
### Introducing the Vast.ai Startup Program: AI Infrastructure Without the Burn Rate

Startups can now get [$2,500 in free GPU credits](https://vast.ai/startup) when they move workloads to Vast.ai.

To apply, get in touch with our team to share a bit about your project and compute needs. From there, we'll finalize your credit offer, help you configure the right setup, and get you up and running quickly. You'll also have access to **24/7 priority support** throughout the credit period.

Compute credits apply across Vast.ai's global GPU marketplace, with access to everything from consumer-grade cards to high-end accelerators like A100s and H200s.

The Vast.ai Startup Program gives startups an on-ramp to the same platform that teams rely on well beyond the early stage. That's because Vast.ai is designed to scale alongside you - without long-term commitments or runaway costs.

### Built for Real AI Workloads

From day one, Vast.ai supports the tools and workflows that teams already use. Launch preconfigured environments with Jupyter, Docker, RAPIDS, and Hugging Face to train deep learning models, run large-scale experiments, and process massive datasets on demand.

We offer a growing selection of ready-to-run AI models in our [Model Library](https://vast.ai/model-library), plus thousands of [templates](https://cloud.vast.ai/templates) to get started immediately - or deploy your own custom model anytime. Your team can even collaborate across departments and campuses, avoiding duplicated infrastructure and fragmented workflows.

Most importantly, with Vast.ai, your team can cut costs without cutting capability. Reduce infrastructure expenses by **up to 81%** compared to major cloud GPU providers, while paying only for what you need, when you need it.

Plus, most startups can migrate existing workloads with minimal changes and begin running jobs shortly after approval.

### Secure, Compliant, and Production-Ready

Vast.ai is **SOC 2 Type II** certified and designed to support all of your [security and compliance](https://vast.ai/compliance) needs from the start.

Run workloads in isolated instances with end-to-end encryption and full control over compute environments to keep your sensitive data protected. Our [Secure Cloud](https://cloud.vast.ai/?secure_cloud=true) option lets you filter for GPUs from our vetted datacenter partners for even higher security needs.

Notably, these safeguards support reproducible research via consistent environments, as well as compliance-friendly configurations aligned with GDPR and HIPAA requirements.

With [Vast.ai Serverless](https://vast.ai/article/vast-ai-serverless-automated-gpu-scaling), moving from development to production is even easier. Inference workloads run through a fully serverless API that automatically routes to the most cost-effective GPUs available, according to your performance requirements. This eliminates manual instance management and capacity planning - so you can leverage our distributed GPU fleet with minimal overhead.

### Get Started with Vast.ai Today

For early-stage teams, access to affordable GPU compute is crucial. But there's more to it than that. You also want infrastructure that scales as you grow and supports evolving security and compliance needs - all without locking you into long-term contracts.

The Vast.ai Startup Program helps you get there faster.

[Apply today](https://vast.ai/startup) to get up to 2,500 free compute hours and start building on Vast.ai on your own terms.


Startups can get up to $2,500 in free GPU credits when moving workloads to Vast.ai, with priority support and production-ready infrastructure.

Introducing the Vast.ai Startup Program


As 2025 comes to a close, we're looking back at the milestones and updates that shaped the past year for us here at Vast.ai.

What a year it's been! From new hardware to major security achievements and product launches, here are our top moments of 2025.

## Expanding Our Global GPU Fleet

The past year saw some of the biggest hardware expansions in Vast.ai's history.

In 2025, we added NVIDIA's latest RTX 5000-series GPUs – including the RTX 5090 – followed by high-performance machines like the RTX PRO 6000 WS and the H200 and B200. These additions supported higher-throughput workloads and more efficient AI inference across a [wide range](https://cloud.vast.ai/) of hardware profiles.

Another highlight: our RTX 5090 inventory surpassed 1,000 machines this year!

## Security and Compliance Milestones

This year, Vast.ai achieved both **SOC 2 Type I** and **Type II** certification. These independent audits validated that the security infrastructure we've built over 6+ years of trusted service continues to meet rigorous compliance standards.

We take this responsibility seriously. Data privacy and integrity remain core priorities in everything we build. As we continue to advance our [security and compliance](https://vast.ai/compliance) roadmap, we're committed to upholding the trust you place in us.

