Runpod's AI report reveals Qwen's dominance and Llama 4's surprising flop
A new report from Runpod challenges common assumptions about AI adoption. The State of AI Report reveals surprising trends in how businesses deploy large language models (LLMs) and other AI tools. Data from real-world serverless deployments—rather than benchmarks or surveys—paints a different picture of industry preferences.
One standout finding is the dominance of Alibaba Cloud's Qwen, now the most-deployed self-hosted LLM. Meanwhile, Llama 4, despite its high-profile launch, shows almost no adoption in production environments.
Runpod's analysis draws from anonymised logs of serverless AI deployments. This approach avoids the biases of surveys or synthetic benchmarks, offering a clearer view of actual usage patterns. The report highlights a shift toward performance, efficiency, and workflow control in production-grade AI.
In model deployment, Qwen leads the field, surpassing Llama variants. Llama 4, in particular, has seen near-zero adoption despite media attention. The data suggests developers prioritise reliability and integration over hype. For image generation, ComfyUI has become the dominant tool, powering over two-thirds of all image endpoints. AI video workloads also show a clear trend: upscaling tasks outnumber raw generation by roughly two to one. This indicates a focus on refining existing content rather than creating new material from scratch. Industry adoption varies by sector. HealthTech and FinTech represent nearly two-thirds of enterprise users on Runpod's platform. The report breaks down usage further: technology (45%), healthcare and pharma (22%), finance (15%), and research or academia (12%). Geographically, North America accounts for 52% of deployments, followed by Europe (28%), Asia-Pacific (12%), and other regions (8%). Runpod supports these workloads with on-demand pods, autoscaling serverless endpoints, and instant cluster services. The infrastructure caters to both AI-native companies and traditional enterprises integrating AI into existing systems.
The findings underscore a gap between public perception and real-world AI adoption. Qwen's lead and Llama 4's minimal uptake reflect practical choices over speculative trends. Businesses appear to favour tools that enhance efficiency and control rather than chasing headline-grabbing releases.
With video upscaling and image refinement dominating workloads, the data suggests a maturing AI landscape. Enterprises are optimising workflows rather than prioritising raw content creation.
Read also:
- Executive from significant German automobile corporation advocates for a truthful assessment of transition toward electric vehicles
- Crisis in a neighboring nation: immediate cheese withdrawal at Rewe & Co, resulting in two fatalities.
- United Kingdom Christians Voice Opposition to Assisted Dying Legislation
- Democrats are subtly dismantling the Affordable Care Act. Here's the breakdown
