Broadcom, Apollo and Blackstone Launch $35B AI XPV Platform

On June 9, 2026, Broadcom, Apollo, and Blackstone launched a $35 billion financing platform called the AI XPV Platform, designed to fund more than 20 gigawatts of AI compute capacity through 2028, with Anthropic’s first gigawatt already slated for Fluidstack sites starting in mid-2026. Apollo is the lead investor, Blackstone is the partner, and Broadcom is the chip supplier anchoring the structure. Apollo called the deal “the largest private financing ever executed.”

The arrangement puts Broadcom at the center of a new template for AI infrastructure finance, where Wall Street banks, private credit, and chip designers pool capital to underwrite multi-gigawatt compute builds. It also extends Broadcom’s role beyond silicon. The company is on the hook for the senior debt on a single-customer, single-operator credit, and the stock has its own verdict, with AVGO down 8.2% over the past 30 days.

The $35 Billion Tranche and the 20-Gigawatt Target

Broadcom, Apollo, and Blackstone on June 9 launched the AI XPV Platform, a strategic financing vehicle the three parties set up to deploy more than 20 gigawatts of AI compute capacity through 2028. The platform opens with a $35 billion initial tranche led by Apollo, in partnership with Blackstone, to finance more than 1 gigawatt of compute infrastructure for Anthropic at Fluidstack-based sites starting in mid-2026. That initial deployment is the first under the platform, per the official announcement of the AI XPV Platform. Roughly half of the $35 billion has been syndicated to other investors, with the rest concentrated on Broadcom’s books as the senior backstop.

“We are at a historic inflection point where the demand for AI compute is fundamentally reshaping the global economic landscape,” said Hock Tan, President and CEO of Broadcom, in the same release. The platform, in his framing, “synchronizes the world’s most sophisticated capital with Broadcom’s advanced technological roadmap” to meet what Tan called a “once-in-a-lifetime opportunity.” Tan’s quote signals that Broadcom sees XPV as an operational extension of the chip business, not a side project.

The XPV is built specifically for Broadcom’s custom XPU accelerators and the company’s networking silicon, the same components it designs for Google’s Tensor Processing Units and OpenAI’s custom chips. The platform is designed for “leading frontier AI labs, including Anthropic and OpenAI,” per the announcement, with OpenAI’s October 2025 10-gigawatt custom accelerator deal with Broadcom already in the pipeline. The structure establishes a “scalable framework for future deployments of XPU-based compute capacity and networking” to enable frontier model training and inference at lower per-token delivery costs. The 20-gigawatt ambition is not a chip target. It is a financed compute target, with each gigawatt backed by long-dated capital partners rather than spot GPU sales. That distinction is the spine of the XPV pitch.

• $35 billion initial tranche, led by Apollo with Blackstone
• 20+ gigawatts of compute capacity targeted by 2028
• $1.03 trillion Apollo assets under management (March 31, 2026)
• $1.3 trillion Blackstone assets under management

Apollo and Blackstone on the Front End

Apollo’s $35 billion capital solution details make clear that Apollo-managed funds and affiliates are leading the initial tranche, with Blackstone’s Credit & Insurance Business as the partner anchor. As of March 31, 2026, Apollo had approximately $1.03 trillion in assets under management, and Blackstone’s AUM is over $1.3 trillion, giving the two alternative asset managers the balance-sheet scale to commit capital across multi-year draw schedules. Apollo Partner Jamshid Ehsani framed the deal as “the largest private financing ever executed,” leaning on what he called Apollo’s “integrated platform across High-Grade Capital Solutions, Apollo Capital Solutions and ATLAS SP Partners.” The lead bank syndicate is Goldman Sachs, Wells Fargo, and Citi, with Wells Fargo as global coordinator on the A1 tranche and BNP Paribas, Citi, and UBS as joint bookrunners.

Blackstone President Jon Gray called the platform “an unprecedented opportunity to invest at scale across the AI infrastructure ecosystem, including providing financing through our credit and insurance business.” Blackstone’s role is the credit and insurance side of the structure, not the asset-management flagship, which is a deliberate choice for a deal that needs long-dated, insurance-balance-sheet-style capital. Broadcom is advised by Morgan Stanley, with JPMorgan Chase as co-advisor and Sullivan & Cromwell as legal counsel, per the Apollo release. Apollo is advised by Latham & Watkins as lead counsel, with Paul, Weiss as special counsel, PwC as accounting advisor, and Milbank as investors’ counsel. The advisor stack is the same one a $35 billion leveraged buyout would carry, signaling that the AI compute deal has been engineered to bank-credit standards.

