# AI SPE Phase 2: Retrospective **URL:** \ **Category:** Treasury\ **Created:** [December 19, 2024, 5:01pm UTC](https://forum.livepeer.org/t/ai-spe-phase-2-retrospective/2691 "2024-12-19T17:01:18Z")\ **Posts on this page:** 1\ **Page:** 1
**Author:** ![rickstaa](https://yyz2.discourse-cdn.com/flex030/user_avatar/forum.livepeer.org/rickstaa/32/917_2.png) [@rickstaa](https://forum.livepeer.org/u/rickstaa)\ **Post date:** [December 19, 2024, 5:01pm UTC](https://forum.livepeer.org/t/ai-spe-phase-2-retrospective/2691/1 "2024-12-19T17:01:18Z")
## Introduction At the start of this year, the [AI SPE](https://www.notion.so/15c0a348568780b1bbdff3eb71e472e4?pvs=21) set out with a mission to **validate the impact, potential, and competitiveness of bringing AI based video compute jobs onto the Livepeer network** by driving adoption through real-world deployments, as outlined in [this treasury proposal](https://explorer.livepeer.org/treasury/82843445347363563575858115586375001878287509193479217286690041153234635982713). Our initial focus was on proving the **cost-efficiency** and **performance** of new generative AI video capabilities on the network. Since then, we’ve collaborated closely with the community, design partners, and startups to accelerate adoption and demonstrate the potential of AI jobs on Livepeer. During the first phase, we harnessed Livepeer’s decentralized payment and compute infrastructure to deploy generative AI pipelines, enabling orchestrators to process AI jobs and earn initial fees while allowing pioneering builders to experiment and innovate on an alpha-stage network. This effort resulted in a stable network and a successful prototype launch at ETH Berlin. As outlined in our [first-phase retrospective](https://forum.livepeer.org/t/ai-spe-stage-1-retrospective/2417), these pipelines were a crucial first step toward our broader mission of **establishing Livepeer as the leading decentralized platform for media-related AI compute jobs within three years.** Building on this foundation, the second phase of Livepeer’s AI journey, as detailed in [this treasury proposal](https://explorer.livepeer.org/treasury/42084921863832634370966409987770520882792921083596034115019946998721416745190), focused on enhancing AI subnet performance, improving developer experience, scaling the network, and expanding service offerings. We prioritized delivering **media-centric AI pipelines** , such as Whisper, Text-to-Speech, and Segment Anything 2, enabling expanded GPU support and implementing key network upgrades for scalability and reliability. We also released developer tools, including SDKs and documentation, while strengthening open-source collaboration through bounty programs, hackathons, and ecosystem initiatives. Shifting from a foundational approach in the first phase, the second phase embraced a **demand-driven** strategy by collaborating closely with startups from the Startup Program and Hackathon participants. Real-world use cases directly informed our development roadmap, driving increased demand and higher fees on the network while addressing practical challenges and gathering valuable insights that shaped the 2025 roadmap, as outlined in the [Livepeer Cascade post](https://mirror.xyz/livepeer.eth/bCruUtv0PJWWlFxfbCMATGM_h15hUncwvyFW1A3z7Ag). These insights, supported by extensive market research conducted by the INC product and leadership teams, ultimately led to a **strategic focus on real-time AI video** —a domain where our existing network is uniquely positioned to excel and potentially become a market leader. In this forum post, we reflect on the key milestones, share critical learnings, provide a financial overview, and outline next steps as we transition into the third phase of the AI SPE journey. ## Achieved Milestones Since our second [treasury proposal](https://explorer.livepeer.org/treasury/42084921863832634370966409987770520882792921083596034115019946998721416745190) passed in May 2024, we have steadily progressed toward its deliverables and milestones. Despite some engineering tasks being deprioritized due to shifting strategic priorities, we collaborated closely with the community to drive significant improvements in the Livepeer network’s scalability, stability, and capabilities, while enhancing