Capital & Industrial Strategy

South Korea unveiled what amounts to the most ambitious state-coordinated AI industrial investment plan announced to date, with Samsung Electronics and SK Hynix leading commitments totalling a reported $1.3 trillion over a decade, spanning memory chips, data centers, and robotics. President Lee Jae Myung convened both companies' leadership at a formal briefing — a deliberate signal that this is sovereign industrial strategy, not voluntary corporate alignment. Bloomberg reports the investment details are still to be fully disclosed, meaning the headline figure should be treated as a declared intention rather than a confirmed capital commitment with binding terms.

The strategic logic is clear: South Korea's dominance in high-bandwidth memory (HBM) — the critical component enabling GPU clusters to function at AI-relevant scale — is a structural advantage that risks erosion if competitors, particularly Micron in the US and CXMT in China, close the technology gap. By locking in decade-scale investment now, Seoul is attempting to extend its lead before the HBM market matures. The choreography of having Samsung and SK Hynix leadership physically present beside the president mirrors the playbook used by Taiwan and the US CHIPS Act — government acting as convener and signal-amplifier for private capital.

The Bank for International Settlements published its annual economic report warning that the AI spending boom exhibits characteristics of 'exuberance' — elevated asset prices, weak near-term returns relative to capital deployed, and systemic financial fragilities if a correction triggers a funding pullback. The Financial Times and Semafor both report the BIS framing this as a potential multi-year 'investment bust' with global economic contagion risk, not merely a sector correction. The BIS is not a market participant, but its analytical credibility as a central bank institution means this warning will circulate in sovereign wealth, pension, and regulatory circles.

The key analytical tension is that the BIS warning coincides with accelerating capital deployment — $200 billion in US power sector M&A, $1.3 trillion in Korean chip commitments, and record venture activity. The BIS is essentially arguing that the infrastructure build-out is front-running monetisable demand by a dangerous margin. This is a structurally important risk for investors with long-dated AI infrastructure positions: if enterprise adoption plateaus or revenue-per-dollar-of-compute declines faster than expected, the correction in AI-adjacent equities and private valuations could be abrupt.

AI-driven demand for data center capacity has propelled M&A in the US power sector to a record $200 billion, per The Financial Times, as hyperscalers, independent power producers, and utilities engage in a dealmaking blitz to secure generation and transmission capacity. This is no longer speculative — it is confirmed deal volume at a scale that restructures the power industry's ownership landscape. The strategic intent is vertical integration of energy supply into the AI stack: companies that cannot secure power agreements face hard limits on data center expansion regardless of chip availability or software capability.

The European parallel is instructive: Bloomberg reports that European investors seeking AI exposure are rotating into power suppliers and banks, reflecting the same thesis — that the infrastructure enabling AI (power, cooling, financing) will capture a disproportionate share of value as direct AI equity becomes expensive. This convergence of US M&A activity and European equity rotation confirms that energy-as-AI-infrastructure is now a consensus institutional investment theme, not an emerging one.

Baidu's AI chip subsidiary Kunlunxin is reportedly targeting a $50 billion Hong Kong IPO, according to The Information as cited by CNBC and Reuters. Note this remains a reported target from a single publication and has not been confirmed by Kunlunxin or Baidu — treat as unconfirmed intention. Baidu's Hong Kong-listed shares rose over 6% on the news, indicating market credibility even absent official confirmation.

The strategic significance extends beyond Baidu's balance sheet. A $50 billion listing would validate China's domestic AI chip ecosystem under public market scrutiny, providing a funding mechanism for Kunlunxin to compete with Nvidia at scale without reliance on US capital markets. Listing in Hong Kong specifically — rather than mainland China — signals an intent to attract international institutional capital while maintaining access to Chinese sovereign support. This mirrors the logic behind SMIC's Hong Kong listing: using international markets to fund capabilities that US export controls are designed to retard.

Google has placed limits on Meta's use of Gemini AI models due to insufficient compute capacity, per the Financial Times as reported by Bloomberg. This is a confirmed operational constraint, not a commercial dispute — Google's own infrastructure cannot meet demand from a hyperscaler-tier customer. The implication is that frontier model compute is genuinely scarce, even among the largest infrastructure operators, and that API-tier access to leading models cannot be taken for granted by any enterprise or partner, regardless of their negotiating scale.

Simultaneously, Austria is lobbying the EU to host Anthropic within European borders in response to US access restrictions on advanced AI models, per Bloomberg. This is an early-stage lobbying effort — not a confirmed relocation or EU decision — but it signals that US export control policy is actively creating geopolitical arbitrage opportunities for European states competing to attract AI infrastructure and talent. The combination of Google rationing Gemini access and the US restricting frontier model exports creates a structural incentive for non-US jurisdictions to develop or attract sovereign AI capacity.