Capital & Industrial Strategy

Broadcom, Apollo Global Management, and Blackstone's credit and insurance unit have finalised a $35 billion platform to finance Anthropic's AI infrastructure expansion, with Google backstopping lease payments across five data centers — effectively providing the credit enhancement that makes the deal bankable. Bloomberg, WSJ, and FT have all confirmed the deal is finalised. Structurally, this is a leveraged lease arrangement: Anthropic accesses compute without owning it outright, Apollo and Blackstone earn yield on long-duration AI infrastructure debt, Broadcom secures a captive customer for its custom silicon, and Google — already a major Anthropic equity investor — deepens its strategic lock-in by acting as guarantor.

The deal's architecture reveals how the AI infrastructure financing market is maturing. Traditional bank lending is insufficient at this scale and risk profile; instead, private credit giants with insurance balance sheets are stepping in to hold long-dated paper. Google's backstop is the critical piece — without it, the creditworthiness of a pre-profitable AI lab is insufficient collateral for a $35 billion commitment. This creates a template: hyperscaler as credit guarantor enables AI frontier labs to access compute capital markets at scale. Expect to see this structure replicated as OpenAI, xAI, and others pursue similar build-outs.

OpenAI has confidentially filed its S-1 with the SEC — with management pre-empting the leak by announcing it publicly — joining Anthropic, SpaceX, and others in a generational IPO cohort. PitchBook estimates the aggregate implied value of pending AI-linked IPOs at $3.6 trillion. Bloomberg and Semafor note OpenAI may be the most expensive bet on a price-to-revenue basis in the cohort. The MANGOS acronym — Microsoft, Apple, Nvidia, Google, OpenAI, SpaceX — is gaining traction as shorthand for a new generation of market-cap dominants, per TechCrunch.

The macro concern flagged by the FT is structural: this IPO wave, combined with slowing buybacks, could remove the equity supply contraction that has supported market valuations for a decade. For AI-sector investors, the more immediate risk is valuation compression in the private market as public comparables get priced — and as Semafor notes, the capital requirements of these giants risk crowding out second-tier AI companies seeking public market access.

China is preparing a $295 billion state-directed plan to fund a nationwide AI infrastructure buildout, according to Bloomberg via Reuters. This is a confirmed government proposal but implementation details remain subject to political ratification. Simultaneously, Taiwan is actively considering aligning its AI chip export controls with US measures, per Bloomberg — a move that would tighten the semiconductor smuggling routes that have partially circumvented existing US restrictions. TSMC's May sales were up 30% year-on-year, and Taiwan's May exports hit their second-highest monthly value on record, driven by AI chip demand, per Reuters.

China's semiconductor export surge — up 110% year-on-year per Semafor — suggests domestic chip production is accelerating even under sanctions pressure. The combination of a $295 billion state buildout, surging domestic semiconductor output, and the prospective tightening of Taiwan's export controls creates a decisive fork: China is engineering supply-chain autarky in AI compute while Western allies are closing the gaps in their containment architecture. For investors in the AI infrastructure stack, the geographic segmentation of the semiconductor market is becoming structural, not cyclical.

Super Micro Computer announced a $7 billion equity raise — confirmed, with terms to be structured across multiple offering tranches — to purchase AI server components and fulfil existing customer orders, per Bloomberg and CNBC. The stock fell sharply on the announcement, reflecting investor discomfort with the dilution quantum even as the underlying demand signal — customers placing orders that require $7 billion in components — is unambiguously positive. This follows a pattern seen across the AI server supply chain: demand is real and large, but the working capital requirements to fulfil it are stretching balance sheets.

Microsoft-backed D-Matrix is entering full production of an inference chip it claims is 10x faster than a GPU and bypasses the memory bottleneck that constrains current architectures, per CNBC. Lumentum's CEO separately highlighted indium phosphide optical chips as a critical enabler of next-generation data center interconnects, with AI-driven bandwidth demand accelerating the shift from copper to fibre. These developments together suggest that AI infrastructure differentiation is moving up the stack from raw compute to memory architecture and optical interconnect — areas where Nvidia's dominance is less entrenched.

Meta has confirmed a partnership with Reliance Industries to build its first AI data center in India, per Bloomberg and Reuters. The Reliance tie-up follows a familiar playbook — local infrastructure partnerships to navigate regulatory scrutiny and build political goodwill in large emerging markets. India represents one of the few markets where Meta can still drive material user and engagement growth, making AI-native WhatsApp services a strategic priority.

Simultaneously, the EU has ordered Meta to open WhatsApp's messaging infrastructure to rival AI chatbots under the Digital Markets Act, a decision Meta has publicly condemned as enabling competitors like OpenAI to free-ride on its distribution. The order is confirmed as issued but subject to Meta's legal challenge. The strategic tension is acute: Meta is investing to make WhatsApp an AI platform globally while being compelled in its largest regulatory jurisdiction to commoditise that distribution advantage.