Why Broadcom Is Lending Anthropic $42 Billion to Fund Custom AI Chips

Broadcom has reportedly agreed to provide Anthropic with up to $42 billion in financing for AI training infrastructure and custom chip development, a deal that, if confirmed, would make the AI lab Broadcom's largest custom silicon customer.

Broadcom has reportedly agreed to extend Anthropic a credit facility of up to $42 billion, earmarked for AI training capital expenditure and custom chip development.

Broadcom's standing as a custom-silicon destination has been built over years of co-design partnerships with a small number of hyperscale clients. In its most recent quarter, AI chip revenue grew 221% year-over-year, with the company reporting $16.7 billion in AI engine revenue — a pace that has drawn sustained analyst attention to whether Broadcom's valuation reflects a temporary cycle or a structural shift. OpenAI's selection of Broadcom to develop its first custom AI accelerator, with Broadcom retaining a key role in that programme even after OpenAI negotiated a separate manufacturing arrangement with Samsung, illustrates how entrenched these co-design relationships become once the initial investment of design cycles is sunk. One report had already characterised Anthropic as Broadcom's largest potential custom chip engagement; the current financing deal, if confirmed, would formalise that framing.

The strategic logic behind the credit facility aligns with a thesis Macquarie analysts had outlined: they identified approximately $40 billion in capital opportunity linked to Anthropic. Broadcom's chief executive, asked about Anthropic's own CEO having publicly called for a slowdown in AI development, stated that the company's AI revenue targets remain unchanged — targets that include $115 billion in total revenue and $230 billion in AI-related revenue by 2028, with power delivery identified as the principal bottleneck. In September, Broadcom's joint venture with Toppan opened an advanced substrate manufacturing plant in Singapore — a critical component for high-density AI chip packaging — extending the company's reach further up the supply chain that its custom designs depend on.

The current picture is not without meaningful risk. China's audit of Broadcom's dominant position in state data-centre networking raises the prospect of formal restrictions on foreign networking chip suppliers, and export controls have already affected the company's AI chip business in that market. On the competitive front, Broadcom is challenging Nvidia's dominant position in AI networking with its Tomahawk switch ASICs; coverage notes that Broadcom's $16.7 billion AI engine faces margin pressure from intensifying competition in that segment. A warning from GPU cloud provider CoreWeave — which reported it cannot scale custom chip deployments — serves as a reminder that even well-funded, committed custom-silicon programmes can encounter execution friction on the way to volume.

Broadcom's own capital discipline frames the picture in a useful way: Slicast's compilation of SEC XBRL data shows FY2025 capital expenditure at $623 million on $63.89 billion in revenue, a 1% intensity that ranks last among the fifteen chip peers tracked. This reflects the fabless model, where foundry partners absorb manufacturing capex, but it also underscores that the $42 billion Anthropic credit facility is a financial structure, not an industrial commitment — and carries its own distinct category of risk. Three concrete signals will shape the near-term narrative: whether binding terms on the Anthropic deal are disclosed and what supply or exclusivity agreements accompany the financing; how China's audit of state data-centre networking infrastructure concludes and whether it produces formal restrictions on foreign chip suppliers; and whether Broadcom's margin in the AI networking segment holds as Nvidia contests the Ethernet interconnect layer. The company has assembled an unusually durable custom-silicon franchise; the Anthropic transaction will test whether deepening that franchise into the balance sheets of major AI labs strengthens it or introduces a new class of exposure.