Outdated enterprise benchmarks fail to evaluate modern, high-performance infrastructure. In this case, legacy metrics penalize Nvidia's specialized AI focus.
Industry analysts love grid evaluations. They offer a convenient metric for enterprise procurement teams to evaluate technology vendors side by side. But when evaluation frameworks try to assess fundamentally different paradigms using the same outdated checklist, the results can range from inaccurate to completely absurd.
An example of this is the latest Forrester Wave for Data Center Networking Solutions, in which Nvidia ranks at the bottom of the “Contenders” quadrant. When you think of data centers today, Nvidia comes to mind, as it has driven much of the evolution of what a data center is today. This didn’t pass the eyeball test for me, so I took a deeper look at why.
The point I most disagree with is Forrester awarding Nvidia a 1 out of 5 for vision. That’s the lowest possible score for the company that laid the foundation for the modern AI data center. Forrester argues that to be a “leader,” one must fit all traditional data center frameworks, not just those for AI data centers, which creates an inaccurate scoring tool for Nvidia or any other specialized vendor.
The methodological flaw: Evaluating a special-purpose engine as a general-purpose switch
The fundamental flaw in this evaluation is criteria design. For as long as this Wave has existed, scoring criteria have been tailored towards a traditional data center environment. It asks questions such as:
- How well does the switch handle general-purpose enterprise traffic?
- What features support legacy campus, SMB, or traditional enterprise LAN workloads?
- How does the platform integrate with traditional enterprise management workloads?
When Forrester evaluated Nvidia against the metrics, the vendor naturally scored low. Like with vision, Nvidia received a score of 1.0 in enterprise management types, administrative experience, and switch portfolio scope.
However, judging Nvidia on its ability to serve general-purpose enterprise campus environments is like scoring Apple low in 2007 because it didn’t have a keyboard phone. Nvidia does not build a general-purpose enterprise switch for standard IT workloads, and it shouldn’t. In an analyst Q&A, CEO Jensen Huang once described the company he runs as a market maker, not a share taker.
In the context of data center switching, the products are specialized, high-throughput AI fabrics designed to prevent multi-billion-dollar GPU clusters from sitting idle while waiting on network I/O. The results speak for themselves. At a run rate of more than $11 billion, Nvidia sells more data center networking equipment than most vendors on the Wave combined, because that’s the high-growth segment of the industry. Trying to build a product to address the general-purpose part of the market would be a waste of time and money for Nvidia.
For the Wave, evaluating an AI factory fabric using a legacy enterprise networking matrix produces misleading results.
Flawed weighting in an AI-driven era
The mismatch becomes even clearer when examining the underlying questionnaire weightings. In a technical evaluation with dozens of criteria, Forrester allocated just 2% of the overall weight to AI infrastructure support, which seems light-years from market reality. Today, every enterprise board, CIO, and data center architect is restructuring infrastructure strategies around AI, yet Forrester assigned AI networking capabilities a 2% weighting.
By assigning a 2% weight to AI infrastructure, the evaluation framework heavily penalizes vendors focused on AI-optimized fabrics while over-rewarding legacy platforms designed for enterprise environments. As expected, Nvidia scored a perfect 5.0 in the AI Infrastructure category, but the 2% weighting effectively erased that technical lead because lower scores in categories such as legacy campus support outweighed it.
Misunderstanding innovation and market vision
However, the most striking misalignment in the scorecard is what I started with: Nvidia’s score of 1 for Vision. To claim that Nvidia lacks vision in data center networking ignores the structural shifts in the networking industry over the past five years.
Nvidia did not enter the networking market to become a secondary supplier of commodity Ethernet switches. Instead, they redefined how data center fabrics operate:
- Creating Ethernet for AI: Nvidia pioneered high-throughput, loss-free Ethernet fabrics designed specifically for non-blocking AI workloads.
- Architecture evolution: They led the industry transition from simple scale-out networks to scale-across topologies, introducing co-packaged optics and integrated DPUs (data processing units) to optimize workload efficiency.
- Ecosystem enablement: Rather than trying to capture every legacy enterprise edge deployment, Nvidia partnered with established enterprise networking vendors, such as market leader Cisco, enabling those partners to integrate Nvidia AI networking components directly into their enterprise management architectures.
Assigning a low vision score to the company driving these architectural shifts suggests that the evaluation framework measures adherence to historical enterprise models rather than to forward-looking technology direction. There’s a famous Wayne Gretzky quote about skating to where the puck is going to be, and this Wave is skating to where the puck was last season.
Why buyers need better frameworks
Industry research reports play a key role in enterprise decision-making. However, when they apply legacy evaluation models to specialized technology transitions, they risk confusing IT buyers.
Enterprise infrastructure buyers fall into distinct categories:
- Traditional enterprise IT: Organizations building standard cloud or on-premises environments for legacy applications need vendors such as Cisco, Arista, or HPE. These platforms provide the management tools, legacy protocol support, and operational models required for those workloads.
- AI and high-performance compute: Organizations building dedicated AI infrastructure require specialized architectures optimized for high throughput, low latency, and efficient GPU clusters.
Forcing both buyer profiles into a single evaluation framework does a disservice to both. It evaluates specialized AI fabrics against missing enterprise features they never intended to build, while giving enterprise buyers the impression that high-performance AI platforms are underperforming. One wouldn’t put pickup trucks and Formula 1 cars in an “automobile Wave.” Why would that be the case for data center networking?
Moving beyond rearview-mirror benchmarks
Market evaluation grids often reflect historical vendor positioning rather than emerging architectural trends. As data centers shift toward accelerated compute and AI-driven workloads, evaluation models must evolve as well. Nvidia isn’t misplaced in the Data Center Wave because its technology is lacking. It is misplaced because the evaluation relies on a legacy enterprise benchmark that fails to capture the requirements of modern AI networking.
IT decision-makers should look beyond generic evaluation grids. First, define your workloads’ specific operational requirements—and choose evaluation frameworks that measure what truly matters in your environment.