## New Templates and Tools

We introduced a number of ready-to-use templates this year, to help you move from idea to execution quickly and seamlessly. Some notable ones included:

- [Langflow + Ollama](https://docs.vast.ai/langflow-ollama) for visual AI workflow design
- [Ostris AI Toolkit](https://cloud.vast.ai/?ref_id=62897&template_id=81e6e9dd266d9bdbcbb0fd48b9afe357) for custom Stable Diffusion and LoRA training
- [Open-Sora](https://cloud.vast.ai/?ref_id=62897&creator_id=62897&name=Open-Sora) and [LTX Video](https://cloud.vast.ai/?ref_id=62897&creator_id=62897&name=ComfyUI%2520%252B%2520LTX%2520Video) (ComfyUI) for video generation workflows
- [ComfyUI ACE Step](https://cloud.vast.ai/?ref_id=62897&creator_id=62897&name=ACE%2520Step%2520Music%2520Generation) for AI-assisted music creation
- [InvokeAI](https://cloud.vast.ai/?ref_id=62897&template_id=c309df384010cf9d3dc9abb9e0a5a6fd) for image generation in a professional web interface
- [Dia 1.6b TTS](https://cloud.vast.ai/?ref_id=62897&template_id=6ba071acb5f590913c5eaf6c2dc3d0db) for lifelike text-to-speech

But the biggest update of the year was still ahead...

## Introducing Vast.ai Serverless

The headline story of 2025 was the [recent launch](https://vast.ai/article/vast-ai-serverless-automated-gpu-scaling) of **Vast.ai Serverless** – a fully automated GPU compute layer designed for AI inference at scale.

With our Serverless offering, you can easily run workloads without managing instances on your own or overpaying for capacity. You simply send inference requests and the platform automatically selects the right hardware, routes workloads across regions as needed, and predictively scales up and down based on usage.

In short, Vast.ai Serverless gives you access to a diverse, globally distributed GPU fleet with a serverless model that applies predictive optimization and flexible scaling – with none of the overhead. We're thrilled to finally make truly serverless AI inference a reality on Vast.ai!

## Looking Ahead

To all of our users and the wider Vast.ai community, thank you for being part of our journey this year. It's been an incredible year of growth, and we're grateful to everyone building and innovating with us.

Here's to an even stronger year ahead!


As 2025 comes to a close, we're looking back at the milestones and updates that shaped the past year for us here at Vast.ai.

Vast.ai's 2025 Year in Review


# Running ComfyUI on Vast.ai Serverless: Simple, Scalable Image Generation

ComfyUI has become one of the most popular open-source AI tools for image generation, for good reason.

Rather than typing a prompt and hoping for the best, you build the entire process yourself through a **node-based interface that gives you full, visual control** over how images are created. Every part of the workflow – from models, prompts, and parameters – can be adjusted or branched however you like, and workflows themselves are also shareable and reusable.

While ComfyUI runs well locally for smaller projects, it's a whole other challenge to use it for heavier custom workflows or at production scale. Complex ComfyUI graphs can exceed consumer GPU memory or demand more throughput than a single machine can handle – or simply take too long to iterate on locally.

That's where **Vast.ai Serverless** comes in.

### ComfyUI Meets Vast.ai Serverless

Vast.ai Serverless makes it easy to run your ComfyUI workflows at scale, thanks to a ready-to-use template that handles all of the setup for you – and no need to manage GPUs or monitor capacity.

Simply send requests to ComfyUI and have your generated assets automatically uploaded to S3-compatible storage. The template will return pre-signed URLs in response to requests, as well as detailed process updates while the workflow runs.

It's a straightforward way to integrate ComfyUI into your applications or pipelines without worrying about infrastructure overhead.

### What Comes in the ComfyUI Serverless Template

Our ComfyUI Serverless template provides [everything you need](https://www.google.com/url?q=https://docs.vast.ai/documentation/serverless/comfy-ui&sa=D&source=editors&ust=1766188232265630&usg=AOvVaw0k4v0rLcf_0MltEhly3_ZW) right out of the box:

- **ComfyUI** and its **API wrapper** (which manages asset uploads)
- **Stable Diffusion 1.5** for benchmarking
- **PyWorker** for processing JSON workflows
- **Provisioning script** for custom configuration

The template saves you from building or maintaining your own Docker image and ensures you get a consistent environment every time a worker spins up. Once the template is in place, several capabilities come built into the workflow by default.

#### 1. Built-In Asset Handling with S3 Storage

Generated assets are automatically uploaded to your private S3-compatible storage buckets, and pre-signed URLs provide secure, temporary access to download results directly. This keeps storage and retrieval simple, even as workloads scale.

