Photo credit: www.cnbc.com
DeepSeek has recently made waves in the U.S. AI landscape with the launch of its newest model, significantly impacting Nvidia’s market valuation, which has seen a decline of hundreds of billions. While leading firms in the sector contend with this shift, smaller AI enterprises perceive a favorable opportunity for growth alongside the emerging Chinese startup.
Several representatives from various AI-focused companies have conveyed to CNBC that the arrival of DeepSeek presents a “massive” opportunity rather than an impending threat.
“Developers are eager to transition from OpenAI’s costly, proprietary models to more accessible solutions like DeepSeek R1,” stated Andrew Feldman, CEO of Cerebras Systems, a competing artificial intelligence chip startup. Cerebras is known for its advanced computing clusters and services that rival Nvidia’s graphic processing technology. Feldman noted that the debut of the R1 model triggered one of the most significant increases in demand for Cerebras’ offerings.
He further elaborated, “The R1 indicates that growth in the AI market won’t be controlled by a single entity — the notion of hardware and software barriers fails to apply to open-source models.”
Open source refers to software for which the source code is publicly accessible, allowing modification and redistribution without restrictions. In contrast to competitors like OpenAI, DeepSeek’s offerings are characterized by their open-source models.
DeepSeek contends that its R1 reasoning model can compete with top-tier American technology while operating at reduced costs and without reliance on cutting-edge graphic processing units. However, these claims are met with skepticism from some industry analysts and competitors.
“Similar to the dynamics seen in the PC and internet sectors, declining prices incentivize widespread adoption. The AI market appears to be on a comparable growth trajectory,” Feldman remarked.
Inference Chips
Industry insiders predict that DeepSeek may accelerate the adoption of innovative chip technologies by enhancing the transition from AI training to the “inference” stage.
The inference process involves applying AI to generate predictions or make decisions based on new inputs, distinguishing it from the model training phase.
“In straightforward terms, AI training is about creating the tool or algorithm, while inference focuses on utilizing these tools in real-world applications,” explained Phelix Lee, an equity analyst at Morningstar, concentrating on semiconductor developments.
Despite Nvidia’s stronghold on GPUs essential for AI training, various competitors are identifying opportunities for growth in the “inference” market, where they promise enhanced efficiency at lower costs, as noted in a piece from CNBC.
AI training demands substantial computational resources, whereas inference can function on less powerful chips designed for a more focused range of tasks, Lee explained.
A number of AI chip startups have reported to CNBC that interest in inference chips and technology is rising as businesses increasingly implement and build upon DeepSeek’s open-source model.
Sid Sheth, CEO of AI chip startup d-Matrix, remarked, “[DeepSeek] has shown that smaller open models can achieve comparable or superior capabilities to larger proprietary ones at a fraction of the cost.” He further noted a growing influx of global clients eager to enhance their inference strategies.
Robert Wachen, co-founder and COO of AI chipmaker Etched, mentioned that the release of DeepSeek’s reasoning models has sparked interest from numerous companies, saying, “Companies are reallocating their budgets from training infrastructure to inference infrastructure.”
Wachen added, “DeepSeek-R1 has established that inference-focused computing is now the benchmark method for every significant model vendor, and accommodating this demand will necessitate increasing our computing capacities to support millions of users.”
Jevon’s Paradox
Experts in the field share the view that DeepSeek’s innovations are advantageous for both AI inference and the broader AI chip market.
A recent report by Bain & Company noted, “DeepSeek’s achievements seem rooted in a series of engineering advancements that substantially lower inference costs while also optimizing training expenditures.”
The report suggested that if ongoing efficiency improvements persist, they could facilitate cheaper inference, ultimately promoting greater adoption of AI technologies.
This phenomenon aligns with Jevon’s Paradox, a principle indicating that cost reductions in new technological advancements lead to increased demand.
In a recent analysis, Wedbush, a financial services and investment firm, maintained that there is substantial potential for AI utilization among enterprises and retail consumers on a global scale.
During an interview on CNBC’s “Fast Money,” Sunny Madra, COO at Groq, a firm specializing in chips for AI inference, emphasized that as demand for AI continues to escalate, smaller companies are poised for significant growth opportunities.
Madra stated, “With the impending need for more tokens — the units of data processed by AI models — Nvidia will be unable to meet the demand for chips across the board, presenting fertile ground for us to expand our market presence.”
Source
www.cnbc.com