Blackstone Leads $18 Billion Debt Tranche to Finance Anthropic Chips as Part of Broader Package
Key Facts
Blackstone is leading an $18 billion junior debt tranche under arrangements to finance access to chips and computing infrastructure for Anthropic and other artificial intelligence companies, according to a Bloomberg report published on October 2. The group plans to allocate $9 billion from its various funds to the tranche and distribute the remainder to investors, according to the report, which cited people familiar with the matter for details of the arrangement. The Class B tranche forms part of a broader $60 billion financing package that a Broadcom consortium has begun raising. The package also includes a $42 billion Class A senior secured debt tranche, according to Investing.com, citing Bloomberg. The financing is tied to providing computing capacity used by AI companies, including access to chips and the infrastructure needed to run them. The package was at the investor fundraising stage when Bloomberg's report was published, with participating banks preparing to market the senior secured debt tranche.
In a more recent development on the financing terms, Reuters cited the Financial Times on October 16 as reporting that Anthropic is offering Blackstone a return above 10% to finance its latest chip deal. According to the newspaper, the debt is backed by chip lease payments Anthropic is committed to making over 5 years. This structure ties the financing to payments for using the hardware during the lease term, according to the terms reported by the Financial Times. The financing arrangement comes as Broadcom supplies hardware to Anthropic and participates in financing it, according to Anthropic's offering prospectus cited by Reuters. The prospectus warned that combining these roles creates a "potential conflict of interest." It said this conflict could affect Anthropic's ability to access the computing capacity it needs for its business.
The computing expansion agreed with Google and Broadcom includes Google's tensor processing unit chips supplied through Broadcom, according to Google's announcement of the partnership. Anthropic already uses multiple hardware platforms to train and run its Claude model, according to its statement on the agreement. These platforms include Amazon Web Services' Trainium chips, Google's tensor processing units and Nvidia's graphics processing units. The company said "this diversity of platforms translates into better performance and greater flexibility," describing the impact on customers of distributing workloads across different hardware. Alongside the expansion announcement on April 6, Anthropic said its annual revenue run rate had exceeded $30 billion. The company compared that with a run rate of about $9 billion at the end of 2025, the previous level it cited in the same announcement.
Blackstone previously participated in another financing platform to support the expansion of Anthropic's computing capacity, alongside Apollo and Broadcom. When that platform was launched, Blackstone announced an initial financing tranche of $35 billion. Apollo led the earlier tranche in partnership with Blackstone, according to the group's statement announcing the strategic platform. The financing was intended to support the deployment of computing infrastructure at Fluidstack sites, serving Anthropic's expansion of its available capacity. The statement set mid-2026 as the target for starting deployment of that infrastructure at Fluidstack sites.
In the new package, participating banks were preparing to send loan marketing letters to investors, according to Bloomberg's October 2 report. These letters concern the $42 billion senior secured debt tranche. Blackstone's arrangement covers the junior debt portion, with a planned contribution from its funds to the tranche it is leading. On hardware deployment, Anthropic announced in April that it expected the new tensor processing unit capacity agreed with Google and Broadcom to come online in 2027. The expected launch date for the new capacity is thus tied to the agreement the company announced with its suppliers to expand the computing infrastructure used to train and run Claude.