The Customer Pipeline Filling the Roster

Anthropic is the first gigawatt on the platform. The Claude maker’s previously announced capacity expansion of more than 1 gigawatt of compute is expected to deploy in Fluidstack-based sites starting in mid-2026, with the $35 billion tranche structured around it. Anthropic itself committed $50 billion to American AI infrastructure in November 2025, with Fluidstack as its primary build partner for sites in Texas and New York.

Anthropic’s growth makes that scale digestible, per Anthropic’s multi-gigawatt TPU agreement announcement in April. The company’s run-rate revenue has surpassed $30 billion, up from approximately $9 billion at the end of 2025. More than 1,000 business customers are now each spending over $1 million on an annualized basis, double the 500 customer count Anthropic cited in February at its Series G. Anthropic trains and runs Claude on AWS Trainium, Google TPUs, and NVIDIA GPUs, so the new XPU financing slots into a multi-vendor strategy rather than replacing an existing one.

OpenAI sits in the next slot, per OpenAI’s 10-gigawatt accelerator deal with Broadcom from October 13, 2025. Under that collaboration, OpenAI will design accelerators that Broadcom will co-develop and deploy, with deployment starting in 2026. Google is also a customer, with Anthropic’s April 6 deal adding “multiple gigawatts of next-generation TPU capacity” from 2027. Beyond those three, Broadcom has signed Meta, TikTok owner ByteDance, and a fifth XPU customer, with custom AI accelerator programs covering Google’s TPUs, OpenAI’s custom chips, and Meta’s MTIA. The roster is the demand side of the XPV capital, and the financing is the supply side.

1. October 13, 2025 – OpenAI and Broadcom announce 10 gigawatts of custom AI accelerators
2. November 2025 – Anthropic commits $50 billion US infrastructure with Fluidstack
3. February 2026 – Anthropic Series G
4. April 6, 2026 – Anthropic signs multi-gigawatt TPU agreement with Google and Broadcom (capacity from 2027)
5. Mid-2026 – First 1+ gigawatt of Anthropic compute goes live at Fluidstack sites
6. June 9, 2026 – AI XPV Platform launches with $35 billion first tranche
7. 2027 – Google TPU capacity for Anthropic comes online
8. 2028 – Platform’s 20+ gigawatt target reached

Broadcom’s Hidden Credit Risk

Apollo called the deal “the largest private financing ever executed.” Bloomberg, reporting on the three-tranche debt structure and syndication, called the package “one of the biggest private credit transactions in history.” Both characterizations are accurate, and both leave out the most consequential line in the structure.

Broadcom is backstopping payments on the largest senior portions of the $35 billion debt, per people familiar with the matter cited by Bloomberg. Roughly half the $35 billion of debt was syndicated to other investors, leaving Broadcom with concentrated credit exposure on the remainder. The capital funds Google’s custom TPUs that Anthropic will lease, which means Broadcom’s risk sits on top of a TPU supply chain it does not own and a single operator, Fluidstack, it does not control. The structure is closer to a project finance deal than a chip supply contract.

The way Apollo describes the asset class underscores the bet.

AI compute is rapidly emerging as one of the most compelling new asset classes in finance, characterized by contracted cash flows, mission-critical utility and a supply-demand dynamic that continues to intensify.

Jamshid Ehsani, Apollo Partner, said that in the June 9 release. That framing is correct, and it relies on the contracts holding. The cash flows that the senior debt service depends on are Anthropic’s lease obligations on the TPUs the $35 billion is funding. Broadcom’s backstop is, in effect, a guarantee that those contracts perform.

The size of the bet is also a step change from the prior high-water mark in private AI data-center financing. Meta’s $27 billion Hyperion data center joint venture with Blue Owl Capital, announced October 21, 2025, was the previous record, and it financed the building and the power, not the chips. The XPV tranche is $8 billion larger, and the chips it finances are the most valuable asset inside the data center. The capital has moved upstream.

