Open source LLM frameworks continue to favour Nvidia’s CUDA platform over Cerebras’ faster AI hardware, showing that software compatibility and developer adoption remain the biggest drivers of AI infrastructure leadership.
Open source large language model (LLM) frameworks continue to reinforce Nvidia’s CUDA software ecosystem, limiting the adoption of faster AI hardware from rivals such as Cerebras.
Independent benchmarks show Cerebras’ wafer-scale AI processor delivers up to a 21x speed advantage over Nvidia hardware for latency-sensitive, low-batch LLM inference. However, most major LLM frameworks and enterprise AI developer stacks remain natively optimised for CUDA, forcing Cerebras users to depend on specialised compilation and custom engineering instead of out-of-the-box deployment.
The analysis argues that software compatibility and developer adoption—not raw chip performance—have become the defining factors in AI infrastructure leadership.
Nvidia’s financial results reinforce that advantage. The company posted Q1 FY27 revenue of $81.61 billion, up 85.2% year over year, with its Data Center business generating $75.25 billion, a 92% increase. Networking revenue jumped 199%, driven by InfiniBand, NVLink and Spectrum-X, while Nvidia also reported a 75% non-GAAP gross margin and $48.55 billion in quarterly free cash flow.
“The only platform that runs in every cloud, powers every frontier and open source model, and scales everywhere AI is produced,” said Jensen Huang, CEO, Nvidia.
Meanwhile, Cerebras reported Q1 GAAP revenue of $193.4 million, up 94%, and cloud services revenue growth of 178%. Although it secured a multi-year OpenAI inference contract worth more than $20 billion, the company expects full-year operating margins of negative 28% to negative 32% as it invests in expanding AI infrastructure.
