Are AMD Stock’s AI Hopes Losing Steam?

AMD’s (NASDAQ:AMD) shares fell by over 6% in Friday’s trading after semiconductor and infrastructure software provider Broadcom reported blockbuster quarterly results and disclosed a $10 billion order for custom AI chips from a new client, widely reported to be OpenAI. The deal sent Broadcom stock up nearly 11%, sparking concerns over AMD’s positioning in the increasingly competitive AI hardware market. Nvidia (NASDAQ:NVDA)  stock also slipped about 3% on the news, though its well-entrenched dominance in GPUs and deep software ecosystem are giving investors less reason to panic. The bigger takeaway is that the AI hardware market itself may be entering a new phase – one where custom silicon begins to reshape infrastructure spending. GPUs have powered the boom in generative AI training thus far, but ASICs are emerging as more efficient alternatives for inference workloads.

Image by Pete Linforth from Pixabay

AMD vs Nvidia vs Broadcom

While AMD remains far behind Nvidia in the AI accelerator game, it has made some progress in recent months. Data center revenue, often viewed as a proxy for AI chip sales, rose 57% in Q1 to $3.7 billion, driven in part by strong uptake of Instinct GPUs. However, the following quarter told a different story, with revenue of $3.2 billion falling short of consensus estimates and guidance pointing to an 11% sequential decline and just about 14% year-over-year growth.

In comparison, Broadcom saw Q3 AI revenue growth accelerate to 63% year-over-year to $5.2 billion, while Nvidia’s data center sales surged 56% last quarter to $41.1 billion, over a much larger base. This has reinforced the view that AMD is still very much in the early innings of its AI race, with Nvidia controlling the lion’s share of training workloads and now Broadcom threatening to carve out share with its custom silicon. Nvidia Stock To Fall 50% As AI Cycle Turns?

For investors looking for growth opportunities while minimizing exposure to the high volatility of individual stocks, we encourage you to explore the High Quality Portfolio. Since its launch, this portfolio has shown an exceptional performance track record, consistently exceeding its benchmark—a mix of the S&P 500, Russell, and S&P MidCap indices—and achieving cumulative returns over 91%, as evident in HQ performance metrics.

GPU To Custom Chips

The AI landscape may be entering a turning point. The past three years have seen outsized capital poured into GPUs for training massive models, but this phase is beginning to look more saturated. Incremental performance gains from larger models are diminishing, while the supply of high-quality training data is becoming a binding constraint – much of the easily accessible data available on the Internet has already been consumed. That suggests the focus of AI spending could shift from training, which is front-loaded, toward inference, where models are deployed at scale and used continuously in real-world applications.

This is precisely where ASICs hold an edge. The parallels with cryptocurrency are striking: Bitcoin mining began on GPUs, but efficiency and cost imperatives quickly drove a wholesale migration to ASICs. The same logic applies to AI inference, where custom chips can outperform GPUs in both power consumption and cost-effectiveness. Broadcom appears well-positioned for this shift, with CEO Hock Tan signaling that hyperscale customers are increasingly adopting XPUs – the firm’s custom chip architecture – for a growing share of their compute footprint.

If the client behind Broadcom’s $10 billion order is indeed OpenAI, it would mark the most visible signal yet of diversification away from Nvidia GPUs for ChatGPT workloads. Such a pivot would likely have repercussions across the industry, encouraging other hyperscalers to hedge their reliance on GPUs and be more open to experimenting with custom silicon. OpenAI, which has historically run its workloads on Microsoft Azure using Nvidia or AMD GPUs, now could be shifting toward in-house kit with custom Broadcom silicon.

AMD Challenges

For AMD, this raises uncomfortable questions. Its challenge is now twofold: fending off Nvidia, whose GPU ecosystem and CUDA software remain deeply entrenched, while also countering the rise of ASICs that could blunt demand for general-purpose accelerators. The company’s MI300 series shows promise, but still lacks high-profile customer validation at scale. Meanwhile, Broadcom’s strength in other areas of the AI stack including networking, through its Tomahawk switches and interconnects, strengthens its hand, offering hyperscalers a bundled proposition. Now, AMD risks being relegated to the middle ground, squeezed by Nvidia at the high end, and ASIC providers at the efficiency frontier.

Even post the sell-off, AMD stock trades at about 40x estimated 2025 earnings. Revenues are projected to grow by about 28% this year, led in part by a recovery in its CPU business. In comparison, Broadcom commands an even richer valuation at roughly 49x forward earnings, despite consensus calling for slower revenue growth of about 23% this year. However, the premium multiple looks warranted given Broadcom’s accelerating momentum in AI. Its $10 billion custom silicon deal, combined with a 63% year-over-year surge in AI revenues last quarter and strength in networking interconnects, could provide a more durable growth narrative for investors.

The Trefis High Quality (HQ) Portfolio, which comprises 30 stocks, has a history of consistently outperforming its benchmark, inclusive of the S&P 500, Russell, and S&P midcap. What accounts for this? As a collective, HQ Portfolio stocks have delivered superior returns with reduced risk compared to the benchmark index; a smoother ride, as evident in HQ Portfolio performance metrics.

Invest with Trefis Market-Beating Portfolios

See all Trefis Price Estimates