# Infrastructure

We're building the next generation of composable AI infrastructure, a decentralized network of GPU clusters that provides the simplicity of serverless platforms. Our vision is to democratize access to GPU computing by creating a scalable platform where developers can run AI workloads without managing infrastructure complexity.

#### The Problem 

Today's AI developers face a difficult choice:

* **Serverless platforms** (Lambda, Modal): Easy to use but limited GPU access and control
* **Dedicated instances** (RunPod, AWS EC2): Full GPU control but complex management and scaling
* **Managed services**: Expensive and vendor-locked

We're eliminating this by creating composable AI infrastructu**re** that gives you the best of both worlds.

**The Solution: Taste Metal**

* **Composable**: Mix and match different GPU types, models, and workloads
* **Decentralized**: Distributed across multiple clusters and regions for reliability and performance
* **Serverless Experience**: Submit jobs and get results without managing infrastructure
* **GPU-Optimized**: Built specifically for AI workloads, not retrofitted from general-purpose cloud
* **Cost-Effective**: Pay for what you use, with intelligent resource sharing
* **Web3-Native**: Built-in asset registry with provenance, lineage, and ownership protection

### How It Works

####  **Submit & Forget**

<figure><img src="https://1865526980-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FmU77N3i0x34qj4of1y6c%2Fuploads%2F9ADrduUHQHlvgPNquvZ8%2FTASTE-Metal-Infrastructure_3%402x.png?alt=media&#x26;token=b1ede2a7-6845-4a6c-bd60-78febd4807de" alt=""><figcaption></figcaption></figure>

* Submit your AI workload through a simple API call or queue message
* Our system automatically finds the best GPU resources across our decentralized network
* Get results back without thinking about infrastructure, scaling, or resource management

**Which means:**

* **No infrastructure management**: Focus on your AI models, not servers
* **Automatic scaling**: Handle 1 request or 1000 requests seamlessly
* **Global distribution**: Your workloads run where it makes the most sense

**Shared Intelligence Across the Network**

<figure><img src="https://1865526980-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FmU77N3i0x34qj4of1y6c%2Fuploads%2FbBCuxJoyLTnZos7PP4yf%2FTASTE-Metal-Shared-Intelligence_1%402x.png?alt=media&#x26;token=bfbdc0e9-10b9-4b6a-bbf1-1af7ae2c0850" alt=""><figcaption></figcaption></figure>

* Once you upload a model, it's available everywhere (but still private to you)
* Models cached in Sydney are instantly available in Singapore
* Foundation models and supporting assets are stored once
* Popular models are automatically cached where they are needed

**Which means:**

* **Faster startup**: No waiting for model downloads
* **Cost savings**: No paying for duplicate storage across regions
* **Global reach**: Deploy your AI workloads anywhere in the world
* **Collaboration**: Share models and workflows with your team globally

**Blockchain Asset Registry**

<figure><img src="https://1865526980-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FmU77N3i0x34qj4of1y6c%2Fuploads%2FEgsTgVkXSu1EO1j8t1DO%2FTASTE-Metal-Blockchain-Asset-Registry%402x.png?alt=media&#x26;token=1c8d35fd-1150-4be3-bbc3-a97dd2f57a68" alt=""><figcaption></figcaption></figure>

At the core of our platform is Taste Protocol, a revolutionary asset registry that ensures **complete provenance** and **ownership protection** for all AI workloads:&#x20;

**Which means:**

* **Data ownership**: Your models, datasets, and AI outputs are cryptographically yours
* **Provenance tracking**: Complete lineage from input data to final AI model
* **Access control**: Granular permissions for who can use your AI assets
* **Compute-to-data**: AI workloads run on your data without compromising ownership
* **Audit trails**: Immutable records of all AI operations and transformations