Broadcom's AI chip revenue tripled while Applied Digital simultaneously locked in $36 billion of data center capacity, according to a Dealroom data roundup. The paired disclosures capture the twin pressure points now defining AI infrastructure deployment: chip supply at the top of the stack and power capacity at the bottom.

On the silicon side, Broadcom confirmed its AI chip revenue tripled over the comparable period.

Broadcom supplies custom AI accelerators — often referred to as XPUs — to hyperscale customers. It also sells merchant networking ASICs and Ethernet switch silicon. The threefold revenue jump positions Broadcom as one of the few vendors outside Nvidia moving meaningful volume at the AI training and inference layer.

Applied Digital, a specialist in large-scale data center operations that has pivoted into AI and HPC workloads, locked in $36 billion of capacity commitments. The figure marks one of the largest single commitments of pre-construction AI capacity publicly disclosed to date.

Applied Digital typically builds and leases entire buildings — and increasingly entire campuses — to anchor tenants under multi-year agreements that bundle land, power, cooling, and shell.

What do the two announcements together signal?

The $36 billion of locked-in capacity at Applied Digital reflects the procurement reality for hyperscalers. Compute demand now outpaces the speed at which utilities, grid operators, and developers can deliver gigawatts. Customers increasingly sign capacity agreements well before power is delivered, betting on multi-year silicon roadmaps rather than near-term spot capacity.

For Broadcom, the AI chip revenue triple demonstrates that custom accelerator programs have crossed the early-adopter threshold. Hyperscale customers that historically bought only merchant GPUs now operate dual-track silicon strategies, pairing Nvidia systems with second-source accelerators.

How do the contracts reshape the supply chain?

• Applied Digital's $36 billion commitment transfers financing risk from the developer to the anchor tenant. The structure lets the operator pre-pay for power equipment, transformers, and switchgear at a time when lead times for high-voltage equipment routinely exceed 18 months.
• Broadcom's revenue triple signals that foundry capacity at TSMC and advanced packaging at CoWoS lines will absorb another wave of demand exactly when Applied Digital brings gigawatts online.
• The pairing of $36 billion of capacity with tripled chip purchases tightens both ends of the stack simultaneously, leaving limited slack for late entrants.

What did the two companies disclose?

Broadcom reported only the headline revenue figure without segment-specific dollar amounts in the Dealroom summary. Applied Digital disclosed the $36 billion aggregate, but the Dealroom roundup did not enumerate individual counterparties, contract lengths, or site locations.

Industry observers will read the pair of disclosures as confirmation that AI infrastructure now moves on two-year, capital-intensive planning cycles rather than quarterly procurement cycles. Customers committing $36 billion to a single developer and tripling silicon purchases at a single vendor reflect a procurement model closer to long-cycle utility contracts than traditional enterprise hardware purchasing.

Die Signal will track the next quarterly disclosures from both companies for confirmation of the headline figures.