NVIDIA

Just now: NVIDIA plans $1B investment in AI coding startup led by Copilot veteran, valued near $100B in just two years

Honghao Wang

— 4 min read
Just now: NVIDIA plans $1B investment in AI coding startup led by Copilot veteran, valued near $100B in just two years

NVIDIA to Invest Up to $1 Billion in AI Coding Startup Poolside

On October 30, Bloomberg — citing informed sources — reported that NVIDIA plans to invest up to USD 1 billion in AI startup Poolside, potentially quadrupling its valuation.

image
image
image

---

Funding Round Details

  • Target Raise: USD 2 billion
  • Pre-money Valuation: USD 12 billion
  • NVIDIA Contribution: At least USD 500 million (up to USD 1 billion if the round closes successfully)
  • Current Commitments: Over USD 1 billion, with USD 700 million from existing investors

Poolside develops an AI-powered coding assistant designed to enhance coding and debugging efficiency.

Neither NVIDIA nor Poolside have commented publicly on the report.

---

1. Leadership & Rapid Growth

Founding & Relocation

  • Founded: 2023 in San Francisco, USA
  • Current HQ: Paris, France
  • Focus: Generative AI for software development and the long-term pursuit of AGI (Artificial General Intelligence)

Founders

Jason Warner (CEO)

  • Former GitHub CTO
  • Founded GitHub Copilot and oversaw GitHub Actions, Packages, Codespaces, and security tools
  • Led engineering teams at Canonical and Heroku
image

Jason Warner

Eiso Kant (CTO)

  • Serial entrepreneur
  • Co-founded Athenian, an engineering analytics startup
image

Eiso Kant

The co-founders met in 2017 and spent six years planning AI-driven developer tools before launching Poolside.

---

Vision for Specialized AI

Warner left Microsoft believing AI is underestimated in its impact on software development. He sees the need for specialized AI models — not generic ones like GPT — to achieve full program synthesis.

> "The future lies in AI built specifically for software development, with deeper reasoning and context-awareness."

---

Industry Context: AiToEarn Example

Platforms like AiToEarn官网 demonstrate a convergence of AI for development and content monetization — enabling content creators to:

  • Generate AI content
  • Publish cross-platform
  • Analyze and track AI model performance
  • Rank models and optimize output

---

Funding History

  • May 2023 – Seed: USD 26M led by Redpoint
  • Aug 2023 – Series A: USD 100M led by Felicis, Redpoint China, and Xavier Niel
  • Oct 2024 – Series B: USD 500M led by Bain Capital Ventures, valuing Poolside at USD 3B

---

2. NVIDIA’s Interest in Poolside

RLCEF: Reinforcement Learning from Code Execution Feedback

Poolside’s proprietary RLCEF methodology:

  • Trains AI by writing and running code
  • Provides deterministic feedback for iterative improvement
  • Explores millions of tasks across 130k real codebases
  • Expanded training cluster to 10,000 GPUs

> Founders believe AGI in software is achievable with unlimited data and compute.

---

Technical Strategy

Key Differentiator: Full self-development of data, algorithms, and training frameworks:

  • Avoids reliance on pre-trained open-source models
  • Enables deep optimization for software-specific scenarios
  • Maximizes control over the model stack

---

3. Solving Data Quality Bottlenecks

Problem: Only ~10% of open-source code is high-quality for model training

Solution:

  • Automated pipeline scans and filters repositories
  • Generates synthetic training data
  • Uses positive/negative reinforcement to optimize correctness

Impact:

  • Processed over 200,000 repositories
  • Estimated 3 trillion tokens of high-quality code globally
  • Target: 1.5 trillion synthetic tokens for top-tier model training

---

4. Model Iteration & Deployment

Parallel Experimentation

  • Multiple small-scale experiments with architectures, loss functions, and optimization methods
  • Select best candidates before scaling up
  • First Core Model Launch: Expected in 2025
  • Pre-trained with RLCEF and synthetic data

Current Offerings

  • Foundational model + Maidu AI assistant via AWS Bedrock
  • Deployable in client cloud environments (e.g., AWS VPC)
  • Ensures enterprise data privacy

Capabilities:

  • Chat-based coding assistance
  • Autocompletion & explanations
  • Documentation generation
  • IDE support (e.g., Visual Studio, VS Code)
  • RAG-like mechanism for legacy systems
  • Role-based access control for sensitive data

---

Measuring Productivity

Monteiro: "Productivity is not a single number."

Key metrics:

  • Usage frequency
  • Feature invocation rates
  • Developer adoption & feedback quality

---

Looking Ahead

Monteiro’s vision:

> "The developer of the future will co-build systems with AI — not just write code."

Poolside aims to create an AI that genuinely understands software development tasks.

In parallel, platforms like AiToEarn开源地址 are emerging to connect model output with global content distribution and monetization.

---

References

---

Disclaimer: Translated and collated by InfoQ. Reproduction without permission is prohibited.

---

---

WeChat Video Channel:

Follow for updates on:

  • Technical trends
  • Startup experiences
  • Lessons from business pitfalls
image

---

Reading this rewritten version should be easier to navigate thanks to the clear headings, bullet lists, and emphasis on technical concepts. The next logical enhancement could be condensing certain funding details into a visual timeline chart for quicker reference. Would you like me to prepare that?

Read more