Why AMD and Amazon Could Be Better AI Growth Bets Than Nvidia

Nvidia's current investment case is under pressure from two fronts: stretched valuation and a structural shift in chip demand. The company's dominance is undeniable, but its premium price may already be pricing in near-perfect execution for years to come.

A discounted cash flow analysis suggests a significant gap between price and perceived value. Using earnings as a proxy, the GuruFocus DCF model calculates an intrinsic value of $125 for Nvidia, implying a current price of $185.41 is deeply overvalued. Even using a traditional free cash flow approach, the intrinsic value drops to $98.31. This creates a stark margin of safety problem, leaving little room for error in the company's growth trajectory.

The more profound threat is a long-term erosion of profit margins. Major cloud providers, the backbone of Nvidia's sales, are actively building alternatives. Amazon, for instance, has installed 1.4 million of its Tranium2 AI chips in its data centers and is monetizing them aggressively, with custom chip revenue hitting a $10 billion annual run rate and growing over 100% annually. The company claims its chips offer 30% to 40% improved performance-per-dollar over comparable GPUs. This trend is not isolated; Google, Meta, Microsoft, and OpenAI are all developing custom ASICs. As these chips mature, they could pressure Nvidia's pricing power and reduce its long-term market share.

Yet Nvidia's current software ecosystem lock-in remains a formidable barrier. Despite technically competitive hardware from rivals like AMD, Nvidia's market share stays near 90% while competitors struggle to gain more than 5-8%. This is because customers are buying an entire stack of tools and expertise, not just silicon. The deep integration of CUDA and Nvidia's developer network creates a switching cost that challengers must overcome. For now, this lock-in protects Nvidia's cash flows. But it also means that any alternative must offer not just better specs, but a superior total solution to displace it.

The bottom line is that Nvidia's growth story is becoming more expensive to own and more vulnerable to internal disruption. The valuation suggests the market has little patience for a stumble, while the rise of custom chips signals a potential long-term margin compression. This sets the stage for investors to look at companies like AMD and Amazon, which may offer more attractive entry points into the AI infrastructure wave.

AMD: Scalable Growth with a Competitive Architecture

Advanced Micro Devices presents a compelling case for scalable growth, powered by robust financial momentum and a strategic push toward open infrastructure. The company's top-line trajectory is undeniable, with revenue surging 34% last year to $34.6 billion. This expansion, driven by its data center and client segments, has been accompanied by disciplined cost management, boosting net income to $4.3 billion. Analysts project continued acceleration, with revenue growth forecast at 34% in 2026 and 37% in 2027. This financial scalability provides a solid foundation for investing in the next generation of AI hardware.

The key to capturing market share, however, lies beyond raw performance specs. AMD is betting heavily on an open architecture strategy, exemplified by its "Helios" rack-scale system. Unveiled at the Open Compute Project summit, Helios is built to Meta's new Open Rack Wide standard and is engineered for frontier AI workloads. It leverages the next-generation MI450 Series GPUs to deliver generational leaps in performance and memory capacity. This initiative is a direct play for hyperscalers and enterprises seeking a future-proof, interoperable platform that avoids vendor lock-in. By aligning with open standards from the silicon up to the rack, AMD aims to build an ecosystem that competes with Nvidia's proprietary stack.

Yet the critical challenge remains software ecosystem lock-in. Despite technical parity on benchmarks-where AMD's MI355X chip is seen as just as good as Nvidia's Blackwell-market share has only grown to 5-8%. This gap exists because customers are buying an entire software and developer ecosystem, not just silicon. The switching cost from Nvidia's CUDA, which has been cultivated for nearly two decades, is immense. Rewriting code and retraining engineers can take months and cost millions. For now, this lock-in protects Nvidia's dominance. But it also creates a vulnerability that AMD's open approach is designed to exploit. If Helios gains traction and attracts a critical mass of developers, it could lower the switching cost for new AI deployments, opening a path for AMD to capture a larger slice of the expanding market. The growth thesis is intact, but its realization hinges on converting technical capability into ecosystem adoption.