developer accessibility through robust SDKs and improved documentation—positioning Livepeer AI for success in 2025. In parallel, the ecosystem team effectively showcased these advances through events, competitions, hackathons, and programs, boosting community engagement, social interactions, and open-source contributions. Originally scheduled to conclude on September 15th, this phase was extended to December thanks to Sarah’s exceptional financial leadership, which enabled additional deliverables. We also want to acknowledge Doug, Rich, and Shannon for their invaluable guidance on the project vision and roadmap, as well as the rest of the ecosystem team for their continued operational support. Lastly, we thank Livepeer.cloud for providing the public gateway. The sections below detail all engineering and ecosystem key deliverables accomplished during this phase, together with the main stakeholders. For a more comprehensive overview of engineering progress and individual contributions, please refer to our [changelog](https://livepeer-ai.productlane.com/changelog) or review the release notes in the associated [Livepeer AI repositories](https://github.com/livepeer). ## Main deliverables 1. **Enhanced AI Subnet Performance, Metrics, and Orchestrator-Gateway Operations** 2. **Provided a Clear Entry Point for App Developers and Expanded Service Offerings** 3. **Launched the First Livepeer Software Bounty Program** 4. **Launch of the First Cohort in the Livepeer AI Startup Program** 5. **Hosted the First Livepeer AI Hackathon and Multiple Successful Side Events** 6. **Extra deliverable: Implemented the first realtime AI pipelines** ## Postponed Initiatives As we executed our roadmap, we made strategic adjustments in response to emerging challenges and evolving opportunities. While prioritizing core infrastructure upgrades and operational improvements, some features outlined in our [Back-end Deliverables](https://www.notion.so/978f0d9adbf646dd8b9dbff65d366625?pvs=21) were deprioritized due to resource reallocation. The most notable of these was the **Compute-Based Pricing Mechanism** , intended to enable orchestrators to set prices based on actual compute usage rather than per pipeline. Similarly, **On-Chain Ticket Distinction** , aimed at improving the accuracy and reliability of the [Network Fee Dashboard](https://dune.com/rickstaa/livepeer-ai), was deferred due to its complexity and resource needs. Initial efforts toward **VRAM Optimization** revealed trade-offs, such as slower processing speeds, leading to a temporary pause for further research. Additionally, work on a **Data Aggregator** for centralized job metrics and an **Efficient Job Distribution Mechanism** based on VRAM estimation and availability was postponed to ensure focus on higher-priority network upgrades. These features remain strategically important and are planned for inclusion in future proposals by our group or have been handed over to Livepeer INC. We remain committed to revisiting these initiatives, building on the robust technical foundation established during this phase, and aligning them with our roadmap to ensure continued progress. ## Network impact During the second phase of Livepeer’s AI journey, several promising signals emerged, highlighting the potential of running AI-powered workloads on the Livepeer network: - **41 AI Node Orchestrators** actively processed jobs, contributing to network growth and stability. - **Hundreds of thousands of jobs** were successfully completed across **8 distinct AI pipelines** , showcasing the network’s capacity to handle diverse workloads. - **8 Startups** were onboarded onto the AI network, accompanied by **multiple smaller applications** launched through hackathons and ecosystem initiatives, reflecting growing interest and adoption. - **Over [7,000 winning tickets](https://dune.com/rickstaa/livepeer-ai)** were paid out to orchestrators, amounting to approximately **9 ETH** (\>$24K at the current price). - **The highest-ever weekly payout** was recorded, highlighting the increasing economic opportunities for orchestrators and validating Livepeer’s viability as a decentralized AI compute platform. Although the network is still in its early stages and has not yet achieved exponential adoption, these milestones