and the distribution of digital products.
State of Phala Network Q3 2024
- Phala conducted the first-ever benchmark study on Nvidia's TEE-enabled H100 GPUs. The study revealed minimal performance impact — less than 7% for typical AI tasks — and excellent scalability for larger models.
- Phala processed over 66 million contract executions in Q3 — a 3% decrease from the previous quarter.
- Monthly active users (MAUs) dipped 5% QoQ. MAUs peaked in March 2024, when Phala contracts logged 774 unique users.
- Total Phala supply-side network participants (workers) expanded by 8% in Q3 2024. This growth kicked off after the Phala blockchain went live in April 2023.
- Phala partnered with multiple AI-focused protocols to drive the usage of AI agent contracts: Hyperbolic AI, Morpheus, Lumerin, Allora, Artela, Aligned Layer, Theoriq, and io.net.
Phala Network (PHA) is an AI coprocessor that enables smart contracts to control and interact with AI Agents. It uses a blockchain-TEE (trusted execution environment) hybrid design to incorporate offchain elements — such as artificial intelligence (AI) — into blockchain-focused applications. This design creates the foundation for onchain smart contracts to control AI agents (run by offchain worker nodes) while maintaining security and user privacy. Phala consists of four core components:
- Decentralized Cloud Computing Network: Phala is a backend host of scalable blockchain technology and a trusted execution environment (TEE) to ensure the verifiability and security of compute execution.
- TEE (SGX) Framework: The core Intel TEE architecture scheme that ensures secure and verifiable code execution within an isolated hardware-protected area of a computer’s processor.
- AI Agent Contracts: AI computing programs that enable EVM smart contracts to extend their capabilities offchain.
- Blockchain/Phala Nodes: Attested secure worker nodes that retrieve and execute transactions offchain. These nodes then use the blockchain to organize the decentralized network and incentivize workers.
In 2020, Phala secured a funding round at a $10 million valuation with participation from Candaq Group, InfiChain, Blue Mountain Labs, Incuba Alpha Group, and Exoplanet Capital. Its product was developed to offer AI-enabled, Turing-complete smart contracts, aiming to address the challenges posed by AI and centralized cloud computing. Following this momentum, Phala won a Polkadot parachain slot auction with 1,685 contributors in March 2022. As of Q3 2024, Phala is an AI coprocessor that provides a secure and verifiable compute solution for integrating crypto and AI at the application layer. For a full primer on Phala, refer to our Initiation of Coverage report.
Website / X (Twitter) / Discord
Key Metrics
Phala Produces First-Ever GPU-TEE Benchmark StudyIn 2024, Nvidia launched the H100 Tensor Core GPUs with advanced Confidential Computing (CC) capabilities. These GPUs — featuring TEE functionality built directly into the architecture — include encrypted memory, secure boot, and hardware firewalls, ensuring sensitive data remains protected during processing.
Phala's employment of Nvidia's TEE-enabled GPUs underscores a key advancement in decentralized AI: a foundation for secure, transparent AI systems that are not controlled by any single entity. By partnering with decentralized cloud computing networks io.net and Hyperbolic, Phala aims to broaden access to global GPU resources. This move aligns with the broader goals of decentralized AI, which aims to democratize access to AI technologies.
Phala’s recent benchmark on TEE-enabled Nvidia H100 GPUs offers valuable insights into secure AI performance. Highlights from the benchmark include:
- Low overhead: The integration of TEE results in a performance overhead of under 7% for most LLM queries, with negligible impact on larger models such as LLaMA-3.1-70B. For smaller models, the overhead mainly stems from CPU-GPU data transfers over PCIe rather than GPU computation.
- High scalability: The report demonstrates that, for larger AI models, the overhead becomes insignificant, showing that TEE mode does not hinder the GPU's compute-intensive tasks.
These findings confirm that TEE-enabled GPUs are a viable solution for developers aiming to create secure, decentralized AI applications without sacrificing performance.
