Here's What Jim Cramer Is Saying About Semis in August

Cramer highlights semiconductors—especially the AI silicon complex—as earnings have reinforced a bifurcated setup of (1) merchant GPU compute (NVIDIA $81.61B Q1 FY2027 revenue, +85% YoY; AMD $11.54B Q2 revenue, +107% Data Center) and (2) custom ASIC/networking (Broadcom AI semis $10.80B Q2, +143% YoY; Marvell Q1 FY2027 revenue $2.418B, +28% YoY). Guidance and demand signals are broadly upbeat, with NVIDIA Q2 revenue guided to $91B at 75% gross margin and Broadcom Q3 AI semis guided to $16B (+200% YoY). The article frames this as an investor overweight “lane” into 2H26 rather than a new stock pitch, with sentiment skewing bullish based on strong growth and raised outlooks.

Analysis

This is no longer a broad semiconductor call; it is a ranking exercise. The market is rewarding names that monetize AI capex directly and punishing those still tethered to slower handset/consumer cycles. The second-order effect is a widening revenue-quality gap: merchant compute, custom silicon, and networking/optics should keep taking share of the same hyperscaler budget, while legacy exposure becomes a multiple trap even if absolute revenue is stable.

Near term, flow will likely stay constructive for NVDA, AVGO, and MRVL because these are the names investors can buy as a proxy for durable AI spend, but the upside is increasingly revision-driven rather than multiple-expansion-driven. NVDA is the cleanest center of gravity, yet also the most crowded; any slip in forward margin, shipment cadence, or capex commentary could hit the stock harder than fundamentals alone justify. AVGO has the best risk/reward if custom-silicon adoption keeps broadening, because it benefits from customers diversifying away from merchant GPUs without needing a single hero product.

The contrarian read is that QCOM may be more interesting as a value/trajectory debate than the market allows, but not yet as a rerating story. Until non-handset revenue can visibly offset phone cyclicality, every AI-positive tape likely compresses its relative multiple versus the rest of the group. Over 6-18 months, the key falsifier for the whole bullish complex is a slowdown in hyperscaler or sovereign AI spending; if that happens, the best-defended names should be the ones with recurring design-win content and the highest free-cash-flow conversion, not the highest headline growth.

Key Decisions for Investors

• Long AVGO / short QCOM for 2-4 months: best relative-value expression of AI monetization versus handset cyclicality; target 10-15% spread capture, stop if QCOM shows two consecutive quarters of non-handset acceleration or AVGO AI guidance misses.
• Buy NVDA only on pullbacks rather than chasing strength; 1-3 month catalyst path is continued estimate revisions, but the stock is crowded and vulnerable to any guide/margin disappointment. Use call spreads if you want convexity with defined downside.
• Add MRVL tactically as a higher-beta beneficiary of AI networking/optics for the next earnings cycle; smaller size than NVDA/AVGO because execution risk is higher, but upside is better if optical attach rates keep surprising.
• Hold AMD as a selective long, but demand proof that data-center mix can keep outrunning margin compression. Best entry is post-earnings weakness or a market-wide semi selloff; thesis breaks if compute growth decelerates or merchant GPU pricing turns competitive.
• Use SOXX as a hedge against being wrong on the breadth of the AI capex cycle only if hyperscaler commentary weakens; otherwise the index dilutes the strongest relative-value opportunities in AVGO/MRVL versus QCOM.