Just add your credentials once in your account settings, and every worker can use them. If you need different buckets for different jobs, you can override these variables on a per-request basis.

#### 2. Flexible Configuration When You Need It

With ComfyUI Serverless, you can keep things simple or tailor your setup with optional configurations. For instance:

- Add a webhook to be notified after generation completion or failure
- Use a custom PyWorker git repository URL
- Adjust benchmark settings to better reflect your workload

With these configuration tools, you're free to adapt the template to your unique needs.

#### 3. Distributing Work Through Benchmarking

Each worker runs a quick benchmark when it initializes, to validate GPU performance and calculate a performance score. This score is what determines how requests are distributed across workers.

The template includes a default Stable Diffusion 1.5 benchmark for this purpose, but you can configure it to match your actual workload complexity or even supply your own custom benchmark.

### Try It Out Interactively Before Going Serverless

The ComfyUI Serverless template is designed for serverless execution, of course, but you can test it as an interactive instance first. Interactive testing helps confirm that everything works as expected before you start sending requests through Serverless.

You'll have to enable port forwarding to access applications over the public interface, but the benchmarking process will be visible in the instance logs either way.

### A Smoother Path to Production-Scale ComfyUI

Vast.ai Serverless and our ComfyUI template lets you go from a local workflow to a scalable, on-demand system with minimal setup. The environment is preconfigured – with customizable adjustments as needed – and assets are handled automatically while workloads are routed to the right GPUs based on real performance.

For creative work beyond images, we offer Serverless templates for [text-to-video](https://www.google.com/url?q=https://docs.vast.ai/documentation/serverless/comfyui-wan-2.2&sa=D&source=editors&ust=1766188232275066&usg=AOvVaw33TRQTBXX80tZ040XucSdk) and [text-to-music](https://www.google.com/url?q=https://docs.vast.ai/documentation/serverless/comfyui-acestep&sa=D&source=editors&ust=1766188232275242&usg=AOvVaw3Dnzq-bbQYJjA76I3vHcU3) workflows, as well.

Ready to see ComfyUI in action? Just launch the [ComfyUI Serverless template](https://www.google.com/url?q=https://cloud.vast.ai/?ref_id%3D62897%26creator_id%3D62897%26name%3DComfyUI%2520(Serverless)&sa=D&source=editors&ust=1766188232275749&usg=AOvVaw0clIXR2u6vdhxqRdqoPmHE) on Vast.ai and start running workflows at scale. It's that easy.


ComfyUI has become one of the most popular open-source AI tools for image generation, for good reason.

Running ComfyUI on Vast.ai Serverless: Simple, Scalable Image Generation


Vast.ai is thrilled to announce the launch of our Serverless offering for GPU workloads, enabling automatic, pay-per-use scaling for AI inference without hyperscaler pricing or fixed infrastructure constraints.

With [Vast.ai Serverless](https://vast.ai/products/serverless), users run inference workloads through a fully serverless API on Vast's globally distributed GPU cloud. That means no manual instance management and no capacity planning. Instead, the platform automatically applies predictive optimization and flexible scaling, with access to a diverse GPU fleet -- so teams can deploy production AI systems that scale when needed and yet stay cost-efficient at all times.

How does this serverless model work in practice? Let's take a closer look!

<div style={{ margin: "48px auto", maxWidth: "960px", width: "100%" }}>
  <div style={{ position: "relative", paddingBottom: "56.25%", height: 0, overflow: "hidden", borderRadius: "12px", background: "#000" }}>
    <iframe
      src="https://www.youtube-nocookie.com/embed/0PAPzSZa3tA?rel=0"
      title="Vast.ai Serverless overview"
      allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share"
      allowFullScreen
      referrerPolicy="strict-origin-when-cross-origin"
      loading="lazy"
      style={{ position: "absolute", top: 0, left: 0, width: "100%", height: "100%", border: 0 }}
    ></iframe>
  </div>
</div>

### Where GPU Cloud Meets Serverless AI

Vast.ai Serverless provides access to a wide range of GPUs, from RTX-class consumer GPUs to enterprise-grade machines like A100s, H100s, and B200s. Rather than managing individual instances, all you have to do is define performance targets, and Vast handles the rest.

Behind the scenes, Vast.ai continuously benchmarks GPUs across a global network of more than **17,000 GPUs hosted by 1,400+ providers in over 500 locations worldwide**. As workloads run, the platform dynamically selects, provisions, and routes jobs to the most efficient hardware available at that moment.

The key is predictive optimization.

