OpenAI CEO Sam Altman has acknowledged the disruptive impact of DeepSeek’s R1 model, calling it “impressive” and announcing plans to accelerate the release of new OpenAI models.
“We will obviously deliver much better models and also pull up some releases,” Altman wrote on X. His statement highlights the pressure on OpenAI as it faces its first major challenge from a cost-efficient competitor whose achievements have shocked the AI industry.
DeepSeek’s R1 has demonstrated a new level of efficiency in artificial intelligence development, achieving top-tier performance metrics at a fraction of the cost typically associated with advanced models. The company’s app has risen to No. 1 on Apple’s App Store, dethroning ChatGPT.
By relying on just $6 million in resources and 2,000 Nvidia H800 GPUs, R1 has outperformed expectations and set a new standard for what is achievable with limited hardware. The model has outpaced OpenAI’s o1 in several reasoning benchmarks, including a 97.3% score on MATH-500 and 79.8% on AIME 2024.
A New Model Forces a Shift in Strategy
Altman’s response reflects both admiration and urgency. While he praised DeepSeek’s achievements, he also reaffirmed OpenAI’s commitment to large-scale infrastructure investments. “More compute is more important now than ever before to succeed at our mission,” Altman noted, emphasizing the organization’s belief in computational power as the foundation of advanced AI development.
but mostly we are excited to continue to execute on our research roadmap and believe more compute is more important now than ever before to succeed at our mission.
— Sam Altman (@sama) January 28, 2025
the world is going to want to use a LOT of ai, and really be quite amazed by the next gen models coming.
OpenAI’s recently unveiled $500 billion Project Stargate initiative, launched in partnership with SoftBank and Oracle, underscores this approach, aiming to build expansive data center infrastructure to support AI innovations.
Despite these efforts, the emergence of DeepSeek R1 has raised questions about the sustainability of resource-intensive strategies. DeepSeek’s lean approach, combined with its use of Nvidia H800 GPUs—a throttled version of high-performance chips restricted under U.S. export controls—has demonstrated the potential for optimization over brute computational power.
Founder Liang Wenfeng explained this focus, stating, “We estimate that the best domestic and foreign models may have a gap of one-fold in model structure and training dynamics. For this reason, we need to consume four times more computing power to achieve the same effect. What we need to do is continuously narrow these gaps” (source: DeepSeek).
Industry-Wide Turmoil and Market Reactions
The release of R1 has sent shockwaves through the global technology sector. Nvidia’s stock plunged nearly 17% after the announcement, wiping $589 billion from its market value, as investors recalibrated their expectations for future demand for high-end GPUs.
Similarly, the Nasdaq 100 saw significant declines as the financial markets absorbed the implications of DeepSeek’s cost-efficient success. The broader industry impact has been palpable, with companies like Meta and Google grappling with the disruption.
At Meta, employees have described their AI division as being in “panic mode” following R1’s release. Anonymous discussions on the professional networking platform Blind reveal internal frustration with Meta’s reliance on resource-heavy strategies and organizational inefficiencies.
One engineer noted, “Management is worried about justifying the massive cost of GenAI org. How would they face the leadership when every single ‘leader’ of GenAI org is making more than what it cost to train DeepSeek V3 entirely?”
Meta’s Chief AI Scientist, Yann LeCun, acknowledged DeepSeek’s achievements in a LinkedIn post, attributing part of the company’s success to its open-source approach. “They came up with new ideas and built them on top of other people’s work,” LeCun wrote, referring to DeepSeek’s use of frameworks like PyTorch and its transparent release of model details.
A New Benchmark for Efficiency
DeepSeek R1’s reliance on Nvidia H800 GPUs is a particularly notable achievement. These GPUs, designed to comply with U.S. export restrictions, feature throttled performance compared to Nvidia’s top-tier H100 chips.
DeepSeek’s ability to achieve world-class results with these constraints has reignited debates over the effectiveness of U.S. policies aimed at limiting China’s access to advanced technology. By stockpiling H800 GPUs and focusing on optimization, DeepSeek has turned hardware limitations into an advantage.
This efficiency-first approach extends to DeepSeek’s broader philosophy. The company’s release of R1 and its subsequent Janus model series under the open-source MIT license has enabled global collaboration and transparency.
The Janus models, which include advanced multimodal capabilities, have already surpassed competitors like OpenAI’s DALL-E 3 in image-generation benchmarks
Altman’s comments suggest that OpenAI is adapting to this new competitive environment. While the company remains committed to its large-scale infrastructure investments, the accelerated release timeline indicates a recognition of the urgency created by DeepSeek’s success.