Broadcom is seeking roughly $60 billion tied to AI chips, according to a Moneycontrol report, as Anthropic races to expand its computing power and joins the queue of AI labs buying custom silicon rather than merchant GPUs alone.

The $60 billion figure is the story's hardest number, and it lands at a moment when every major AI developer is rethinking where its compute comes from. Broadcom has positioned itself as the leading merchant designer and supplier of custom AI accelerators — application-specific integrated circuits built for a single cloud or AI customer — and Anthropic's push to scale training and inference capacity puts it in contention for that pipeline.

Why does Anthropic need custom silicon?

AI labs face a structural problem: demand for high-end accelerators from incumbent suppliers outstrips supply, and general-purpose GPUs force labs to pay for capabilities their workloads may not use. Custom ASICs, co-designed with Broadcom and manufactured at leading-edge foundries, let an AI developer tune memory bandwidth, interconnect and compute density to its own models.

For Anthropic, which races rivals to expand computing power for training frontier models, locking in a dedicated silicon program would convert compute from a spot-market scramble into a planned supply line. That is the commercial logic behind the reported Broadcom discussions.

What does the $60 billion signal?

The scale matters more than the specifics. A figure of that magnitude implies multi-year volume commitments across multiple chip generations — not a single-product order. Moneycontrol's report frames the number as what Broadcom is seeking, so it should be read as a target under negotiation rather than confirmed contracted revenue.

It also signals how the economics of AI infrastructure have shifted. The largest compute buyers are no longer customers of chip companies in any conventional sense; they are co-investors in silicon programs, underwriting design and capacity years ahead of deployment. For Broadcom, a handful of such relationships now drives the majority of its AI revenue trajectory, and each new anchor customer compounds the foundry capacity commitments behind it.

How does this reshape the competitive picture?

Every major AI lab now weighs the same trade-off: buy merchant accelerators immediately, or commit to custom silicon that takes longer but locks in supply and cost curves. Broadcom sits on the custom side of that fork, and Anthropic — under pressure to expand computing power fast enough to stay competitive in frontier-model development — embodies the pull.

The competitive dynamics cut three ways:

• For incumbent GPU suppliers, each lab that shifts spend toward custom ASICs erodes a share of the largest AI capex pool.
• For Broadcom, winning Anthropic-tier customers converts its AI franchise from a two-customer story into a broader platform.
• For Anthropic, dedicated silicon is insurance against allocation risk — compute it can count on when rivals cannot.

The report does not specify process nodes, wafer volumes, or a timeline for the $60 billion figure, and any capacity details behind the arrangement remain undisclosed. What is confirmed is the direction of travel: AI labs are moving from renting compute toward owning the silicon supply chain beneath it.

Whether the full $60 billion materializes as contracted orders will depend on how quickly leading-edge foundry capacity frees up and how Anthropic's model-training roadmap justifies multi-year volume commitments — the two variables that will define pricing and allocation across the entire AI accelerator market.