Why These Semiconductor Stocks Could Surge 50% in the Next 12 Months

The article argues AMD and Broadcom each have potential for more than 50% upside over the next year, driven by AI infrastructure spending and a shift toward inference and custom chips. AMD is highlighted for inference and agentic AI demand, with a path to roughly $175 billion in revenue and about $50 in EPS by 2030, while Broadcom is projected to exceed $100 billion in custom chip revenue in fiscal 2027 and could reach $180 billion in AI sales by 2028. The piece is bullish on both names but is largely forward-looking commentary rather than new company-reported results.

Analysis

The important second-order read-through is that AI capex is no longer just a GPU story; it is becoming a broader memory, networking, and CPU monetization cycle. That matters because it widens the beneficiary set from the obvious accelerator winner to the “picks and shovels” layer that gets paid on every rack expansion, regardless of which model ultimately wins. The market is still underappreciating how inference-heavy workloads compress the gap between compute and system design, which should keep ordering power strong for suppliers that can bundle silicon, interconnect, and software-enabled stickiness.

AMD’s setup is more about mix than headline TAM. If inference and agentic workloads force a more balanced CPU/GPU architecture, the company’s earnings power becomes less dependent on single-node accelerator share and more dependent on being embedded in the full server bill of materials. The risk is that the path to those economics is lumpy: design wins can create a revenue cliff several quarters later if deployment schedules slip, and a lot of the implied upside requires both pricing discipline and sustained supply tightness rather than just unit growth.

Broadcom is the cleaner expression of the structural trade because custom silicon can scale even if hyperscalers keep diversifying away from merchant GPUs. The underappreciated issue is concentration: a handful of customers can create very large numbers, but they can also negotiate hard once their internal chip teams mature, so the market may be overextrapolating the 2027-2028 run-rate. In other words, the near-term setup is strong, but the valuation debate will shift from growth to customer control and margin durability once the buildout phase peaks.

The main contrarian is that consensus may be too linear on AI infrastructure spend. If inference efficiency improves faster than expected, capital intensity per token could fall, which would not kill demand but could delay the revenue cadence for the whole supply chain. That creates a tactical window where the stocks can keep grinding higher on booking momentum, while the medium-term risk is multiple compression if investors start discounting a second-half 2026 digestion phase.

Key Decisions for Investors

• Stay long AVGO versus AMD as the cleaner way to express AI infrastructure durability; use a 6-12 month horizon and expect lower earnings volatility, but trim if customer concentration headlines intensify.
• Buy AMD on pullbacks only, ideally after confirmation of next-quarter deployment timing; treat it as a higher-beta execution trade with attractive upside if inference ramps, but tighter risk control if ramp schedules slip.
• Pair trade: long AVGO / short NVDA to isolate custom-ASIC and networking share gains versus merchant-GPU intensity; thesis works best over 3-9 months if hyperscalers continue diversifying supplier mix.
• Own GOOGL as an indirect beneficiary of TPU adoption and internal AI stack monetization; this is a lower-beta way to participate if custom silicon gains adoption without paying the full semiconductor multiple.
• Consider short-dated call spreads on AMD and AVGO into catalysts rather than outright stock if you want defined risk; the trade benefits from continued capex optimism while limiting downside if the market starts worrying about 2026 digestion.