Broadcom, Apollo, and Blackstone Form $35B Private-Capital AI Infrastructure Consortium
Broadcom has teamed with Apollo Global Management and Blackstone to co-fund a $35 billion push into AI compute, custom silicon, networking, and data center infrastructure — with Anthropic named as a key partner. The deal signals that private equity and credit markets, not just tech company balance sheets, are becoming direct funders of the AI build-out cycle.
A new financial architecture is forming around the AI infrastructure boom, and it is not being built exclusively from tech company cash flows. Broadcom, Apollo Global Management, and Blackstone have formed a $35 billion consortium to co-fund AI compute capacity, custom chip development, networking fabric, and data center physical infrastructure — with Anthropic named as a strategic partner in the first phase of deployment.
The consortium represents something qualitatively different from prior AI investment rounds. Apollo and Blackstone are not venture capital firms, nor are they technology companies. They are the two largest alternative asset managers in the world, with combined assets under management exceeding $2 trillion, and their primary capital instruments are private credit, leveraged buyouts, and infrastructure equity — not equity risk capital. Their entry into direct AI infrastructure funding signals that the asset class has crossed a threshold from speculative venture territory into bankable, cash-flow-producing infrastructure.
What the Consortium Is Funding
The first phase of the $35 billion commitment covers three areas:
Custom Silicon: Broadcom brings its ASIC design business — which has quietly built the world’s second-largest custom AI chip program behind NVIDIA, serving Google’s TPUs, Meta’s MTIA, Apple’s Neural Engine, and ByteDance’s custom training accelerators. The consortium will fund expanded capacity for Broadcom’s custom silicon pipelines, targeting the growing number of AI labs and enterprises that want to avoid dependency on NVIDIA’s pricing and allocation dynamics. Anthropic, which has been building a custom inference chip program, is the named first customer for this component of the consortium.
Networking Fabric: Next-generation spine-and-leaf networking for AI clusters is one of the most underappreciated bottlenecks in AI infrastructure. Broadcom’s networking silicon — including its Tomahawk and Jericho switch chips — currently handles the majority of high-performance AI cluster interconnect outside of NVIDIA’s NVLink ecosystem. The consortium will fund new generations of these switches designed specifically for the bandwidth demands of 100,000+ GPU clusters.
Physical Data Center Buildout: Apollo and Blackstone bring their existing infrastructure portfolios to bear here. Both firms have made significant investments in data center real estate, power infrastructure, and cooling systems over the past three years. The consortium will accelerate that buildout, with a particular focus on sites with access to high-capacity power — a constraint that has become the single largest bottleneck to deploying additional AI compute in the United States and Europe.
Why Private Capital Is Here Now
The entry of Apollo and Blackstone into direct AI infrastructure funding reflects a calculation that hyperscaler balance sheets alone cannot sustain the capital intensity of the current build-out cycle, and that private capital can fill the gap with returns that justify infrastructure-style risk.
Morgan Stanley estimated in May 2026 that AI-related debt issuance — structured products, infrastructure bonds, and credit facilities backed by AI compute assets — could reach $570 billion globally during the year, up from near zero in 2022. That estimate, which analysts greeted with skepticism when published, now looks conservative. The Broadcom-Apollo-Blackstone consortium alone accounts for $35 billion of that figure.
The bet underlying private credit’s entry is that AI compute infrastructure has predictable revenue characteristics: long-term contracts with creditworthy offtakers, high asset utilization rates driven by demand that is structurally growing rather than cyclical, and physical assets (buildings, power equipment, networking hardware) with meaningful recovery values. These are the characteristics that make infrastructure debt financeable at competitive rates, and they stand in sharp contrast to the software-equity risk that defines most AI investment.
Anthropic’s Role and the Broader Ecosystem
Anthropic’s involvement as a named partner is notable given the company’s parallel S-1 filing process and its stated ambition to reduce its dependency on third-party cloud infrastructure as it scales. The Rubin platform announced by NVIDIA today and the custom silicon component of the Broadcom consortium both speak to the same underlying strategic concern: that frontier AI labs which don’t control their compute stack face a structural cost and availability disadvantage versus those that do.
For Broadcom, the consortium validates a business model that has been building quietly for over a decade. CEO Hock Tan has consistently rejected the consumer narrative around AI, positioning Broadcom as the infrastructure layer beneath the AI labs and hyperscalers rather than a competitor to them. The custom silicon business, which contributed roughly $12 billion in revenue in Broadcom’s fiscal 2025, is now the company’s fastest-growing segment and the anchor of the consortium’s investment thesis.
What It Means for the Industry
The formation of a $35 billion private-capital consortium for AI infrastructure has several second-order implications. First, it establishes a template for how AI compute will be financed going forward — not just through hyperscaler capex, but through a layered capital structure that includes infrastructure equity, private credit, and strategic manufacturing partnerships. This mirrors how the energy sector finances large-scale power projects, and that analogy is not accidental given the power-intensity of AI data centers.
Second, it creates a new source of competitive pressure on NVIDIA. A Broadcom-powered custom silicon program with Apollo and Blackstone’s capital backing can credibly offer large AI customers an alternative to the NVIDIA stack at a scale that pure-software plays or smaller chip startups cannot. The consortium essentially bundles chip design, manufacturing scale, networking, and physical infrastructure into a single offering — the kind of vertically integrated solution that has historically been the domain of hyperscalers alone.
Third, and perhaps most consequentially for the broader economy, it signals that AI infrastructure has become a mature enough asset class for the most conservative large-scale institutional capital to enter directly. When Blackstone and Apollo move, pension funds and sovereign wealth funds typically follow within 18 to 24 months. The AI infrastructure buildout, in other words, may be just entering its most capital-intensive phase.