Perplexity Will Run Its AI Workloads on IREN’s Blackwell Ultra GPUs

The Perplexity IREN agreement announced on Friday will put the AI search company’s workloads on NVIDIA Blackwell Ultra GPUs at an IREN data-centre campus, IREN said. “Under a multi-year agreement, Perplexity will access NVIDIA Blackwell Ultra GPUs at an IREN data center campus,” the Nasdaq-listed AI cloud provider wrote in a blog post dated 30 October.

IREN has not given a value, a term, a GPU count, a megawatt figure or the site for the agreement. In a post on X later that evening, it said the capacity will support “the models and agents behind Perplexity Computer”, the product Perplexity uses to run multi-step tasks for its users.

What the Perplexity IREN agreement covers

IREN’s page describes Perplexity as “the AI company built on the foundation of accuracy” and says it is “partnering with IREN to run AI workloads on IREN AI Cloud”. According to IREN, more than 50,000 organisations use Perplexity, and Perplexity Computer “takes complex projects from a single request to a finished result, coordinating agents that research, code and create”. IREN’s X post adds that Perplexity’s users ask “more than 1.5 billion questions a month”. Both products draw on models from several providers, IREN says.

Perplexity’s voice in the announcement comes from Nate Kupp, its vice-president of engineering, quoted on IREN’s page. “Every efficiency gain, from new models to new chips, should show up as better outputs at lower cost,” he said. “Dedicated compute is one element of making that possible, and we’re excited to work with IREN as we invest in the next phase of Perplexity’s growth.” TSN found no separate statement from Perplexity on its own channels by Saturday afternoon.

Tim Delcourt, commercial director for AI Cloud at IREN, said: “At IREN, our job is to keep capacity ahead of AI ambition. Owning and operating our infrastructure end to end gives us the control to bring capacity online reliably, drive greater efficiency and give our customers the confidence to scale.” IREN says it has “5GW of grid-connected power worldwide”.

How Perplexity fits IREN’s customer list

Perplexity is not a new name for IREN. In a 20 July release, filed with the US Securities and Exchange Commission, IREN said it had signed new multi-year cloud contracts worth $2.8bn in total and listed Perplexity in a customer base that also included Microsoft, NVIDIA, Figure AI and Together AI. IREN said its contracts have a weighted average term of about four years. Its full-year results on 27 August again named Perplexity among “other recent signings”, alongside Cohere and Prometheus.

The word “agreement” in IREN’s Friday blog links to the PDF of that July release. IREN has not said whether Friday’s post describes the same contract or a further one.

The announcement comes as IREN is turning more of its capacity into cash. This week it raised $363m against GPUs already in service, and earlier this month it proposed a 900MW campus near Wagga Wagga in New South Wales. The Information has also reported that Anthropic has an unannounced multiyear cloud deal with IREN, which neither company has confirmed.

A named workload, not a new number

The most useful thing in IREN’s announcement is not a figure but a name. Until now, Perplexity appeared on IREN’s customer lists without detail. On Friday IREN tied it to a specific chip generation, Blackwell Ultra, and a specific product, Perplexity Computer. That tells investors which kind of demand is filling the capacity: agent workloads that chain many model calls together, rather than one-off chat queries.

What it does not tell them is how big the commitment is. IREN’s August figures put contracted annualised run-rate revenue for 2026 capacity at $4bn, against $1bn operating at the time. A customer name with no value, term or megawatt figure cannot be placed inside that gap, so the announcement adds visibility rather than measurable revenue.

For Perplexity, Kupp’s wording matters. He calls dedicated compute “one element” of making outputs cheaper, not the whole answer, and Perplexity still relies on models from several providers. The company has been widening where its AI runs, including a recent deal for London’s Valarian to run Perplexity inside client systems. Reserved GPUs give it more control over cost and speed for the work it does itself, but its answers will still lean on other companies’ models.

IREN reports again on 5 November

IREN has said it will publish results for the quarter to 30 September on Thursday 5 November. That is the next point at which it may say how much of its contracted capacity, including Perplexity’s, is running.

Sources

• IREN: Perplexity and IREN: Powering Fast, Trustworthy AI at Scale (30 Oct 2026)
• IREN on X: Perplexity agreement post (30 Oct 2026)
• IREN: IREN Signs $2.8bn in New Customer Contracts with Leading AI Developers, SEC Form 8-K Exhibit 99.1 (20 Jul 2026)
• IREN: IREN Reports FY26 Results, SEC Form 8-K exhibit (27 Aug 2026)
• IREN investor news: Q1 FY27 results date announcement (22 Oct 2026)

Related stories

• IREN Unlocks $363m From Its Deployed AI GPUs in Wingspire Deal
• Anthropic Reportedly Has an Unannounced Multiyear Cloud Deal With IREN, The Information Says
• IREN Proposes Redbank, a 900MW Data-Centre Campus Near Wagga Wagga
• Perplexity Picks London’s Valarian to Run Its AI Inside Client Systems