customers — but the divergence between his messaging and the market's behaviour in Seoul is notable. Bloomberg

Why it matters

A sustained equity rout concurrent with peak AI issuance volume would compress the funding window for mid-tier AI infrastructure players, force repricing of private rounds, and potentially defer the IPO timelines that frontier labs have been building toward.

What to watch

The pricing and reception of the first major AI-related public offering in this environment will function as a real-time referendum on whether institutional capital has genuinely rotated out or is waiting for better entry points.

China's Frontier Model Arms Race: Moonshot AI Raises at $30 Billion

Moonshot AI — maker of the Kimi model — is in confirmed funding talks seeking up to $2 billion at a $30 billion valuation. This would be its third capital raise in six months, a cadence that reflects both the escalating compute costs of staying competitive in China's frontier model market and the availability of domestic capital willing to back the race. Bloomberg The deal is described as being in talks — terms are not confirmed — but the valuation trajectory and frequency of rounds signals that Chinese frontier labs are on a capital consumption path comparable to their U.S. counterparts.

The competitive context matters: Moonshot is racing against ByteDance, Alibaba, Baidu, and Zhipu AI, all of which have deep corporate balance sheets subsidising their model development. Independent frontier labs in China face a structural disadvantage in a war of attrition against platform giants, making the speed of external fundraising an existential variable. The $30 billion valuation — if achieved — would make Moonshot one of the most valuable private AI companies globally, comparable to mid-tier Western frontier labs, despite operating under U.S. chip export restrictions that constrain compute access.

Why it matters

Sustained capital availability for independent Chinese frontier labs despite export controls suggests that domestic investors have concluded model capability can be maintained through efficiency gains and alternative hardware — a direct challenge to the strategic premise of the U.S. export control regime.

What to watch

Whether Moonshot closes this round at or near its target valuation, and which investors participate — sovereign funds, corporate strategics, or purely financial — will indicate how the Chinese AI capital ecosystem is structuring itself for the next phase of the race.

UK AI Capital Formation: PhysicsX Raise and Government Procurement Signal

London-based PhysicsX — which applies AI to physics simulation for aerospace, automotive, and energy applications — confirmed a $300 million raise at a $2.4 billion valuation, led by Singapore's Temasek. The deal is closed and terms are confirmed. Financial Times This makes PhysicsX one of the UK's most valuable AI companies and is notable for being a deep-tech, sector-specific application play rather than a foundation model or enterprise SaaS bet — indicating that sophisticated capital is increasingly flowing toward AI with defensible domain moats.

Separately, the UK government announced — per a Telegraph report cited by Bloomberg — a commitment to purchase AI chips from British technology companies as an incentive for firms to maintain UK presence. Bloomberg The detail on scale, budget, and qualifying criteria has not been confirmed in primary government documents, and the announcement should be treated as a policy intention rather than a confirmed procurement contract. Nonetheless, the pairing of a Temasek-backed deep-tech close with a government procurement signal in the same week is a meaningful indication that the UK's AI industrial strategy is gaining traction with international capital.

Why it matters

PhysicsX's raise validates a thesis that AI companies with proprietary scientific simulation capabilities and capital-intensive switching costs can command premium valuations independent of the foundation model cycle — a differentiated capital formation path for European AI.

What to watch

The structure and legislative backing behind the UK's chip procurement commitment — if formalised, it would represent a meaningful demand-side industrial policy lever comparable to U.S. procurement preferences and could influence where semiconductor and AI hardware companies choose to incorporate and operate.

Private Equity Software Buyouts at Pandemic-Era Lows

The value of private equity software acquisitions in the first five months of 2026 fell to $50 billion, the lowest level since the pandemic, according to Financial Times analysis. Financial Times The proximate cause is AI-driven valuation uncertainty: buyers cannot confidently underwrite future cash flows for software businesses whose product categories face displacement risk, while sellers anchored to 2024-era multiples are unwilling to accept discounts. This bid-ask spread has frozen deal-making at precisely the moment when PE firms are under pressure to deploy dry powder.

