Broadcom management meeting: AI inference demand surges, even exceeding current capacity, and is not reflected in current expectations

Broadcom's latest management meeting has released a clear signal—demand for AI inference is not only rapidly increasing but is also still in the early stages of an upward trajectory. Future reassessments of market size and capacity allocation may lead to systematic profit upgrades.

According to the Wind Trading Desk, during an investor meeting organized by JP Morgan recently, Broadcom's management revealed that the company is experiencing an unexpected surge in demand in the AI inference sector, even "exceeding current capacity," and this trend has not yet been incorporated into previous market size forecasts, potentially providing room for profit upgrades in the future. Meanwhile, non-AI businesses are also beginning to recover, with VMware continuing to scale up, and the company overall is "firing on multiple fronts."

Broadcom CEO Hock Tan and CFO Kirsten Spears stated that over the past year, AI demand has primarily come from training workloads, especially the training of "cutting-edge models." However, in the past two months, the company has seen a significant increase in orders for AI inference. Management pointed out that this wave of inference demand stems from customers wanting to monetize their AI investments more quickly, "the market capacity for inference may be severely underestimated."

AI Inference Demand Exceeds Expectations, Likely to Push Market Size Forecast Upper Limit

Broadcom's management indicated that the custom AI XPU chip business remains strong, with a clear growth runway. Over the past 9-12 months, AI demand has mainly focused on training workloads such as cutting-edge models. However, in the past two months, as customers have worked to monetize their investments, the company has experienced a significant surge in inference demand. Management believes that current inference demand has exceeded capacity.

This strong AI inference demand has not been included in its 2027 market size forecast (which predicts $60-90 billion for three existing AI customers), potentially representing an upside opportunity.

Broadcom is closely collaborating with four potential AI XPU customers and is confident in building a 1 million XPU AI cluster-type infrastructure for these customers. The company plans to complete the first generation of AI XPU product tape-outs for two major potential customers (which JP Morgan believes are Arm/SoftBank and OpenAI) this year.

At the technological forefront, Broadcom is leading the industry in transitioning to the next-generation 2nm 3.5D packaged AI XPU architecture and plans to complete the tape-out of the 2nm 3.5D AI XPU as scheduled this year.

AI workloads require higher-performance networking products, with a spending ratio of computing to networking at 3:1. Broadcom sees strong content opportunities in scale-up networking, which is valued 5-10 times higher than scale-out networking.

In terms of new technologies, the adoption of Co-Packaged Optical Devices (CPO) faces challenges due to the high failure rate of optical components (5-8%), which will require replacing the entire CPO solution rather than just swapping out pluggable optical modules.

Non-AI Business at Recovery Inflection Point, VMware Platform Transition Drives Continued Growth

After several quarters of cyclical pressure in non-AI semiconductor businesses (servers/storage, enterprise networking, general data centers, broadband), Broadcom is seeing a gradual "U"-shaped recovery in its business, reflected in the current booking/order situation JP Morgan believes this could drive a positive EPS revision cycle next year.

VMware is leveraging its cloud infrastructure (VCF) platform to provide a comprehensive solution stack, including hardware virtualization, management, and deployment, for large enterprise clients through software-defined data center solutions.

Management expects the VCF platform transition to maintain strong growth until the end of 2026/2027, at which point VMware's revenue scale is expected to reach approximately $20 billion in annualized revenue. Thereafter, growth is expected to return to a more normalized mid-to-high single-digit percentage level.

Margin Expansion Offsets Gross Margin Pressure

Despite the high demand for customized AI XPUs potentially putting pressure on gross margins (which are relatively low), Broadcom expects operating margins to continue expanding under the operational leverage of the model. For example, AI revenue is expected to grow by 60% year-over-year in fiscal 2026, but management anticipates that operating expenses will not grow at such a high rate. Equally important, the gross margin of AI network revenue (which accounts for about 30% of AI revenue) is above the company average, with strong operational leverage effects, and its growth rate is comparable to that of AI XPU products.

JP Morgan expects Broadcom's EPS for fiscal 2025 to grow by 38% year-over-year to $6.71, continuing to grow by 25% to $8.38 in 2026. Free cash flow is expected to reach $39 billion in 2026, with the net debt ratio decreasing from 0.9 in 2024 to 0.2 in 2026, indicating a solid financial position.

Management is highly focused on seven key AI XPU customer collaboration opportunities. They plan to deepen customer collaboration through reinvestment in the business rather than pursuing mergers and acquisitions that could distract attention and resources. Additionally, the company will continue to return 50% of free cash flow and prioritize debt repayment.

JP Morgan maintains an "Overweight" rating on Broadcom and sets a target price of $325, which is 16.9% higher than the current stock price. Broadcom's stock price has risen nearly 20% year-to-date.