Wall Street Is Wrong About These 2 Chip Stocks

AMD reported Q2 2026 revenue of $11.5B, up 50% year over year and above roughly $11.3B consensus, while non-GAAP EPS rose 245% to $1.66 versus $1.62 expected; data-center revenue surged 107% to $6.7B. Broadcom posted Q3 sales of $29.6B, up 86%, and non-GAAP EPS of $3.32, up 96%, both ahead of estimates, while management projects AI revenue will double to $100B in 2027. The article argues recent share volatility reflects concerns over an AI-spending slowdown that are inconsistent with projected global AI-infrastructure spending of more than $1T next year, versus $750B this year, and highlights AMD's CPU opportunity and Broadcom's estimated 60% ASIC market share.

Analysis

The investable distinction is not simply AI demand durability but where hyperscalers direct incremental capex. AVGO benefits when large customers shift mature, high-volume workloads from merchant GPUs to internally optimized accelerators; that shift is structurally negative for NVDA's unit economics at the margin, even if total AI silicon spend continues rising. The bottleneck also migrates toward networking, memory bandwidth, advanced packaging and power delivery, creating secondary beneficiaries in ANET, MRVL, TSM and VRT rather than a uniform semiconductor uplift.

AMD's CPU upside is more conditional than the bullish framing implies: agentic deployments can raise host-CPU, memory and networking requirements per accelerator, but improved inference efficiency could reduce total server count. AMD needs sustained server-CPU share gains and accelerator gross-margin improvement simultaneously; revenue growth alone will not support a premium multiple if MI-series pricing is used to win deployments. The next 1-3 months hinge on customer-specific accelerator commentary, 2027 supply commitments and gross-margin guide trajectory, while the 6-18 month issue is whether enterprise inference expands beyond a handful of hyperscaler buyers.

Consensus is likely too binary on an AI-capex "bubble." A capex digestion phase would not affect AVGO and AMD equally: custom-chip programs are harder to cancel once taped out, but they are exposed to a small number of customers reprioritizing programs; AMD is more exposed to competitive pricing and product execution. A broad risk-off move can therefore create an attractive AVGO entry, but a generic AMD dip is not sufficient without evidence that its data-center mix is improving rather than merely scaling at lower incremental margins.

Key Decisions for Investors

• Prefer long AVGO / short NVDA as a 3-6 month relative-value expression after confirming both names' next guidance updates: custom silicon share gains can compress NVDA's long-duration premium without requiring an AI-spending downturn. Target a 10-15% relative move; exit if NVDA reports re-accelerating hyperscaler GPU demand with stable pricing or AVGO discloses a major customer program delay.
• Maintain AMD as a watch-list long, not a dip-buy mandate. Initiate only if the next earnings release shows data-center gross-margin expansion and explicit evidence of server-CPU share gains; use a 6-12 month horizon and size smaller than AVGO because execution risk remains higher. Falsification: flat-to-down data-center margin, delayed accelerator ramps, or guidance that relies on lower pricing to preserve unit growth.
• Add a basket hedge against an AI capex digestion event through selective short exposure to high-multiple AI infrastructure beta or a long SOXX puts structure around major hyperscaler earnings. The key near-term catalyst is whether GOOG and META capex guidance converts into deployed compute demand rather than longer lead-time commitments; a reduction in 2027 capex plans would hit AMD faster than AVGO's multiyear design pipeline.
• For second-order exposure, favor ANET and VRT over adding broad semiconductor beta if evidence emerges that inference clusters are scaling. Their revenue sensitivity is to network fabric and data-center power density, respectively, and they are less directly exposed to the merchant-GPU versus ASIC winner selection.