Broadcom, Intel CEOs warn capital could become AI's biggest bottleneck

When Broadcom and Intel's CEOs simultaneously flag capital as AI's binding constraint, the signal transcends corporate strategy—it marks a structural shift in what limits frontier compute. Technical cascade: Every 100bps rise in financing cost compresses a $1B data-center IRR by two to three points. This directly throttles HBM stacking and CoWoS advanced-packaging utilization. If downstream deployment slows, TSMC and Samsung will likely trim 2026 capex, shortening ASML's EUV order visibility and forcing the equipment chain into passive destocking. Compliance and risk: The rate cycle compounds export controls, creating a dual squeeze on non-US AI infrastructure—both technology access and capital access tighten simultaneously. Firms must now trade off compliance overhead against financing premiums, systematically compressing supply-chain redundancy. Market dynamics: Tighter capital actually favors Broadcom's custom-ASIC thesis. Under TCO pressure, hyperscalers prefer three-to-five-year locked contracts over chasing the latest GPU. Intel's window is closing; without a scaled PDK ecosystem by end-2025, its AI accelerator narrative risks further hollowing out. Forward look: Over the next eighteen months, AI infrastructure investment pivots from compute arms race to inference ROI. Mid-tier cloud providers accelerate consolidation; capital concentrates at the top. Inference energy-efficiency-per-token will replace peak FLOPS as the primary procurement metric. The bottleneck is no longer fabrication—it is the balance sheet.