Should You Forget Nvidia and Buy These 2 Artificial Intelligence (AI) Chip Stocks Instead?

The article sees AMD and Cerebras as potential long-term AI-chip outperformers relative to Nvidia, citing AMD’s market capitalization at less than one-fifth of Nvidia’s and its $100 billion deals with each of OpenAI and Meta. Cerebras, valued below $40 billion, says its systems are about 30 times faster than GPUs, and has a $10 billion OpenAI deal to supply 750 megawatts through 2028, but its technology remains niche and must gain mainstream adoption. The outlook is positive but speculative, especially for Cerebras; Nvidia is still described as a strong option.

Analysis

The investment question is not whether inference grows, but who captures its economics. Faster decode can support premium-priced specialist systems, yet total cost per useful token—including power, utilization, cooling, and software integration—may matter more than headline latency. Cerebras’s benchmark claims therefore do not establish a broad displacement case; a large customer commitment is not equivalent to sustained, diversified utilization or attractive returns on deployed capacity.

AMD has the more credible risk-adjusted challenger setup, but customer commitments and a large addressable market are not revenue, margins, or market share. Winning through price cuts could lift shipments while weakening the earnings case. Its hybrid work with Cerebras could also broaden AMD’s offering rather than simply transfer workloads away from Nvidia. Conversely, any faster-than-expected inference shift would pressure Nvidia’s pricing and mix at the margin, not automatically overturn its software and systems advantage.

Over 1–3 months, watch AMD data-center revenue, accelerator gross-margin commentary, and evidence that announced customer programs are converting into shipments. Over 6–18 months, HBM availability, advanced packaging capacity, and customer concentration may be as decisive as chip performance; shortages could constrain AMD’s growth while reinforcing incumbent supply-chain access. The article’s forward valuation comparison is not enough to establish relative cheapness without checking estimate dates, earnings quality, and execution assumptions. The contrarian risk is treating inference growth as a single-winner market: workloads may remain split among GPUs, CPUs, and specialized systems, limiting both Cerebras’s upside and any simple Nvidia-to-AMD rotation.

Key Decisions for Investors

• Keep NVDA as the core exposure; do not short it on market-cap size or an inference narrative alone. Reassess if accelerator pricing or data-center margins weaken alongside evidence of sustained workload substitution.
• Put AMD on a relative-value watchlist. Consider a staged, market-neutral long AMD / short NVDA only after quarterly results show accelerator revenue converting without material margin deterioration; falsify on slowing data-center growth, discount-led share gains, or HBM/packaging constraints.
• Treat CBRS as a high-volatility event-driven watch, not a proven mainstream inference trade. Verify contract terms, deployment timing, utilization, customer concentration, and power economics before sizing; missed deployments or continued reliance on Nvidia for production workloads would weaken the thesis.
• Track customer capex and supply-chain conversion over the next 1–3 quarters: shipment visibility, HBM allocation, and advanced-packaging capacity are more actionable signals than announced deal values or vendor speed claims.