highlight its strong potential for scaling AI workloads on Livepeer. ## Key Learnings 1. **Scaling supply to match demand remains a challenge** 2. **Real Demand Generation Requires a Unique Differentiating Factor** ## Finance The AI SPE received 25K LPT in funding from the Livepeer Treasury for [Phase 2](https://explorer.livepeer.org/treasury/42084921863832634370966409987770520882792921083596034115019946998721416745190), bringing the total funding for 2024 to 50K LPT. Over the second half of 2024, the SPE spent over $435K in USD terms, bringing the total spend for the year to nearly $690K. The financial breakdown demonstrates a strategic allocation across contributors, infrastructure, and programmatic initiatives to drive forward the objectives outlined in the [Phase 2 pre-proposal](https://forum.livepeer.org/t/ai-video-spe-stage-2-pre-proposal/2419). ### Phase 2 Key Spending Highlights - **SPE Contributors** : The largest share of the budget — nearly $300K — went to AI-focused contributors, reflecting the core development effort. - **Infrastructure & SW** : Investments of nearly $58K enabled scaling and network improvements during H2 2024. - **Hackathon and Bounties** : Programs fostering decentralized participation, such as bounties and hackathons, totaled nearly $60K for Phase 2. We expect ~25K in additional expenses in the second half of December. For a detailed breakdown of historical spending, please refer to the [AI SPE public financials](https://docs.google.com/spreadsheets/d/1CCF_41_ymyP9hKNzEOW6mkEO24Edv4Ypo-IcOn9hLbw/edit?gid=298648302#gid=298648302). ### Holdings As of December 16, 2024, the AI SPE holds the following assets on its balance sheet: | **AI SPE BALANCE SHEET (as of December 16, 2024)** | **# Tokens** | **$ Value/Token** | **$ Value** | | --- | --- | --- | --- | | _Arbitrum_ | | | | | LPT | 690 | $16.50 | $11,382 | | ETH | 3.9 | $4,000 | $15,684 | | USDC | 13,777 | $1.00 | $13,777 | | Staked LPT | 7,270 | $16.50 | $119,955 | | _Ethereum_ | | | | | ETH | 0.49 | $4,000 | $1,960 | | USDC | 1 | $1.00 | $1 | | **Total Assets** | | | **$162,760** | The majority of our remaining assets are in the form of [LPT staked](https://explorer.livepeer.org/accounts/0x9004593Bea2CEcD59EF4c58a1920C60f70F70136/delegating) toward the [AI SPE Orchestrator](https://explorer.livepeer.org/accounts/0x16a72bdb3017196825bc53809b87f96fbee31f6c/orchestrating) to maintain its position in the top 100, support the network, and earn rewards on our idle assets. Throughout the year, holdings in LPT, ETH, and USDC were actively managed to support ongoing operations. ## Conclusion The second phase of the AI SPE has showcased the Livepeer network’s growing potential as a leading platform for AI-powered video compute. Through strategic infrastructure upgrades, robust developer tools, and dynamic ecosystem initiatives—including hackathons, the startup program, and other demand-focused efforts—we’ve built a solid foundation while gaining valuable insights to guide the next phase. Looking ahead, we will focus on scaling real-time AI offerings, enhancing network scalability and performance, and empowering developers to seamlessly integrate both real-time and batch media-related compute jobs into our permissionless network. These advancements will unlock new revenue opportunities for orchestrators, diversify service offerings, and strengthen Livepeer’s competitiveness in the decentralized AI landscape. A detailed proposal with key milestones and our [2025 roadmap](https://www.notion.so/15c0a348568780b1bbdff3eb71e472e4?pvs=21) will be shared soon, charting the path toward establishing Livepeer as the premier decentralized platform for media-related AI compute. With several core contributor groups, many open-source developers, and Livepeer INC united behind this common goal, we are better positioned than ever to succeed. The lessons learned and the groundwork laid in earlier phases have paved the way for broader adoption, innovative applications, and sustained growth in the years ahead. We extend our heartfelt gratitude to the Livepeer community for your unwavering support, insightful feedback, and invaluable contributions. Your dedication drives our progress, and we look forward to building the future of decentralized AI—together 🚀.