Alternatives to Thunder Compute (YC S24)

Paperspace is a cloud GPU platform from DigitalOcean offering per-second-billed NVIDIA H100, A100 and V100 instances, hosted notebooks, and managed ML training

CoreWeave is the AI-native GPU cloud for training, reinforcement learning post-training, and low-latency inference on NVIDIA cluster-scale

Rent GPU pods, serverless inference endpoints and multi-node clusters by the second on Runpod's AI developer cloud.

Vast.ai is a GPU rental marketplace where market-set prices and per-second billing let you provision 20,000+ GPUs from code.

Serverless GPU compute for AI agents: forkable microVMs, sandboxes, and inference endpoints billed by the second.

Verda (formerly DataCrunch) is a European GPU cloud: on-demand NVIDIA instances, InfiniBand clusters, and serverless inference.

Zettascale builds XPU accelerator chips that cut AI inference and training energy by keeping data close to compute.

Training-free distributed inference that splits SDXL generation across multiple GPUs for up to 6.1x lower latency

Fireworks AI runs open-weight model inference, orchestration and training on one platform with per-token and per-GPU-second pricing.

Free Apache-2.0 command-line framework for fine-tuning and post-training LLMs, vision-language and MoE models on GPUs you own or rent

LLaMA-Factory fine-tunes 100+ LLMs and VLMs with a zero-code CLI and the LlamaBoard web UI

Salad Cloud rents idle consumer GPUs from $0.015 per GPU-hour, billed per second, for retryable inference and batch work.

European edge AI inference accelerators — the Metis and Europa AIPU lines run vision, agentic and GenAI workloads from embedded modules to server cards.

Run 20+ pre-installed open-source AI apps — ComfyUI, Stable Diffusion, RVC, LoRA training — in your browser on rented cloud GPUs

Crusoe Cloud rents NVIDIA and AMD GPU instances by the hour, plus managed inference and fine-tuning for AI teams

Serverless GPU cloud for production AI — declare Python functions and hardware in one file, scale to thousands of GPUs, pay per second.

Managed Ray platform for distributed training, multimodal data curation, batch embedding, and post-training on your own GPUs

DeepSim is an AI-driven multi-scale physics simulator that runs nano-to-macro semiconductor simulations in a single pass, claiming 1000X speedup.

Cerebras runs frontier models at up to 750 tokens/sec for latency-critical AI agents and voice apps.

Open infrastructure giving builders one API for image, video, speech, music, text, agents and MCP tools

Deploy custom ML models on serverless Nvidia GPUs with per-second billing and scale-to-zero.

ViewComfy turns ComfyUI workflows into no-code web apps and autoscaling serverless APIs, billed per active GPU second.

Rent private bare-metal GPU servers in the EU with one-click vLLM, OpenWebUI and ComfyUI AI templates.

An open-source, readable LLM inference engine written in Python with hand-written CUDA and Triton kernels.

EU-sovereign GPU cloud for AI training and inference — Hyperstack on-demand or single-tenant Secure Private Cloud.

Krutrim Cloud rents A100 and H100 GPUs from ₹189/hr with INR billing and Indian data residency.

Run Llama 3.1 405B and other huge language models at home over a BitTorrent-style decentralized inference network.

Freesolo is a managed post-training platform that turns your dataset into a production-ready small language model via Claude Code, Cursor,

Decentralized GPU rental and lending marketplace with pay-as-you-go hourly pricing and crypto rewards.

Enterprise LLM fine-tuning with accuracy SLAs and sub-second inference.