Nvidia experienced significant benefits of Last week was 24.7%This was driven in part by CEO Jensen Huang's statement the day before the stock's big rally, stressing the need to replace outdated data center equipment with new chips as more companies adopt AI.
“The computing industry is experiencing two simultaneous transitions: accelerated computing and generative AI,” Huang said. Nvidia's second quarter financial statement for investors “The $1 trillion global data center infrastructure will transition from general-purpose computing to accelerated computing as enterprises race to apply generative AI to every product, service, and business process.”
Old equipment will be replaced, but at what cost?
However, there are questions about the scale and cost of these alternatives. Implementing AI may require hardware upgrades or specialized processors for optimal performance, but the cost and scale of such replacements will vary depending on each company's specific requirements. Companies must evaluate their needs and costs on a case-by-case basis.
Bradley Shimmin, data and AI industry analyst at Omdia, a technology research and advisory group, said enterprises could take advantage of the emerging AI trends, which may require a new approach to acceleration hardware. I admit that there is. However, Shimmin doesn't fully support his Huang's belief that data centers need to replace all equipment.
“For many use cases, especially those with demanding model training requirements, enterprises are finding ways to reduce costs and accelerate time to market by investing in the latest and greatest AI hardware acceleration. We will aim to do so,” Shimmin said. “However, the current trend is to counter that, with researchers doing more with fewer parameters, highly curated datasets, and smarter training/fine-tuning using PEFT. I'm learning how to do it. [Parameter Efficient Fine-tuning] For example, LoRa. ”
Data center financial hurdles and physical limitations
In addition to the physical limitations of data centers, the pursuit of increased transistor density in data centers is not without hurdles. Building a fab is expensive, especially when combined with the rising cost of cutting-edge nodes. Data center leaders must navigate these financial concerns while meeting the ever-increasing demand for more advanced data center infrastructure.
As the data center industry continues to evolve, finding cost-effective solutions to achieve transistor density along with employee retention will be a key focus for data center operators.
Expanding the ecosystem and chip architecture
Shimin noted that chipmakers are also rushing to support generative AI use cases on smaller target platforms, such as Samsung's efforts to run full-scale models on chips and in phones. This indicates that the entire ecosystem will extend to different chip types and deployment configurations, including backend training and on-edge or in-device inference. Multiple chip architectures such as RISC-V, FPGA, GPU, and specialized solutions such as his AWS Trainium and Inferentium will play a key role in this evolving landscape.
“You can easily see the whole ecosystem exploding,” Simin said.
AI is attracting the attention of investors and data center infrastructure managers due to the increasing demands on scale for AI. This is due to the great success of OpenAI's various GPT models.
However, only a few companies can create powerful language and image models. Previously, it was possible to see significant improvements with smaller models running on data center-scale systems. To continue pushing the boundaries of technology, companies need to invest in better and more advanced hardware, giving Huang's statement great credence.
Karl Freund, founder and principal analyst at Cambrian-AI Research, said in a statement: Data center knowledge I would never bet that Jensen is wrong.
“He is a unique visionary,” Freund said. “Jensen has been saying for years that data centers are going to get faster, and that's becoming a reality. Based on processors, GPU segment accounted for the largest revenue share of 46.1% in 2021. ”
However, NVIDIA investors may want to temper their expectations for continued earnings growth. This means that scaling has already stalled and will soon reach a plateau. Implementing AI may require hardware upgrades or dedicated processors for optimal performance, but the scope of replacement can vary by company. As the technology ecosystem evolves, optimizations and advancements in AI models are expected to provide alternative solutions to balance hardware demands.
Sam Altman OpenAI CEO — Who recently asked Congress to consider proposed AI regulations — Further advances in AI will not come from making bigger models, he said.
“I think we're at the end of an era where big, giant models are going to be the norm.” As reported by Wired, Altman told an audience at an event at MIT in early April.. “We will continue to improve in other ways.”
