In a wide‑ranging keynote, Nvidia CEO Jensen Huang tried to define what makes his company different in the new era of accelerated computing. His answer: Nvidia is a “vertically integrated but horizontally open” computing company, a structure he argued is necessary if the industry is going to keep delivering big performance gains and cost reductions for real‑world applications.
Huang pushed back on the idea that accelerated computing is just a chip or systems problem. The missing phrase, he said, is “application acceleration”. In other words, the point of accelerated computing is not simply to build faster processors, but to make specific applications and domains run dramatically faster and more efficiently than they can on general‑purpose CPUs alone.
What does ‘vertically integrated but horizontally open’ mean?
During the keynote, Huang walked through the layers of Nvidia’s approach. Vertically, the company spans from understanding applications and domains, through algorithms and deployment scenarios, all the way down to systems and chips. Huang argued that to deliver real acceleration, Nvidia has to understand how software in fields like automotive, financial services or robotics actually works, and then design libraries, systems and silicon that are tuned to those needs.
At the same time, Huang stressed that Nvidia is “horizontally open”. By that he meant that, while the company is deeply involved from top to bottom of the stack, it integrates its technology into whatever platforms customers and partners are already using. Nvidia offers software and libraries and works with other companies’ technologies so that its accelerators can show up inside cloud services, on‑premises systems and edge devices rather than forcing customers into a single, closed ecosystem.
This mix of vertical integration and horizontal openness, Huang suggested, is what allows Nvidia to keep building out a library‑after‑library, domain‑after‑domain approach. Each new library targets a specific area—whether that is graphics, data processing, AI training, or inference—while still fitting into a broader ecosystem that includes many different vendors and deployment models.
Why Huang says CPUs have “run out of steam” for AI workloads
Huang contrasted this model with the traditional reliance on general‑purpose CPUs. If a processor could make “everything faster”, he said, that would be called a CPU—but in his view that approach has effectively run out of steam for the kinds of workloads that now dominate AI and data‑intensive computing.
Instead, Huang argued that the only way to keep bringing “tremendous speed‑up” and “tremendous cost reduction” to modern applications is through domain‑specific or application‑specific acceleration. That means building hardware and software that are tuned to particular tasks, and then exposing that work through libraries and platforms that developers can adopt without becoming experts in GPU programming themselves.
He tied this back to Nvidia’s decision to invest in a growing catalogue of domain‑specific libraries. Each library embodies knowledge of a particular field—such as computer vision, recommendation systems or large language models—and implements algorithms in ways that are optimised for Nvidia’s systems. Over time, Huang suggested, this approach lets the company bring accelerated computing into more and more verticals.
How Nvidia works with partners across its ecosystem
Huang also used the keynote to thank what he called Nvidia’s upstream and downstream supply chain, noting that companies ranging from decades‑old industrial firms to more recent technology partners are now part of that ecosystem. Some of those partners provide components, manufacturing capacity or infrastructure, while others integrate Nvidia’s technologies into their own products and services.
He highlighted that many of the cloud service providers in the audience have embraced this model, inviting Nvidia to integrate its libraries and accelerators into their platforms and asking the company to help land more customers on their clouds. In Huang’s telling, Nvidia’s role is to integrate with partners, accelerate workloads, and then help bring those workloads to whichever platforms customers choose.
Financial services emerged as one notable example: Huang pointed out that the largest percentage of attendees at this particular GTC conference came from the financial services industry. He joked that he hoped they were developers rather than traders, but the point was clear—industries that live on data and latency are flocking to accelerated computing to rework risk models, pricing engines and analytics pipelines.
Why Huang thinks this is “the beginning of something very big”
Looking across Nvidia’s supply chain and customer base, Huang argued that the company’s approach is starting to pay off. He said that many of the companies partnering with Nvidia, including those that have been around for 50, 70 or even 150 years, had just recorded record years, and he credited accelerated computing as a key part of that success.
For Huang, the combination of a vertically integrated understanding of applications and a horizontally open approach to partnerships is what will allow Nvidia to keep expanding into new domains. As more industries adopt domain‑specific acceleration, he believes the benefits of faster performance and lower costs will compound across the economy, marking what he described as the beginning of something very big.
Sources
- Live keynote remarks by Nvidia CEO Jensen Huang on accelerated computing, vertical integration and openness
- Public reporting and profiles describing Nvidia’s role in the rise of accelerated computing and domain‑specific AI workloads