A manufacturing bottleneck in Nvidia's next-generation AI infrastructure could finally give competitors a real opening in the high-end data center chip market. Technology research firm SemiAnalysis reported that Nvidia's Kyber NVL144 next-gen AI rack system, a crucial piece of CEO Jensen Huang's strategy to sell complete server solutions rather than just individual chips, may face delays until 2028 due to manufacturing challenges with a critical circuit board. While Nvidia has stated its roadmap remains on track, the potential delay raises questions about whether the company's aggressive technology schedule can keep pace with physical manufacturing constraints.

What Is Nvidia's Kyber System and Why Does It Matter?

Nvidia has fundamentally transformed its business over the past few years. The company no longer just sells graphics processing units (GPUs), the specialized processors that power artificial intelligence (AI) systems. Instead, under Huang's leadership, Nvidia now designs and sells complete server rack systems optimized for specific AI tasks. Kyber represents the next evolution of this strategy, designed to sit at the heart of AI data centers and help push AI models forward. The system comes in different variants tailored for specific AI workloads, including inference (running already-trained models) and agentic AI (systems that can take autonomous actions).

This shift from selling individual components to selling integrated systems has been a major competitive advantage. Nvidia has largely kept competitors out of the high-end market through an aggressive technology roadmap, but if manufacturing delays push Kyber into 2028, that window of opportunity could widen for rivals.

Who Could Actually Benefit From a Kyber Delay?

Two major technology companies stand to gain the most if Nvidia's timeline slips: Advanced Micro Devices (AMD) and Alphabet, Google's parent company. Neither company would need to dethrone Nvidia to win big; capturing even a modest share of the high-end AI infrastructure market would represent a significant business boost.

AMD has historically struggled against Nvidia due to a substantial software ecosystem gap. However, the landscape has shifted considerably. AMD has made major improvements to its ROCm platform, a software framework that allows developers to write code for AMD chips. Additionally, more programmers are working with open-source AI frameworks like OpenAI's Triton, which has helped narrow the gap, especially for inference workloads. AMD's chiplet designs, which allow the company to package more memory onto its chips, combined with its recent acquisition of Mext, a memory optimization software platform, position the company well to offer a high-end server solution specifically designed for inference tasks. The company has already captured some significant GPU deals, and data center sales grew by 57 percent year-over-year in the first quarter of 2026.

Alphabet brings a different strength to the competition. The company's Tensor Processing Units (TPUs) have earned strong recognition in the AI chip market, and its next generation will include chips optimized for both training and inference. Alphabet also possesses a unique advantage: it is the most complete AI play in the market, with both world-class chips and world-class AI models, giving it a cost edge that could make its offerings more attractive to customers seeking a completely optimized system without the risk of delay or the premium cost of Nvidia's platform.

How Could Competitors Capitalize on a Kyber Delay?

• Software Ecosystem Maturity: AMD's improved ROCm platform and the shift toward open-source frameworks like Triton have narrowed the software gap that once made Nvidia's ecosystem nearly impossible to compete against.
• Memory Optimization: AMD's chiplet designs and acquisition of Mext allow the company to offer memory-optimized solutions tailored for specific inference tasks, a key workload for AI data centers.
• Integrated AI Solutions: Alphabet's combination of custom TPU chips and proprietary AI models gives it the ability to offer customers a fully optimized, cost-efficient alternative to Nvidia's premium platform.
• Agentic AI Readiness: AMD's data center central processing units (CPUs) become increasingly important as agentic AI systems grow more complex, allowing AMD to offer a more complete solution beyond just GPUs.

What Does This Mean for the Broader AI Chip Market?

Nvidia's dominance in AI infrastructure has been nearly total. The company earned over 253 billion dollars in revenue over the past year alone, and CEO Jensen Huang anticipates a staggering 1 trillion dollars in orders through 2027 as Vera Rubin, Nvidia's latest chip architecture, begins shipping later this year. Nvidia's stock trades at less than 23 times 2025 earnings estimates, and analysts are calling for annual earnings growth of 51 to 52 percent over the next three to five years.

However, a delay in Kyber could signal that even Nvidia's manufacturing capacity has limits. The chip market is far too large for just one company to serve, and competition remains important for the industry. If AMD and Alphabet can gain meaningful footholds in the high-end market, it would represent a significant shift in the AI infrastructure landscape. AMD's earnings are expected to grow by 55 to 56 percent annually over the next three to five years, while Alphabet's analysts see 16 to 17 percent annual earnings growth over the same period. Neither company needs to displace Nvidia to be a winner; gaining even a piece of the high-end market would be a substantial boost to their businesses and stock valuations.

For now, Nvidia maintains its position as the industry leader, and the company has publicly stated that its roadmap remains intact despite the SemiAnalysis report. But if manufacturing constraints prove real, the window for competitors to establish themselves in premium data center solutions may finally be opening.