Full Breakdown
Broadcom and Micron Ride the AI Inference Wave
8/26/2026, 11:15:20 AM
Background & Context
The semiconductor market is shifting from the massive compute needs of AI model training toward the sustained demand of inference and emerging “agentic” AI workloads. This transition is boosting demand for two complementary components: custom accelerators and networking silicon that connect hyperscaler clusters (Broadcom’s focus) and high-bandwidth memory (HBM) that supplies those clusters (Micron’s focus). Analysts note that the capital allocation toward inference-oriented hardware is now driving a new growth cycle for both firms.
Performance Highlights
- Broadcom reported fiscal Q2 revenue of $22.2 billion, a 48% year-on-year increase, with AI semiconductor revenue of $10.8 billion, up 143%.
- AI semiconductor orders surpassed $30 billion in the quarter, covering programs for Google TPUs, Meta MTIA XPUs, OpenAI silicon, and Anthropic compute capacity.
- Micron posted fiscal Q3 revenue of $41.5 billion and a consolidated gross margin of 84.9%, with non-GAAP earnings per share of $25.11.
- DRAM prices rose in the low-60 percent range sequentially, and Micron’s HBM4 12-high is ramping twice as fast as its HBM3E predecessor after generating over $1 billion in HBM4 revenue.
- Both companies have secured long-term contracts: roughly $100 billion of contracted AI revenue and $22 billion in cash deposits and letters of credit are intended to smooth memory cyclicality.
Official Statements & Responses
Broadcom CEO Hock Tan indicated that networking accounted for almost 40% of the company’s AI revenue in Q2 and projected fiscal 2027 AI silicon revenue to exceed $100 billion, extending visibility into 2028. Micron CEO Sanjay Mehrotra highlighted the rapid ramp of HBM4 12-high and emphasized that the company’s take-or-pay agreements lock in customers for five years at floor prices, positioning Micron to maintain margins “well above” historic peaks.
Why It Matters
The divergent business models—Broadcom’s fabless, custom-accelerator and networking approach versus Micron’s memory-manufacturing focus—are both benefitting from the same inference-driven tailwind. Broadcom’s diversified silicon portfolio offers durability, while Micron’s exposure to memory scarcity provides asymmetric upside. Together, they illustrate a “picks-and-shovels” narrative where infrastructure providers capture the bulk of AI compute spending as enterprises move from experimental training to production-scale inference.
What’s Next
Broadcom’s next catalyst is the execution of 10 gigawatts of fiscal 2027 shipments and early deliveries to Meta MTIA. Micron’s upcoming guidance is expected to target $50 billion in revenue with a gross margin near 86%, alongside qualification of the HBM4E product. Investors will watch whether Broadcom’s networking mix stays near 40% of AI revenue as custom accelerator dollars scale, and whether Micron’s take-or-pay agreements hold if DRAM spot prices soften into 2028.
