Analysis Disrupt? It’s an awful hackneyed term that some analysts, consultants and technologists like to use.
It is currently being applied to stock market darling Nvidia which lifted the covers off a broad range of tech at its GTC event last week, stuff that could “disrupt” all aspects of enterprise infrastructure.
As well as a plethora of reassuringly expensive datacenter infrastructure systems, including supercomputer-level performance in a single rack, the GPU maker announced DGX Station and DGX Spark, a workstations and personal computer respectively.
DGX Spark (formerly Project Digits) is a diminutive desktop box containing a GB10 Grace Blackwell system-on-chip (SoC) and 128 GB of unified system memory, which Nvidia claims is capable of 1,000 trillion operations per second (TOPS) in AI number wrangling – considerably more than your average AI PC.
DGX Station is closer in size to a professional workstation, and is based on Nvidia’s more powerful GB300 Blackwell Ultra desktop superchip with 784 GB of unified memory to speed large-scale training and inferencing workloads.
Both systems are considerably smaller than the GPU giant’s more familiar datacenter platforms, but still offer a decent whack of compute power for AI developers, researchers, data scientists, and possibly even students – although the tiny DGX Spark alone is said to cost $3,000.
Analysts at Omdia believe Nvidia is now working to enmesh itself in other areas of the enterprise after effectively cornering the market in AI training infrastructure: “In the hardware space they will disrupt PCs (desktop and laptop), workstations and storage on top of the revolution they have started in servers and networking.”
This seems like something of a tall order as AI PCs have hardly set the world alight since the concept was unleashed a year ago. As The Register reported late last year, PC sales were showing little sign of rebounding, despite the efforts of vendors to draw buyer attention with AI-capable systems, which have bells and whistles including a neural processing unit (NPU) – specialized circuitry for accelerating certain tasks.
“Nvidia’s approach to AI PC is to provide more compute to developer hands through Spark,” Vladimir Galabov, Omdia Research Director for Cloud and Datacenter, told us.
“I like that this device is usable with any PC and utilizes the same programming platform as the rest of Nvidia’s computers. We all use hard drives for external storage. It’s the same concept but for AI computing.”
By contrast, the mainstream AI PC hype is targeted at consumers and corporate users rather than scientists. The lack of killer applications and the hefty price tag has put off most buyers so far.
Galabov said: “AI PCs align with the push for tools like Copilot to make your life easier. Nvidia is not competing with this. They are creating a new market through a new form factor, a satellite AI device that can be used with any PC platform,” .
That all sounds good, yet a device aimed at AI developers and data scientists is a little too niche to rattle the cage of the mainstream PC market leaders, although Fortune Business Insights does estimate the global data science platform market reached $133 billion during 2024.
Other analysts also view Nvidia’s step into high-margin devices as somewhat specialized kit that isn’t for everyone.
“The DGX Spark, despite its ‘mini PC’ form factor, appears to be positioned as a more specialized and robust solution for AI development and local LLM experimentation than typical consumer-grade AI PCs,” Context chief analyst Antonio Talia told us. “Its focus on large unified memory caters specifically to the demands of larger AI models, a characteristic that differentiates it from more general-purpose AI PCs.”
AI-capable PCs now account for 50 percent of laptops flowing through European distribution channels, Context says. This surge, however, is more about increased product availability and short-term price cuts than a sudden boom in consumers clamouring to buy these systems.
“We have seen an 11-percentage point jump in just one week,” Talia said. “This dramatic week-on-week growth reflects a spike in the availability of AI-capable systems, rather than a fundamental shift in end-user demand.”
DGX Station is different again from AI PCs, presented as a desktop-sized machine explicitly designed for AI and described as “what a PC should look like for serious machine learning, data science, and LLMs,” according to Talia.
Its specifications far exceed those of typical workstations, meaning that while its cost is unknown, it’s likely to be costly given the DGX Spark’s pricing, so again it’s not for the average user.
There’s little indication that these will disrupt the enterprise PC market – and although there are whispers of a possible laptop form factor that Nvidia might have in the pipeline, Acer’s probably not panicking about losing customers that are looking to buy entry-level or mid-range priced notebooks.
In other parts of the infrastructure, Omdia points out Nvidia is gunning for the enterprise stack with its software ecosystem, ranging from OS to development platforms, to pre-trained model-as-a-service (AI apps). At GTC, Nvidia showed off a software framework called Dynamo that it described as the “operating system of an AI factory.”
Galabov points to numerous alliances the GPU maker has already forged to provide what are essentially managed AI services. “Partners like Accenture and HPE are simply integrating Nvidia’s software stack into an existing services portfolio,” he told us.
Nvidia is also moving steadily further into networking. As well as showing off Spectrum-X and Quantum-X switches with co-packaged optics at GTC, last month it teamed with Cisco to make Cisco Silicon One part of the Spectrum-X platform. Switchzilla will build systems using Nvidia’s Spectrum silicon.
“I think it’s important to recognize Nvidia’s inroads in selling switch ICs to Cisco making them an even bigger player in enterprise networking. Same with storage where I think they will gain access to a whole new total addressable market,” Galabov said.
With $70 plus billion in net profit last year, Nvidia has the wherewithal to leap into any market, yet dominance in one space doesn’t necessarily equate to, er, disruption in another.
Still, with that much financial muscle, CEO Jensen Huang can afford to fail, and if he wins then the price tags of this new range of products suggests he’ll win big. And could lead to bigger… disruptions. ®