Amazon: A Hidden $10 Billion+ AI Growth Engine

Amazon's AI story is a masterclass in vertical integration, where the company is both a colossal customer and a rising chipmaker. This dual role creates a unique growth trajectory that is often overlooked. While investors focus on Amazon's massive capital expenditure plans, the real engine is its in-house chip business, which has quietly become a multi-billion-dollar operation.

The scale of Amazon's custom Trainium AI chip deployment is staggering. The company has installed 1.4 million of its Tranium2 AI chips in its data centers, and the financial results are spectacular. Revenue from these custom chips has reached an annual run rate of $10 billion and is growing by more than 100% annually. This isn't just a side project; it's a core business that is expanding faster than the company's overall cloud revenue. The success is driven by clear economic advantages, with Amazon claiming its chips deliver 30% to 40% improved performance-per-dollar over comparable GPUs. This efficiency is already attracting major AI developers, with Anthropic using Tranium2 chips to train its next-generation models.

This growth is being fueled by an unprecedented capital investment. Amazon is ramping up its capital expenditures to $200 billion this year, a move CEO Andrew Jassy framed as monetizing capacity as fast as it is installed. This aggressive build-out is the fuel for the Trainium business. Every Tranium chip deployed represents a GPU that Amazon doesn't need to buy from Nvidia, directly reducing its reliance on the dominant supplier. The company's next-generation Tranium3 is already in high demand, with capacity expected to sell out by mid-2026.

The strategic leverage here is immense. By developing its own chips, Amazon is not only cutting its own costs but also building a powerful counterweight to Nvidia's dominance. As more hyperscalers like Google, Meta, and Microsoft build custom ASICs, the long-term profit margin pressure on Nvidia intensifies. Amazon's in-house development reduces its exposure to Nvidia's pricing power and could eventually allow it to offer more competitive AI services to its customers. For investors, this means Amazon is positioned to capture growth from two fronts: the explosive revenue from its own custom chip business and the massive, capital-intensive build-out of its AI infrastructure, all while using its chips to lower costs and improve margins.

Catalysts, Risks, and Investment Implications

The path for AMD and Amazon to challenge Nvidia is defined by a clear set of catalysts and risks. The near-term catalyst for AMD is the performance and customer adoption of its MI450 chip, which is seen as a potential validation of its growth thesis. This chip, part of the Helios rack-scale system, is designed to compete directly on frontier AI workloads. Its success will be measured not just by benchmarks but by early orders from hyperscalers and enterprises. The market will be watching for signs that the open architecture strategy can convert technical capability into real deployments, which could help the stock recover from its recent sell-off following a guidance miss that triggered a 17% drop in February.

The critical long-term risk for both AMD and Amazon is software ecosystem lock-in. Despite technically competitive hardware-AMD's MI355X is seen as just as good as Nvidia's Blackwell-market share has only grown to 5-8%. This gap exists because customers are buying an entire software and developer ecosystem, not just silicon. The switching cost from Nvidia's CUDA, which has been cultivated for nearly two decades, is immense. For Amazon, while its Tranium chips are already being used by AI developers like Anthropic, building a parallel software stack is a multi-year challenge. This lock-in protects Nvidia's dominance and limits the immediate threat from rivals, regardless of hardware performance.

The most profound structural threat, however, comes from the rise of custom ASICs from hyperscalers like Amazon and Google. These chips, designed in-house, are smaller, cheaper, and more focused. Google's TPUs are considered leaders, and analysts see custom ASICs "growing even faster than the GPU market over the next few years." This trend represents a long-term pressure on Nvidia's profit margins. As more companies build their own chips, they reduce their reliance on Nvidia's expensive GPUs, directly cutting into Nvidia's sales and pricing power. Amazon's own $10 billion annual run-rate custom chip business, growing over 100%, is a prime example of this shift in motion. The bottom line is that while AMD and Amazon offer more attractive entry points than Nvidia, their ability to capture market share is constrained by software, and their growth is ultimately fueled by the same trend that threatens Nvidia's long-term margins.