Tag
NVIDIA still dominates frontier-model training in 2026, but inference tells a different story. AMD MI300X/MI325X with mature ROCm, Intel Gaudi 3, Google TPU v6, and AWS Trainium/Inferentia deliver 20 to 50% lower cost per token without sacrificing quality. Here is when to choose each option.
La factura de IA en las empresas ha dejado de ser anecdótica. Entre tokens de modelos frontera, GPUs reservadas que nadie usa y pipelines RAG con cachés mal configuradas, muchos equipos pagan diez veces lo que deberían. Guía de FinOps específico para IA sin relatos promocionales.
Los primeros racks GB200 NVL72 llevan meses en manos de los hiperescalares y ya empiezan a verse mediciones públicas. La generación Blackwell no es una mejora incremental sobre Hopper, sino un cambio en la forma de entrenar modelos grandes. Repasamos qué cambia y qué no.
vLLM remains the reference engine for serving LLMs on GPU in 2025: automatic prefix caching sharply cuts latency for repeated prompts, speculative decoding speeds up large models, and multi-LoRA support lowers the cost of multi-tenant SaaS, though multi-GPU support and non-NVIDIA hardware remain weak points.
TensorRT-LLM is the NVIDIA inference engine that compiles each model into a binary optimized for the exact GPU and batch size it will serve. It uses hand-written CUDA kernels and native FP8 quantization on H100. Against vLLM it can run 2 to 3 times faster in the best case, at the cost of a 30 to 90 minute build.
vLLM serves language models on GPU using PagedAttention and continuous batching, two techniques that multiply throughput compared with a naive server. It exposes an OpenAI-compatible API, so migrating an existing application only requires changing the base URL and deploying the right binary.
Zed is the editor built by Atom's creators, rebuilt in Rust with a native GPU-rendering UI framework and no Electron. It delivers ~8 ms latency, real-time collaboration with integrated voice and shared cursors, and an open license (GPL v3 + Apache 2.0). A serious VS Code alternative for pairs and small teams.
Text Generation Inference (TGI) is the Hugging Face stack for serving open LLMs in production: continuous batching, 4-bit and 8-bit quantization, streaming, and an OpenAI-compatible API. After a brief restrictive-licence episode in 2023, it returned to Apache 2.0 in version 2.0.