### Predictive Optimization Built for Real-World Workloads

Most serverless platforms react to demand after it appears. Vast.ai Serverless goes further.

Our predictive optimization feature analyzes historical usage patterns, real-time load, and ongoing market benchmarking to anticipate demand before it peaks. Based on those signals, the platform **proactively provisions GPU workers** that balance cost and latency -- ready to activate on demand rather than scrambling to scale once performance has already degraded.

These reserve workers help avoid the laggy cold starts and unpredictable costs common with other GPU serverless offerings. Not to mention having to pay for excess GPU capacity that isn't actively being used.

### Multiple Workergroups per Endpoint

Unlike traditional serverless systems that restrict endpoints to a single hardware profile, Vast.ai Serverless [supports multiple Workergroups per Endpoint](https://docs.vast.ai/documentation/serverless/architecture). A Workergroup is a collection of GPU workers that are managed as a logical unit to automatically adjust capacity to meet demand.

In our Serverless offering, each Workergroup in an Endpoint can specify different GPU types or hardware configurations. This lets a single API endpoint serve workloads using whichever Workergroup is most cost-effective or the best fit for your performance targets. For example, lighter requests might route to consumer GPUs, while heavier inference jobs scale onto H100s -- with no manual intervention needed.

In practice, this makes it possible to optimize performance and cost-efficiency in real time within a single deployment. And that brings us to pricing.

### Lowest-Cost Transparent Serverless Pricing

Vast.ai Serverless is the **lowest-cost autoscaling GPU cloud** on the market today.

Our Serverless workloads are billed per second, with support for On-Demand, Interruptible, and Reserved pricing. There are [no tiers and no limits](https://docs.vast.ai/documentation/serverless/pricing) -- and all you need is $5 to get started.

Because Vast draws from a competitive global market of GPU providers, pricing reflects real supply and demand rather than preset SKUs. As demand scales up, Vast.ai Serverless automatically selects the most cost-effective GPUs available. When demand drops, resources are released immediately, and billing stops.

As a result, teams can run production workloads at a fraction of the cost compared to traditional providers with centralized infrastructure. However, lower costs don't mean lower standards. Security and compliance remain core to our platform.

### Enterprise-Ready Security and Compliance

Vast.ai is built with security and compliance as foundational principles. Our platform is backed by **SOC 2 Type II** [certification](https://vast.ai/article/vast-soc2-typeII-certification), with ongoing audits every 12 months to ensure continual coverage.

For customers with the highest security requirements, Vast.ai offers a tiered compliance structure with our [Secure Cloud](https://cloud.vast.ai/?secure_cloud=true) offering. In the Secure Cloud, workloads run on isolated instances with direct SSH, CLI, and API access, hosted by our vetted datacenter partners that meet **ISO 27001** standards at minimum. Many partners maintain additional certifications. (Full details are available on our [compliance page](https://vast.ai/compliance).)

Other enterprise security features -- including private VPN access, optional audit trails, and enterprise-grade compliance support -- can be enabled as needed.

In all cases, data sovereignty remains fully in your control. Models, data, and workloads persist only as long as you choose and can be deleted at any time.

### How to Automatically Scale GPU Compute for AI Inference

With Vast.ai Serverless, you can go from [zero to compute](https://docs.vast.ai/documentation/serverless) in seconds. In short, here's how the process works:

1. Bring your own machine learning model
2. Send an inference request
3. Vast.ai Serverless routes and provisions compute on ready GPUs
4. GPU executes and returns results
5. Serverless predictively scales up and down based on usage

That's all there is to it! From request to response, Vast.ai handles all of the heavy lifting. You also gain access to ample metrics and debugging tools, including logs and Jupyter/SSH.

### Get Started with Vast.ai Serverless

Our Serverless offering makes it easy to run AI inference without managing infrastructure or overpaying for capacity. Vast.ai Serverless gives you a simple and straightforward path from experiment to production -- and it's ready when you are.

Leverage Vast's distributed GPU fleet on your own terms. Check out our [Serverless Overview](https://docs.vast.ai/documentation/serverless) and start building today!


Announcing Vast.ai Serverless for GPU inference with predictive optimization, multi-workergroup endpoints, pay-per-second pricing, and enterprise-grade compliance.

Posts about: Serverless

Introducing the Vast.ai Startup Program

Vast.ai's 2025 Year in Review

Running ComfyUI on Vast.ai Serverless: Simple, Scalable Image Generation

Vast.ai Serverless: Automated GPU Scaling for AI Inference - Without the Overhead