The Building Blocks of AI: Inside the Open Stack | ZML, LiveKit, ClickHouse & Neo4j | RAISE 2026

Panel Overview

• The session features leaders from the AI infrastructure stack: ZML (chip-level), ClickHouse & Neo4j (data/database), and LiveKit (voice/video application layer).
• All four companies emphasize open-source strategies as a core pillar for the new wave of AI solutions.
• The panel identifies a shift from "Old Regime AI" (Jan 2023 and earlier) to "Agentic AI," driven by the need for software to interact naturally like humans.

ClickHouse (Alexey Kurtenkov)

• Function: An open-source database designed to process quadrillions of records and hundreds of petabytes of data in real time.
• Performance: Executes queries 10 to 100 times faster than legacy database systems.
• Valuation: Recently reached a $15 billion valuation.
• Adoption: Serves major US and European entities including OpenAI, Anthropic, SpaceX, Visa, and Capital One.
• Differentiation: Clients utilize ClickHouse exclusively for massive data volumes because alternative databases cannot handle the scale of events, logs, and metrics required.

Neo4j (Philippe)

• Function: A leading graph database that stores relationships (subjects, objects, and connections) rather than flat tabular structures.
• Value Proposition: Represents the "digital twin" of complex networks (biology, finance, supply chains) to provide high-fidelity context for AI agents.
• Financials: Generates over $250 million in annual recurring revenue (ARR).
• Valuation: Reached a $2 billion valuation in 2021.
• Adoption: Used by 80% of Fortune 100 companies; has enrolled over 700 AI startups in its support program.
• AI Utility: Serves as the knowledge, context, semantic, and long-term memory layer for AI systems, addressing the limitation of SQL in handling connected networks.

LiveKit (Russ)

• Function: An open-source platform for building voice, video, and physical AI agents, enabling real-time human-computer interaction.
• Infrastructure: Launched LiveKit Cloud in late 2022 as a global mesh network with sub-500ms latency.
• Strategic Partnership: OpenAI secretly integrated LiveKit Cloud to power ChatGPT Voice, Voice Advanced, and Vision modes.
• Market Presence: Supports over 400,000 developers globally.
• Valuation: Achieved a valuation exceeding $1 billion.
• Vision: Argues that as computers become more human-like, interaction will shift from text to voice and visual reaction, necessitating a reimagined full-stack infrastructure beyond traditional web protocols.
• Sovereignty: Deployed as open source for sovereign projects, including French government video solutions and US emergency services (911).

ZML (Sebastien Legarde)

• Function: An AI engineering lab providing an inference stack that supports running models across diverse hardware architectures.
• Hardware Support: Currently supports 8 chip architectures, with 12–15 expected by year-end (NVIDIA, AMD, Intel, TPU, Metal).
• Recent Release: Launched the industry's fastest LLM server yesterday, emphasizing universal compatibility across all major hardware types.
• Thesis: Enterprises require operational control over AI similar to databases; ZML aims to provide this by existing at every layer of the stack to achieve 100x performance gains.

The AI Infrastructure Stack & Economics

• Layer Expansion: Panelists propose adding a "Harness Layer" (orchestration, memory, tooling) to Jensen Huang's five-layer model to address undifferentiated application infrastructure problems.
• Data as Moat: Data is identified as the primary long-term differentiator; while models and code become commoditized, enterprise data accumulation remains unique and non-fungible.
• Token Pricing: Contrary to predictions of token prices dropping to zero, costs are rising due to GPU constraints and high demand for agentic workloads.
• Commoditization Risks: Applications face rapid commoditization; future value will reside in brand equity, network effects, and the inability to be trivially copied.
• Hardware Margins: The panel discusses the challenge of competing against NVIDIA's ~91% margins, necessitating software composability and alternative hardware support.

Open Source in the AI Era

• AI's Impact: Panelists argue AI accelerates open source by lowering contribution barriers and enabling AI-assisted code review and experimentation.
• Defensibility: Sustainability relies on community, brand, and proprietary test suites (e.g., 500,000+ tests) rather than code secrecy, which is easily replicable by AI agents.
• Legal Risks: Concerns exist regarding AI-generated code copyright status (US law currently denies copyright) and the potential for license evasion via code rewriting.
• Open Weight vs. Open Source: A distinction is made between proprietary models, open-weight models (weights available), and true open source (weights, scripts, and datasets available); the latter is rare but desired.

Sovereignty and Geopolitics

• European Renaissance: A shift is occurring where funding and talent for deep tech are becoming fungible globally, challenging the US monopoly on software infrastructure (e.g., ZML founding in France).
• Sovereign AI Drivers: Enterprises and nations are driven by the need for data residency, compliance (GDPR), and supply chain resilience ("not being unplugged").
• Cost Curve: Building sovereign infrastructure becomes exponentially more expensive at the hardware level (silicon), making open source software a more viable path for national and enterprise sovereignty.
• Implementation: Sovereignty is being enacted via open-source deployments for critical infrastructure (e.g., French government video, US 911 systems) to avoid vendor lock-in.