Broadcom's Artificial Intelligence (AI) Chip Revenue Just Rose 221%. There's More Growth Coming.

Broadcom said AI semiconductor revenue is expected to reach $115 billion in 2027 and double again to $230 billion in 2028, supported by contracted component supply and demand from hyperscaler customers including Alphabet, Meta, Anthropic, and OpenAI. The company grew AI semiconductor revenue 221% in fiscal Q3 2027, while trailing-12-month total revenue was $89 billion. If achieved, the outlook implies a substantial shift toward AI chips and potential revenue tripling over roughly 2.5 years; Broadcom trades at 31x forward earnings, according to the article.

Analysis

The key valuation question is not whether AVGO can ship contracted capacity, but whether hyperscaler commitments represent durable workload demand rather than pre-emptive capacity reservation. Custom silicon economics improve materially once volumes clear design and NRE amortization thresholds, which could lift AVGO's mix, gross-margin durability and earnings multiple; however, the same concentration creates asymmetric downside if one or two programs slip. The market should discount management's multi-year revenue framework until customer capex, deployed-cluster utilization and ASIC attach rates independently corroborate it.

Second-order beneficiaries are TSMC (TSM), advanced packaging and HBM suppliers—particularly MU and SK Hynix—because custom accelerators still consume leading-edge wafer, CoWoS and memory capacity. Conversely, the incremental ASIC mix is strategically adverse to NVDA at the margin, but only for stable inference/training workloads; rapidly changing models favor programmable GPUs and limit substitution. The more relevant near-term competitive pressure may be on MRVL, whose custom-silicon narrative requires comparable hyperscaler design wins but lacks AVGO's scale in networking and supply-chain leverage.

Over the next 1-3 months, AVGO's upside requires evidence that AI revenue converts to backlog, margins and cash flow rather than merely extending a long-dated target. Over 6-18 months, a hyperscaler capex reset, lower-than-expected utilization, packaging/HBM bottlenecks, or a shift toward more flexible GPU architectures could drive both estimate cuts and multiple compression. Consensus may underappreciate that a successful ASIC ramp can simultaneously reduce NVDA's unit share without reducing total AI infrastructure spending, making this a mix trade rather than a broad AI-demand short.

Key Decisions for Investors

• Maintain a measured long AVGO versus short MRVL over a 6-12 month horizon; AVGO has superior customer access, networking pull-through and manufacturing scale, while MRVL is more exposed to design-win timing. Size as a relative-value position, with a stop/review if AVGO fails to convert AI growth into sequential gross-margin expansion or MRVL discloses a material incremental hyperscaler program.
• Do not chase AVGO on the article's long-range framework alone. Add only following the next earnings release if disclosed AI backlog, customer commitments and supply availability support FY2027 revenue visibility; target a 15-20% upside on estimate revision/multiple support against a 10% downside if guidance merely reiterates rather than raises.
• Use TSM as a lower-beta supplier expression of sustained custom-ASIC demand over 6-18 months, subject to monitoring leading-edge utilization and advanced-packaging capacity. The thesis is falsified by a meaningful cut to hyperscaler capex guidance or evidence that ASIC deployments are displacing, rather than expanding, total compute demand.
• Keep NVDA as a hedge rather than an outright short: buy NVDA calls or maintain a smaller NVDA long against an AVGO overweight if model cadence remains rapid. A GPU-led demand reacceleration would weaken the AVGO-share-gain narrative while preserving overall AI infrastructure exposure.