3 AI Chip Stocks to Buy in September

The article highlights Nvidia, AMD and Broadcom as attractive AI-infrastructure investments, citing growth prospects and relatively low forward valuation multiples. Nvidia trades at 14.5x estimated fiscal 2028 earnings while projecting 70% revenue growth that year; AMD is more than 15% below its high and targets a $220B agentic-AI CPU opportunity by 2030. Broadcom forecasts AI revenue will double to $115B in fiscal 2027 and double again to $230B in fiscal 2028, while trading at about 12x fiscal 2028 consensus earnings.

Analysis

The actionable distinction is not “AI winners” but monetization timing. NVDA retains the highest near-term earnings visibility because its systems, networking and software stack capture a larger share of deployed cluster spend; however, that also leaves it most exposed to a hyperscaler capex digestion signal over the next 1-3 quarters. AVGO’s custom-silicon pipeline is a credible medium-term alternative, but revenue conversion depends on customer tape-outs, packaging capacity and production yields—milestones that are materially less visible than merchant-GPU shipments and can shift by quarters.

AMD offers the more asymmetric operational setup if inference and agentic workloads broaden CPU attach rates, because incremental server-CPU share gains can support earnings even if GPU share remains modest. The key risk is that inference becomes increasingly ASIC-led, limiting AMD’s accelerator margin pool; Intel’s weak competitive position is a secondary beneficiary only if enterprise buyers prioritize supply diversity or x86 pricing over performance. The article’s forward valuation comparisons should not drive positioning: estimates several fiscal years out are highly sensitive to capex assumptions, and purported customer/product claims require confirmation in company filings, earnings calls, or customer disclosures.

Consensus remains too focused on aggregate AI spend and underweights the likely value transfer from general-purpose training hardware to networking, memory bandwidth, advanced packaging, and customer-specific silicon. That shift is structurally favorable to AVGO over 6-18 months, but it may be temporarily negative for all semiconductor multiples if hyperscalers signal lower 2027 capex growth. This is low-immediacy commentary rather than a stand-alone catalyst; the next earnings cycles, hyperscaler capex guidance, and foundry/packaging lead-time data should determine entry timing.

Key Decisions for Investors

• Maintain an overweight in NVDA for the next 1-3 months only while hyperscaler capex guidance remains intact; hedge with a defined-risk NVDA put spread spanning the next earnings date. Reduce if aggregate GOOG/META cloud capex guidance decelerates materially or if NVDA data-center gross-margin guidance misses by more than 200 bps.
• Initiate a 6-12 month pair trade: long AVGO / short NVDA in equal semiconductor-beta dollars after verifying custom-accelerator backlog and packaging supply in AVGO’s next report. The thesis is mix migration toward ASIC/networking; stop out if AVGO AI revenue/backlog conversion is delayed by two quarters or NVDA sustains data-center revenue growth materially above consensus.
• Keep AMD on a watchlist rather than chase momentum; enter on post-earnings confirmation of both server-CPU share gains and accelerator gross-margin expansion. A suitable risk framework is a 9-12 month call spread financed with an out-of-the-money put only after management provides quantified inference revenue or backlog, since current upside otherwise rests on unverified adoption assumptions.
• Avoid using INTC as a direct AI catch-up long. A more defensible tactical expression, if enterprise CPU pricing weakens, is short INTC versus long AMD; cover if Intel demonstrates sustained server-share stabilization or receives a foundry/customer win that changes external-manufacturing utilization economics.