As the factitious intelligence (AI) turf struggle escalates, Elon Musk-owned Grok and Chinese language DeepSeek fashions now stand on the forefront of AI functionality — one optimised for accessibility and effectivity and the opposite for brute-force scale — regardless of the huge disparity in coaching sources, a report confirmed on Saturday.
Grok-3 represents scale with out compromise — 200,000 NVIDIA H100s chasing frontier good points, whereas DeepSeek-R1 delivers related efficiency utilizing a fraction of the compute, signalling that revolutionary structure and curation can rival brute power, in response to Counterpoint Analysis.
Since February, DeepSeek has grabbed international headlines by open-sourcing its flagship reasoning mannequin DeepSeek-R1 to ship efficiency on a par with the world’s frontier reasoning fashions.
“What units it aside is not simply its elite capabilities, however the truth that it was skilled utilizing solely 2,000 NVIDIA H800 GPUs — a scaled-down, export-compliant different to the H100, making its achievement a masterclass in effectivity,” stated Wei Solar, principal analyst in AI at Counterpoint.
Musk’s xAI has unveiled Grok-3, its most superior mannequin so far, which barely outperforms DeepSeek-R1, OpenAI’s GPT-o1 and Google’s Gemini 2.
“Not like DeepSeek-R1, Grok-3 is proprietary and was skilled utilizing a staggering 200,000 H100 GPUs on xAI’s supercomputer Colossus, representing a large leap in computational scale,” stated Solar.
Grok-3 embodies the brute-force technique — huge compute scale (representing billions of {dollars} in GPU prices) driving incremental efficiency good points. It is a route solely the wealthiest tech giants or governments can realistically pursue.
“In distinction, DeepSeek-R1 demonstrates the facility of algorithmic ingenuity by leveraging strategies like Combination-of-Consultants (MoE) and reinforcement studying for reasoning, mixed with curated and high-quality knowledge, to realize comparable outcomes with a fraction of the compute,” defined Solar.
Grok-3 proves that throwing 100x extra GPUs can yield marginal efficiency good points quickly. But it surely additionally highlights quickly diminishing returns on funding (ROI), as most real-world customers see minimal profit from incremental enhancements.
In essence, DeepSeek-R1 is about reaching elite efficiency with minimal {hardware} overhead, whereas Grok-3 is about pushing boundaries by any computational means crucial, stated the report.