The Stock Is Sending a Different Signal

AVGO closed at $380.15, up 45.2% over the past year and about 7x higher over five years, a long arc that reflects Broadcom’s role in the AI build-out. The 12-month consensus analyst price target sits at $523.84, with 48 analysts rating the stock Strong Buy. That target implies about 38% upside from the June close, a wide gap that has not narrowed even as the XPV deal landed. The Street, on average, still thinks Broadcom is mispriced by more than a third.

The short-term tape has not agreed. AVGO is down 8.2% over the 30 days through mid-June, even with the XPV announcement in the window. On June 3, after Broadcom’s fiscal Q2 2026 results, the stock fell 12% to 16% in a single session, with the move triggered by a Q3 AI chip revenue guide of about $16 billion, slightly below the $16.36 billion analyst estimate. The XPV deal arrived into a stock that was already sliding on its own fundamentals.

The Q2 numbers themselves are not the cause of the slide. AI semiconductor revenue hit $10.8 billion in the quarter, up 143% year-over-year, above Broadcom’s own forecast. That line, and the trajectory behind it, is the closest thing to a perfect print that an AI chipmaker has produced in 2026. The market reaction reflects what investors were willing to pay for, which is more than the Q3 guide implied, and the same pattern played out in how Broadcom erased $300 billion after its best AI quarter in the days after the Q2 release. The XPV deal, on its own, has not reset that tension.

How the $35B Could Snap

The $35 billion tranche is concentrated on a single customer, Anthropic, and a single operator, Fluidstack, with Broadcom on the senior backstop. A slowdown in Anthropic’s growth or a delay at the Fluidstack sites would change the contracted cash flow story Apollo pitched. The credit risk is real because the asset is real, and the asset is a 1+ gigawatt TPU build that has to come online on the schedule the senior debt assumes. None of the parties has said the schedule is slipping, but the structure is exposed if it does.

A credit cycle turn or a broader AI demand reset would test the backstop at the same time the stock is already under pressure. Apollo’s framing of “contracted cash flows” assumes those contracts hold through a downturn. The XPV template is engineered to bank standards, and Broadcom’s shareholders are the ones underwriting that bet.

Frequently Asked Questions

What is the Broadcom AI XPV Platform?

A strategic financing platform Broadcom set up with Apollo and Blackstone on June 9, 2026 to fund AI compute infrastructure. The platform targets more than 20 gigawatts of capacity using Broadcom’s custom XPU accelerators and networking products through 2028. The first $35 billion tranche finances more than 1 gigawatt of compute for Anthropic at Fluidstack sites starting mid-2026.

Why did Broadcom need Apollo and Blackstone?

Traditional capital markets cannot fund AI compute at the speed and scale frontier labs now need, per Broadcom’s Head of Corporate Development Won Kim in the June 9 release. Apollo, with approximately $1.03 trillion in assets under management, and Blackstone, with over $1.3 trillion, can deploy committed multi-year capital where banks cannot. The alternative was slower, smaller chip-by-chip sales to a customer base that is asking for gigawatts at a time.

What is an XPU?

Broadcom’s term for custom AI accelerator chips the company designs with hyperscale customers. The XPU line includes Google’s Tensor Processing Units (TPUs) for Google and Anthropic, custom accelerators for OpenAI, and Meta’s MTIA. The XPU category also includes Broadcom’s networking silicon, the connectivity layer that lets thousands of accelerators operate as one system.

How is Broadcom exposed if the loans go bad?

Bloomberg reports Broadcom is backstopping payments on the largest senior portions of the $35 billion debt. Roughly half the deal was syndicated to other investors, leaving Broadcom with concentrated credit risk on the remainder. The risk is concentrated on a single customer (Anthropic) and a single operator (Fluidstack), with Google TPU supply upstream.

Why has Broadcom’s stock dropped since the deal was announced?

AVGO is down 8.2% over the 30 days through mid-June, even with the XPV launch in the window. On June 3, the stock fell 12% to 16% in a single session after Broadcom guided Q3 AI chip revenue of about $16 billion, slightly below the $16.36 billion analyst estimate. The slide reflects the tension between the long-term AI build and the near-term forecast, and the XPV deal has not yet changed that math.