Nvidia owns the headlines and the GPUs, but smart investors are scanning AMD, Intel, cloud giants and chip specialists for the next outsized AI gains — and the traps to avoid.
Illustration by IMF Alpha editorial · Reviewed by Pedro Marini
Nvidia is the easy headline — it powers most large-language models and sits as the favorite in countless portfolios. That concentration creates a familiar itch for investors: if one company is carrying the trade, where do you put the rest of your capital without simply buying more NVDA?
This is not about disliking the leader. Think of it as a practical tour of alternatives that actually matter, why they could benefit from the AI cycle, and the structural risks that can turn promising chips into paper losses.
Why alternatives matter
AI demand is structural but not monolithic. Data centers need raw GPU compute, cheaper accelerators for narrow tasks, and plenty of memory and networking to keep clusters humming. That opens multiple vendor entry points — not only Nvidia.
A different metaphor: GPUs are the lead violin, sure. But an orchestra still needs percussion, brass, and a conductor to make the piece work. Missing any of those parts limits the sound.
Who’s worth watching (and why)
AMD — Competes on both GPUs and server CPUs. Its Instinct accelerators and EPYC processors give it a two-pronged play: win on price-performance with GPUs and use CPUs to bundle into hyperscaler deals. That dual footprint matters more than most headlines admit.
Intel — The comeback nobody wants to count out. Between custom accelerators, the Habana technology, and deep enterprise relationships, Intel moves slower but with strategic depth. Execution and process gains are the make-or-break.
Marvell — Easy to overlook. Networking and interconnect chips are the plumbing of AI clusters. Faster switches and smarter NICs can materially increase effective GPU throughput, which is a subtle but real lever.
Micron — AI eats memory. Improvements in HBM and DRAM density influence how long you can train big models and how quickly you can serve them. Micron sits squarely in that bottleneck.
Cloud and software owners (Amazon, Microsoft, Google) — Lower-risk ways to own AI adoption. They build custom silicon, lock in enterprise customers, and monetize models through services instead of just selling hardware.
What could go wrong
Valuation shock. Lots of alternative names assume perfection. One missed quarter or a delayed accelerator launch will remove optimism fast.
Foundry and supply-chain constraints. Smaller designers rely on external fabs. If TSMC capacity tightens, the smaller players get squeezed first.
Hyperscaler self-sufficiency. Amazon, Google, Meta — they’ll keep designing in-house accelerators. Over time that can shrink third-party total addressable markets for GPUs and other accelerators.
Short, practical strategies
Diversify around the leader. Keep a core NVDA position for upside, but add smaller stakes in AMD, Intel, and Marvell to capture different parts of the stack.
Consider ETFs for execution risk. Semiconductor ETFs smooth single-stock volatility and give sector exposure without the burden of perfect stock-picking.
Monitor revenue exposure and customer concentration. Companies heavily dependent on one hyperscaler are higher risk if that partner decides to move in-house.
Favor durable moats. IP in network fabric, memory processes, and enterprise CPU relationships often outlast a momentary specs lead.
A quick history check
Hardware cycles teach a blunt lesson: performance leadership can slip fast when execution falters. The crypto-driven GPU rush showed how demand can be transient; telecom capex booms turned into long troughs. Treat the AI boom as meaningful, but expect cyclical potholes along the way.
The upshot
Nvidia is the quickest way to bet on AI compute, but it doesn't have to be the only way. A balanced approach — selective single-stock exposure, some sector ETF coverage, and close attention to partnerships and supply signals — gives participation in the AI story while limiting single-company disaster risk. The future of AI compute is layered; often, the plumbing outlives the flashy hero.
Pedro Marini
OpenAI's enterprise revenue has reportedly surpassed $2 billion annually, signaling rapid adoption of its AI services by businesses and solidifying its market position.
Recent fintech earnings reports emphasize the critical role of payment processing volumes and the emerging impact of AI-driven underwriting models on profitability.
Asset managers and hedge funds are quietly building proprietary data lakes to train in-house AI — reshaping competitive moats, privacy risks, and market structure.