en.Wedoany.com Reported - On June 9, Broadcom announced the establishment of the AI XPV platform in collaboration with Apollo and Blackstone's credit and insurance businesses, with Apollo and Blackstone participating as initial anchor investors. The platform's first-phase funding totals $35 billion and will support Anthropic's previously announced plan to expand its computing infrastructure beyond one gigawatt. Related deployments are expected to begin in mid-2026 at Fluidstack sites.
The core arrangement of the AI XPV platform combines Broadcom's custom XPU chips and networking solutions, tailored for frontier AI labs, with long-term institutional capital to provide larger-scale infrastructure for large model training and inference. According to Broadcom, the platform plans to support over 20 gigawatts of computing power deployment by 2028, serving frontier AI labs including Anthropic and OpenAI. The first-phase $35 billion transaction is led by Apollo, with Blackstone participating, and funds will initially be directed toward Anthropic's expansion of over one gigawatt of computing power. Compared to traditional data center financing, this platform-based arrangement is closer to a bundled approach involving "chips, networking, data center sites, long-term leases, and capital structures," transforming the computing needs of AI model companies into infrastructure projects that are financeable, scalable in phases, and deliverable over the long term. For Anthropic, computing power expansion is directly tied to next-generation model training, inference service stability, and enterprise customer delivery capabilities. For Broadcom, this platform integrates its custom XPUs, switching chips, and high-speed networking solutions into a larger AI infrastructure construction chain, helping expand its influence in the AI acceleration hardware market beyond GPUs.
The first deployments are expected to begin in mid-2026, targeting a scale exceeding one gigawatt. The platform's long-term goal extends to 2028, aiming to establish a global AI computing power deployment capacity of over 20 gigawatts.
This collaboration also indicates that competition in AI infrastructure is expanding from single-chip procurement to coordinated capabilities in capital, energy, networking, packaging, and data center operations. Large model companies require sustained access to high-density computing power; chip companies need to deliver customized hardware and networking systems as integrated packages; data center operators must provide sites capable of supporting high-power clusters; and private credit and insurance funds are seeking new digital infrastructure assets with long-term cash flow support. The AI XPV platform integrates these elements into a unified financing framework, enhancing upfront construction certainty and enabling continuous phased deployments over multiple years. As model training scales increase and inference request volumes grow rapidly, frontier AI labs will continue to raise requirements for per-unit inference costs, energy consumption levels, chip supply stability, and network interconnect efficiency. Whether Broadcom's custom XPUs and networking solutions can achieve replicable, large-scale delivery will be a key variable in the platform's subsequent expansion.
The entry of Apollo and Blackstone indicates that AI computing power has become a key new infrastructure direction for large institutional capital allocation. The $35 billion first-phase funding is not merely a chip procurement budget but a capital solution centered on Anthropic's computing power expansion. It will need to align with site construction, equipment delivery, power supply conditions, customer leases, and multi-party risk-sharing mechanisms. If the platform proceeds as planned, AI computing power construction will further exhibit characteristics of large scale, financialization, and long-term contracting, while the boundaries of collaboration among chip manufacturers, cloud infrastructure companies, and capital institutions will continue to be reshaped.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
