Emerging AI: From new investment to open-source models

AI

The artificial intelligence industry is ferociously meeting the hype with innovation. Covering every base, there’s considerable economic investment to match technological expansions in emerging AI, such as OpenAI Inc.’s new text-to-video model, Sora.

Another advancement can be seen in Google’s recent Gemini Pro 1.5 announcement, according to AI and data executive Howie Xu (pictured). 

“It was a huge deal — 1 million tokens, and then 10 million tokens in testing, so it’s huge for the large language model,” he said, adding that both Google and OpenAI have “amazing models, amazing progress.”

Xu spoke with theCUBE Research executive analyst John Furrier, during a CUBE Conversation from SiliconANGLE Media’s livestreaming studio in Palo Alto. They discussed the dynamic interplay between technological innovation and real-world application, offering invaluable insights into the emerging AI landscape and the transformative potential it holds for diverse industries.

AI’s vast potential must be tempered with realism

With the advent of open-source models rivaling proprietary counterparts, the industry is witnessing a seismic shift in adoption and functionality. Notable examples include the aforementioned Vertex Gemini and OpenAI Sora, signaling a paradigm shift in AI capabilities.

Amid the fervor, however, skepticism looms regarding the exaggerated claims of AI’s economic impact. While investment vigor remains strong, projections of trillions in dollar amounts — as mentioned by OpenAI CEO Sam Altman — must be juxtaposed against the logistical challenges of infrastructure scalability to keep expectations realistic and achievable, according to Xu.

“I can see where [Altman] came from,” he said. “They just released Sora, the model, and guess what? He can easily use 100X of the computing power he has today. With that two or three orders of magnitude, that’s 7 trillion. On the other hand, from a technology point of view, it’s more than just a GPU. Once you have the GPU, how do you stand that up? You need a space. You need power. You need cooling — it takes forever to even build that out.”

Navigating the emerging AI landscape: From models to infrastructure

With the industry expanding, there’s been an equal evolution of infrastructure requirements and hosting dynamics within the AI ecosystem. From the proliferation of graphics processing units to the intricacies of data center management, hyperscalers have a critical role in facilitating AI development.

Strategic appointments and investments by companies such as Meta Platforms Inc. and Microsoft Corp. in hardware underscore the symbiotic relationship between technology and infrastructure, according to Xu.

“Meta is going to spend $10 billion on GPUs alone,” he said. “Then I was looking at an earning report of Google, $11 billion, and an earning report of Microsoft, just one quarter, $11 billion for CapEx. I think the service providers, the host data guides, have an advantage because they have the cards, they have the space and they have the cooling.”

On the topic of open-source vs. proprietary AI models, respect must be given to the respective trajectories of each one in AI innovation. While frontier model proponents tout unparalleled computing power, challenges persist in terms of scalability and resource allocation, according to Xu. Conversely, open-source ecosystems foster collaboration and accessibility, driving democratization within the AI landscape.

Here’s theCUBE’s complete video interview with Howie Xu:

Photo: SiliconANGLE