Performance Analysis Usage
AI Agent Contracts are AI computing programs that enable EVM smart contracts to extend their capabilities offchain. AI agent contracts are coded in TypeScript/JavaScript and Phala nodes execute these contracts offchain — though the requests originate onchain. Contract executions (by Phala worker nodes) measure demand for Phala AI Agents
In Q3 2024, contract executions more than doubled compared to the same period last year, but fell by 3% from the previous quarter. Phala processed over 208 million contract executions year to date. Within Q3, Phala partnered with multiple AI-focused protocols: Allora, Artella, and Theoriq. Executions are expected to keep rising as Phala’s recent partnerships and integrations take effect.
Phala increased the number of AI Agent Contracts deployed to its network by 44. This number should also grow alongside Phala’s recent Q3 partnerships, which include Morpheus and Lumerin. The addition of 44 new contracts should diversify the types of compute enabled by Phala, potentially attracting more demand from AI application protocols and their users in the coming quarters.
Monthly Active Users (MAU)
Monthly active users (MAU) measures the number of unique accounts interacting with Phala contracts on a monthly basis. Some users may use more than one account, which is not accounted for in the data. Phala’s monthly active users saw steady growth until peaking in March 2024, when it logged 774 unique users. Phala average MAUs was 510 in Q3 — a slight dip (-5%) from the previous quarter.
Service Providers
Worker Nodes provide compute resources to the network. They host AI Agent Contracts (offchain programs) inside the nodes’ TEEs. The TEE protects the workload for code integrity and privacy. To join the network, miners anywhere in the world can deploy Worker Node software clients inside their TEE-compatible hardware setups, getting them verified onchain.
The Phala mainnet currently operates on both the Khala and Phala blockchains. Khala is a parachain on Kusama (a canary network for Polkadot), and Phala is a parachain on Polkadot. The Kusama canary network is a mainnet with its own set of validators that protocols can launch on or use as testnet-like infrastructure. Because the coprocessor mainnet was launched before the Phala Polkadot parachain was live, the project uses Khala as an additional mainnet that will eventually migrate its miners to Phala.
Total workers increased 20% YoY, with growth largely starting when the Phala blockchain went live in April 2023. While both sets of workers grew into November 2023 — with Phala workers growing more rapidly, but Khala workers started to decline due to the Phala migration. Workers should continue to migrate to Phala. Growth in demand-side activity may also incentivize node runners to deploy more workers, growing the total in the future.
Phala has a non-consensus staking mechanism that manages up to 1 million CPU cores from over 100,000 nodes. Stake Delegation includes a feature called StakePool to connect computing nodes with PHA holders. As such, PoolOwners can create and manage StakePools allowing groups of workers to delegate their PHA to the pool. Workers must have a certain amount of stake to perform network operations — thus, having more PHA staked/delegated increases worker capacity.
The number of delegators measures the number of PHA holders that are contributing stake to the network. Though workers have grown steadily throughout the year, delegators have remained largely stagnant at about 3,500. That said, the percentage of delegators on Phala has been trending upward since the Phala mainnet launched. It may not be intuitive, but a growing number of delegators — or at least the percentage of delegators — on Phala should further attract more worker deployments on the network, as workers seek to gain delegator stake.
Rewards
In Q3 2024, both operator rewards and the PHA price declined. Operator rewards decreased by 27% and the PHA price dropped by 10%, reflecting overall market contraction. This decline aligns with a 7% decrease in the circulating market cap, indicating a challenging period for operator incentives relative to the token’s market performance.
Agent Wars
Agent Wars is a Web3 game that enables users to launch their own AI agents with tokenomic structures tied to the agent in the form of keys (similar to friend.tech). Agents are represented as an ERC-721 NFT with a unique avatar linked to an X account. An agent is created through a three-day onchain auction, with bids based on the account's follower count. The auction funds then cover the hosting and training costs. While Agent Wars is currently only operating on Base, it has seen steady growth since starting in mid-May 2024. It ended the quarter with 322 AI Agent contracts, 70 users, and $14,000 in TVL.