The strategic implication is that the traditional PE software playbook — buy, cut costs, re-sell — is structurally compromised in an AI transition environment. The businesses most attractive to buyers are those that have credibly integrated AI into their product and cost structure, but these are also the businesses where sellers are holding out for AI-premium valuations. The result is a market in which only the most clearly AI-native software companies are transacting, while legacy SaaS and enterprise software sits in limbo.

Why it matters

A sustained freeze in PE software buyouts concentrates exit pressure on the venture-backed AI-native cohort, potentially forcing down Series D and late-stage valuations as secondary buyers retreat and IPO windows remain uncertain.

What to watch

Whether the buyout freeze extends into H2 2026 or whether a valuation reset — driven by distressed sellers or improving credit conditions — unlocks a wave of opportunistic acquisitions that restructures the enterprise software landscape around AI-capable acquirers.

Signals & Trends

National AI Infrastructure as Geopolitical Currency: The Korea Model

The density of Nvidia deal-making in Seoul — spanning memory co-development, gigawatt data centre construction, robotics, and cloud platforms — in the same week that South Korea's president cited AI as a national growth engine and Jensen Huang appeared publicly with the country's corporate leadership, represents a template for how Nvidia is converting geopolitical AI ambitions into long-cycle infrastructure lock-in. South Korea's position is structurally advantaged: it has memory dominance through SK Hynix and Samsung, a diversified tech corporate base with global scale, and a government actively channelling semiconductor tax revenues into AI investment. FT The pattern — Nvidia arrives, bilateral deals are signed, the host government signals national AI commitment — has now been replicated across the Gulf, Japan, and Korea. The signal for investors is that jurisdictions which have not yet secured a comparable Nvidia partnership are increasingly at a structural disadvantage in the global AI infrastructure race, and will face pressure to offer preferential terms to attract the next wave of compute deployment.

The AI Valuation Bifurcation: Domain Moat vs. Horizontal Platform

The concurrent collapse of PE software buyouts, the equity selloff in broad AI-exposed names, and the successful close of PhysicsX at $2.4 billion points to an emerging bifurcation in how capital is pricing AI exposure. Horizontal AI platform plays — foundation models, general-purpose SaaS with AI features, infrastructure aggregators — are facing multiple compression and demand uncertainty. Meanwhile, companies with deep domain specificity, proprietary training data from physical or scientific processes, and high switching costs (PhysicsX in simulation, potential analogues in genomics, materials science, and industrial automation) are closing rounds at premium valuations with sovereign and strategic capital. The FT's investment super-cycle thesis — that AI, clean energy, and defence are reinforcing each other — supports the view that the most durable AI capital formation will occur at the intersection of AI and hard physical domains, not in software abstraction layers. FT Investment strategists should track whether this bifurcation hardens into a persistent valuation premium for domain-specific AI, which would reshape portfolio construction logic across growth equity and late-stage venture.

Power Constraints as the Binding Infrastructure Bottleneck

Ireland's government telling hyperscalers to bring their own power to qualify for data centre approvals is a leading indicator of a constraint that will increasingly shape where AI infrastructure capital can be deployed. Wall Street Journal The gigawatt-scale ambitions announced by Naver in Korea, combined with Nvidia's multi-site infrastructure deals, imply an aggregate power demand that no single national grid can absorb without structural grid investment. The investment implication is twofold: companies with captive power generation capability (nuclear offtake agreements, dedicated renewable assets, co-located generation) will command a durable siting advantage over competitors dependent on public grid capacity; and jurisdictions that proactively solve the power constraint — through grid investment, permitting reform, or captive generation incentives — will attract disproportionate AI infrastructure capital relative to their market size.

Explore Other Categories

Read detailed analysis in other strategic domains

Compute & Infrastructure

Nvidia and SK Hynix have moved beyond supply agreements into co-designing future AI memory architectures, with Huang confirming SK Hynix memory will ship inside the Vera CPU at the platform level. This design-in advantage builds switching costs that Samsung cannot easily overcome by fixing yield problems alone. The strategic irony is that Nvidia's dominance now rests on a supply chain concentrated in a single Korean vendor sharing Taiwan's geopolitical risk profile.