EXCLUSIVE: FuriosaAI CEO Tells Benzinga 'Nvidia's Greatest Strength Is Also Its Achilles' Heel' After $800M Meta Offer, Targets Series D In 2026
EXCLUSIVE: FuriosaAI CEO Tells Benzinga 'Nvidia's Greatest Strength Is Also Its Achilles' Heel' After $800M Meta Offer, Targets Series D In 2026
Paula Tudoran
Fri, October 3, 2025 at 7:01 AM PDT 11 min read
In this article:
FuriosaAI founder and CEO June Paik said his startup’s architecture can reshape an industry dominated by Nvidia (NASDAQ: NVDA). He told Benzinga he is determined to make FuriosaAI the company that powers the transition to sustainable AI, while declining to comment on reports of an $800 million acquisition offer from Meta (NASDAQ: META).
Yet, his message was clear. “We believe Furiosa will have the greatest impact as an independent company," Paik said. "The world urgently needs new, much more energy-efficient ways to run advanced AI. We are committed to building the technology that will bring the benefits of sustainable AI to everyone.”
Paik added that being mission-driven has simplified key strategic choices, guiding Furiosa to remain independent and focused on long-term impact.
Don't Miss:
The Inherent Inefficiency Nvidia Can’t FixPaik sees Nvidia’s market dominance as both formidable and fundamentally vulnerable.
“Nvidia’s greatest strength—a legacy built on general-purpose [graphics processing units]—is also its Achilles’ heel,” he declared.”There is inherent inefficiency and architectural complexity that stems from using a general-purpose GPU architecture for specialized AI computing."
Paik added, "GPU makers have developed extremely sophisticated hardware and software workarounds to mitigate these limitations. But their efforts are directed at advancing on their current path, making it difficult for them to pivot to a new paradigm that is uniquely suited to AI.”
FuriosaAI started from first principles in 2017, designing the Tensor Contraction Processor architecture specifically for AI computations. “Our chip architecture, the Tensor Contraction Processor, works at a higher abstraction level,” Paik told Benzinga. “Tensor contraction is the fundamental operation in deep learning, which makes complex optimization problems much more tractable compared to the lower-level matrix multiplication used by GPUs.”
The architectural choice delivers practical advantages. “This means that when new AI models emerge, our compiler can run them efficiently without a massive effort to write hundreds of new, hand-tuned kernels,” Paik said. “It also means our chips minimize wasteful data movement, resulting in breakthrough power efficiency.”
Paik compared the competition to the shift from gas-powered cars to electric vehicles. Legacy automakers spent a century refining the combustion engine, while newer entrants began with a clean slate and designed around an entirely different paradigm.
According to Paik, Furiosa is following a similar path in AI compute, developing an architecture that delivers performance and cost savings beyond the reach of GPU-based systems.
Trending: Kevin O'Leary Loves ‘Wonderful Recurring Cash Flows' — These Small Industrial Assets Deliver Just That
The Electricity Bottleneck Blocking AI Growth—And FuriosaAI’s SolutionFuriosaAI’s RNGD chip has been described as the “foundational compute layer of the AI stack,” a positioning that Paik believes will fundamentally alter the balance of power in the chip industry.
“We empower enterprises economically,” Paik told Benzinga. “RNGD’s superior efficiency translates directly into lower total cost of ownership and significant operational cost savings. This allows customers to get 3.5x more tokens per rack and run powerful AI applications within standard air-cooled data center racks, where power is often limited.”
The implications extend beyond cost savings. “We enable Sovereign and On-Prem AI,” Paik said. “By providing efficient, locally deployable infrastructure, RNGD directly addresses the cost and data privacy concerns of enterprises. This architecture enables customers, such as LG AI Research, to control and own their own AI stack rather than being reliant solely on large chipmakers and hyperscalers.”
For hyperscalers, Paik said RNGD provides crucial flexibility by increasing compute density in data centers, allowing more tokens per second from the same number of racks. The design also reduces the bottleneck of generating and delivering electricity to power AI services, enabling faster scaling in regions that cannot support the massive energy demands of hundreds of advanced GPU racks.
Three Pillars Of Sustainable AI ComputingPaik’s vision of sustainable AI extends across economic, societal, and environmental dimensions that address interconnected global challenges.
Economic sustainability tackles the fundamental cost crisis. “Economic sustainability is primarily about the fundamental cost of running AI,” Paik told Benzinga. “Because the current infrastructure is prohibitively costly, Furiosa’s superior power delivers substantial operational cost savings and a lower Total Cost of Ownership. This is critical because many AI applications are currently losing money due to high infrastructure costs.”
Societal and infrastructure sustainability emphasizes democratizing access while advancing sovereignty. Paik frames sustainable AI as the ability to scale technology for everyone by removing economic and infrastructure barriers.
This approach supports global sovereign AI initiatives by providing secure, efficient, and locally deployable infrastructure, reducing reliance on a small group of centralized cloud providers and chipmakers.
Environmental sustainability centers on relieving the energy grid crisis that threatens AI growth. Furiosa's RNGD accelerator delivers significantly higher energy efficiency than advanced GPUs, consuming only 180 watts, which lowers the environmental burden from data centers.
Paik told Benzinga the company plans to extend these efficiency gains with future products designed to further reduce strain on global energy systems. |