Qualitative AnalysisKey EventsHackathonsPhala ran two hackathons during Q3: one in July and one in September at ETHGlobal. At EthGlobal in Singapore, 26 teams submitted projects to Phala’s bounty. For the July hackathon, 10 teams submitted their projects to Phala. One standout hackathon project was WarpAds, which leveraged AI agent contracts to generate Facebook Ads for Farcaster.
Partnerships and IntegrationsThroughout Q3 2024, Phala has made some key partnerships and integrations to help accelerate its role in pioneering decentralized AI.
Hyperbolic AIHyperbolic is a cloud computing marketplace that provides GPU access and inference services to AI projects. Phala partnered with Hyperbolic to run AI models and inference services on its coprocessor and provide compute to the Hyperbolic open-access AI Cloud via its TEE-based worker nodes. Phala’s integration with Hyperbolic will enable Hyperbolic’s users to deploy and verify AI models onchain.
Morpheus and LumerinMorpheus incentivizes a network of general-purpose AI agents (i.e., Smart Agents) that can execute smart contracts on behalf of users. Lumerin was originally designed to manage hash power distribution in Bitcoin mining; however, today Morpheus uses Lumerin’s architecture as core node software to decentralize and allocate AI compute. Phala runs the Lumerin Node inside its H100 TEE, integrating its coprocessor into the Morpheus network.
AlloraAllora built a self-improving, decentralized AI network where AI agents use a particular forecasting task to predict the performance of other AI models under live conditions. In short, Allora is a marketplace that incentivizes data scientists to offer high-quality inferences (predictions or decisions based on a trained AI model). Phala partnered with Allora in a collaboration that adds a privacy layer (via TEEs) to Allora’s machine learning processes, while Allora powers AI Agents on Phala with market-responsive inferences.
ArtelaArtela is an extensible blockchain with parallel execution and modular VMs. Artela enables developers to build EVM-based smart contracts while adding WASM-native extensions to their applications. Phala partnered with Artela to provide an AI coprocessor layer to the chain that will facilitate AI-enabled native extensions.
Aligned LayerAligned is a decentralized network that validates Zero-Knowledge and validity proofs, then submits the verified results to the Ethereum blockchain. Its design focuses on achieving high throughput, cost-effective proof verification, and minimal latency. Phala Network is integrating its Trusted Execution Environment (TEE) attestation and advanced zero-knowledge proof (ZKP) techniques into Aligned Layer—enhancing secure, scalable verification and enabling privacy-preserving computations for decentralized applications across various sensitive data-driven industries.
TheoriqTheroiq is a protocol designed to streamline the coordination and communication between AI agents. Phala has partnered with Theoriq to integrate Phala's Trusted Execution Environment (TEE) capabilities, enabling secure, decentralized AI agents to seamlessly operate within Web3, advancing the governance, security, and monetization of multi-agent systems on blockchain.
io.netio.net is a decentralized physical infrastructure network (DePIN) that focuses on providing GPU resources for machine learning (ML) and artificial intelligence (AI) applications. Phala Network's new partnership with io.net will enhance secure, decentralized AI by integrating Phala’s TEE technology with Nvidia's advanced GPUs via io.net’s cloud network. This collaboration provides developers with affordable access to high-performance GPU resources, enabling robust, privacy-focused AI computation without compromising on security.
Closing SummaryIn Q3 2024, Phala achieved significant milestones with the launch of its GPU-TEE benchmark study and the hosting of multiple hackathons. While some key metrics — such as operator rewards, monthly active users, and contract executions — saw slight declines, the network demonstrated growth in other areas, including contract deployments, worker numbers, and delegator count. Phala also expanded its ecosystem through strategic partnerships with seven AI and blockchain-focused protocols. Looking ahead, Phala’s GPU-TEE advancements and new partnerships position it to meet growing demand for secure, decentralized AI. These developments could drive greater network adoption and industry integration in the coming quarters.