Jonathan Ross
Showing 1–5 of 5 transcripts.
Groq Founder, Jonathan Ross: OpenAI & Anthropic Will Build Their Own Chips & Will NVIDIA Hit $10TRN
Jonathan Ross, Harry Stebbings
Jonathan Ross projects that within five years, the AI sector's trajectory will be defined by a critical convergence of energy capacity and compute availability, with NVIDIA potentially reaching a $10 trillion valuation due to a competitive temporal moat. While the market remains concentrated among approximately 36 entities, firms like Grok are leveraging proprietary hardware cycles and massive speed advantages to capture value, contrasting sharply with competitors hindered by supply chain constraints. Ross further argues that this rapid evolution will drive significant deflationary economic shifts and create new labor categories, ultimately positioning energy-sufficient nations with rapid deployment capabilities to dominate the global landscape.
- RAISE Summit26 min
Groq's CEO: A conversation with Jonathan Ross
Grok is establishing a strategic foothold in Europe with a newly operational Helsinki data center that leverages proprietary SRAM-based LPUs to solve critical compute shortages through air cooling and reduced energy consumption. By combining this hardware with a fully automated compiler and sovereign AI partnerships in Saudi Arabia and Canada, the company differentiates itself with a high-volume, low-margin revenue model optimized for sequential token generation. This approach positions Grok as a complementary force to NVIDIA, targeting the lucrative inference market where its architecture achieves superior latency and efficiency while navigating a regulatory landscape it defines as akin to utility infrastructure.
Jonathan Ross, Founder & CEO @ Groq: NVIDIA vs Groq - The Future of Training vs Inference | E1260
Jonathan Ross, Harry Stebbings
Grok CEO Jonathan details a strategy where the company generates $1.5 billion in revenue by utilizing specialized LPU architectures to deliver inference services at five times lower costs than NVIDIA while scaling from 640 to over 40,000 chips. The organization prioritizes geometric capability gains through synthetic data and maintains a lean 300-person team to bypass memory bottlenecks and secure power deals with Saudi entities. This approach positions Grok to capitalize on an impending data center power oversupply while aiming to revolutionize longevity and solve complex alignment challenges within a decade.
Jonathan Ross: DeepSeek Special - How Should OpenAI and the US Government Respond | E1253
Jonathan Ross, Harry Stebbings
DeepSeek-R1 has disrupted the global AI landscape by demonstrating that high-performance foundation models can be achieved through data efficiency and distillation techniques rather than brute-force compute spending. This breakthrough has intensified geopolitical concerns regarding Chinese state surveillance while compelling Western competitors like OpenAI and Meta to pivot toward business models based on network effects, enterprise switching costs, and open-source strategies. Consequently, the industry is shifting from a training-centric paradigm to an inference-focused economy where value accrues to specialized applications and infrastructure providers benefit from surging demand driven by falling model costs.
- RAISE Summit38 min
Chamath Palihapitiya & Groq CEO Jonathan Ross | RAISE Summit 2024 | Paris
Chamath Palihapitiya, Jonathan Ross, Olivier Ollier
Grok CEO Jonathan Ross, a former Google TPU architect, presents a strategic shift in the AI industry where hardware-focused inference is poised to surpass training as the dominant market force. Leveraging proprietary Low Precision Units and an alternative supply chain strategy, the company has attracted 75,000 developers in just 30 days and secured a massive compute partnership with Aramco Digital to challenge NVIDIA's infrastructure monopoly. With projections to deploy 1.5 million LPUs by next year, Grok aims to redefine AI deployment costs and latency standards while positioning non-tech enterprises as primary consumers of global compute capacity.