3 AI Chip Stocks to Buy in September

The article identifies Nvidia, AMD, and Broadcom as undervalued AI-infrastructure opportunities, citing substantial projected growth despite recent semiconductor-stock volatility. Nvidia trades at 14.5x fiscal-2028 estimates while projecting 70% revenue growth that year; AMD is more than 15% below its high and targets a $220 billion agentic-AI CPU market by 2030. Broadcom expects AI revenue to double to $115 billion in fiscal 2027 and again to $230 billion in fiscal 2028, while trading at roughly 19x fiscal-2027 and 12x fiscal-2028 consensus earnings.

Analysis

The actionable distinction is not “AI exposure” but monetization durability. NVDA retains the highest probability of sustaining system-level gross margins because networking, software, and rack-scale integration raise switching costs; its principal 6-18 month risk is that hyperscalers shift incremental inference workloads toward internally designed ASICs rather than merchant GPUs. That makes AVGO a credible hedge to NVDA concentration, but reported custom-silicon revenue aspirations should be treated as design-win pipeline, not contracted revenue: ASIC programs carry concentrated-customer, tape-out, yield, and deployment-timing risk that can create large estimate revisions.

AMD’s upside requires evidence that inference wins become volume deployments rather than technical partnerships. Its server-CPU franchise provides a cleaner agentic-AI beneficiary than its GPU business, but a CPU-per-GPU mix shift is unlikely to translate one-for-one into AMD revenue if ARM-based CPUs and hyperscaler in-house silicon take the incremental attach. Intel is the likely second-order loser if AI-driven server refreshes accelerate, yet its depressed expectations make it a poor outright short without confirmation that share losses persist in unit data.

Near term, this promotional-style bullish framing is unlikely by itself to change estimates; the relevant 1-3 month catalysts are hyperscaler capex commentary, custom-accelerator production disclosures, and AI networking order trends. The contrarian risk is that investors are extrapolating aggregate AI capex while ignoring customer concentration: a single spending pause by GOOG, META, or a frontier-model customer would impair AVGO’s custom-silicon multiple more sharply than NVDA’s diversified platform economics.

Key Decisions for Investors

• Maintain a core long NVDA versus a short semiconductor-beta hedge (SMH or SOXX), not a naked chase: add only on post-earnings volatility or a pullback that is not accompanied by reduced hyperscaler capex guidance. Thesis fails if data-center gross margin or next-generation system demand indicates material pricing pressure; target 6-12 months.
• Initiate a 6-12 month long AVGO / short AMD pair in equal dollar risk only after verifying custom-accelerator backlog, customer prepayments, and networking growth at the next earnings release. AVGO offers better exposure to hyperscaler ASIC diversification; exit if any major customer delays a production ramp or AVGO’s AI revenue guide is cut, as the concentrated-revenue downside can exceed the pair’s intended protection.
• Keep AMD on a catalyst watch rather than treating inference partnerships as revenue: go long only if management discloses named production deployments, MI-series supply commitments, or sustained server-CPU share gains. A practical trigger is a guide raise tied to AI GPU or EPYC demand; absent that, the stock remains vulnerable to multiple compression if NVDA pricing remains rational.
• Avoid shorting INTC solely on AI-server displacement. Use INTC as a tactical short only against long AMD following evidence of sequential server share loss or a data-center guidance reset; cover on foundry subsidy/newsflow or a credible product-execution milestone, which can drive sharp bear-market rallies.