On October 30, according to Bloomberg citing sources, NVIDIA plans to invest up to $1 billion in AI startup Poolside, a deal expected to quadruple the latter's valuation.
Sources said Poolside is negotiating a new funding round at a $12 billion pre-money valuation to raise $2 billion. NVIDIA plans to contribute at least $500 million, potentially reaching $1 billion if the round closes successfully.
Reports indicate Poolside has secured over $1 billion in commitments for this latest round, including about $700 million from existing investors.
Poolside provides an AI-powered coding assistant that helps developers improve code writing and debugging efficiency.
As of press time, neither NVIDIA nor Poolside has responded to media comment requests.
So, what makes this AI startup attractive to NVIDIA?
Microsoft tech guru leads the team, valued at nearly $30 billion in two years
Poolside AI was founded in 2023, initially headquartered in San Francisco, California, USA, with headquarters and operations later moved to Paris, France. The company focuses on the intersection of generative AI and software development, aiming to build an "AI coding assistant" and, in the long term, "general artificial intelligence."
Founders Jason Warner and Eiso Kant are both software engineers. Warner serves as CEO and Kant as CTO at Poolside.
Before founding Poolside, Warner was GitHub's CTO, where he incubated the AI tool Copilot and led products like GitHub Actions, Packages, Codespaces, and Advanced Security. He also led engineering teams at Canonical and Heroku.
Jason Warner
Kant is a serial entrepreneur who previously co-founded several developer-focused startups, including engineering analytics firm Athenian.
Eiso Kant
Warner met Kant in 2017. Over the next six years, they planned an AI-driven toolkit for developers, which evolved into Poolside.
In an interview, Warner explained his motivation for leaving Microsoft to found Poolside.
Warner believes the industry underestimates AI's disruptive impact on software development. He is convinced the future lies in building AI specialized for software development, rather than relying on general models (like those behind GitHub Copilot, the GPT series). Achieving "full program synthesis" (AI automatically generating complete programs) requires specialized AI focused on software, not general models—this was the key motivation for starting an independent company.
Warner said: "Microsoft is no longer innovating, and hasn't for decades. Microsoft is a private equity company holding assets across various fields. Its partnerships with OpenAI, acquisitions of GitHub, LinkedIn, Minecraft, etc., show it's no longer a software development company but a private equity firm. They integrate resources from various industries, which holds them together and gives them their market position. They do it better than anyone. They're ahead of the curve in AI, leading all competitors. As long as they maintain this momentum—which I see no change in—they're in a very strong position. Other companies entering this space have missed the wave."
Poolside has developed its own AI models for tasks like code autocompletion and recommending relevant code based on specific contexts or codebases—similar to competitors' AI coding assistants. Customers are primarily Fortune 2000 enterprises and public sector institutions; few have been publicly disclosed.
As of March this year, Poolside has nearly 100 employees.
Public information shows the company has completed three funding rounds since inception, raising a total of $626 million, with a current valuation of $3 billion.
In May 2023, Poolside completed a seed round raising $26 million led by Redpoint Ventures; valuation undisclosed. Funds were used for AI platform development and initial infrastructure.
Three months later, in August 2023, Poolside raised $100 million in Series A led by Felicis, Redpoint China, and Xavier Niel; valuation undisclosed. Funds expanded the team and accelerated product development.
In October 2024, Poolside raised $500 million in Series B led by Bain Capital Ventures, valuing the company at $3 billion. Funds will focus on R&D for AI-driven software tools, market entry, and overall R&D acceleration.
What makes Poolside special enough for NVIDIA's interest?
After the Series B close, the founders posted on the company blog outlining fund uses, stating AGI is already real in ideal data and compute environments. AI becomes very smart through massive practice, but current learning efficiency is low, requiring too much data.
To make it as good at programming as humans, learning existing code isn't enough; it must "write and run code" itself, getting feedback on execution. This debugging-like process is key to improvement.
Thus, the Poolside team created a reinforcement learning method (RLCEF), letting AI attempt code writing on thousands of real projects and get immediate execution feedback for training. The company is pushing this training, with Series B funds mainly for massive compute resources.
The founders wrote in the blog:
"We believe self-supervised learning based on next-token prediction opens a promising path to AGI. However, current AI learns from data far less efficiently than humans, requiring orders of magnitude more data. While language understanding benefits from massive data, most other capabilities face data shortages, needing new data collection techniques.
We believe AGI is already real in ideal data and compute environments. Due to resource limits, software development will be the first area where AI achieves major economic impact, surpassing human capabilities.
To excel in software development, models need deterministic feedback to enhance reasoning and coding. This is evident in development: developers think, write, execute, iterate based on feedback. LLMs trained on language and code need massive similar feedback to match.
For this, we developed reinforcement learning based on code execution feedback (RLCEF). Our models explore solutions for millions of tasks across 130,000 real codebases, providing execution feedback for each. Though massive, it's far from human-level. With new funding, we've expanded the training cluster to 10,000 GPUs and are scaling RLCEF and model training."
NVIDIA was already an investor in prior rounds.
What makes this under-100-person AI startup appealing to giants like NVIDIA and Bain?
It starts with their tech stack.
Building infrastructure from scratch to break code data quality bottleneck
At an early 2024 AWS customer conference, Poolside's EMEA solutions engineering lead Vitor Monteiro was interviewed by the AWS Developer Blog, sharing the company's journey and tech roadmap. Vitor said Poolside's ultimate goal is true AI understanding and autonomous software development, advancing AGI in software.
Monteiro admitted AI coding tools are plentiful but hitting a "bottleneck"—features amazing two years ago no longer impress. Developers want context-understanding, reasoning AI assistants. That's Poolside's foundation.
"We're not building a chatbot or autocompletion engine," Monteiro said, "but an AI partner that collaborates like a colleague, learns, and suggests improvements."
On uniqueness, Monteiro noted only 10-15 companies globally have compute and tech to train frontier models—like OpenAI, Google, AWS—and Poolside is one.
Unlike most "AI consumer Copilots" calling external APIs and focusing on UI, Poolside builds everything in-house: data, algorithms, training frameworks. The team rejects open-source models, building full data pipelines, training software, and compute stacks from scratch for control.
Monteiro emphasized this costly path yields two advantages: avoiding open-source limits and deeply optimizing for software dev, not general compromises.
In training, the overlooked issue isn't compute, but data quality.
From GitHub/GitLab, only ~10% of open-source code is trainable; 90% is duplicate, buggy, redundant.
Thus, Poolside's CTO, researching code intelligence since 2016, developed a reinforcement system with code execution feedback, using synthetic data generation to overcome quality shortages.
The system scans high-quality internet codebases, validates changes, extracts commits. It uses commit descriptions as prompts to generate up to 50 code variants, filters via unit tests/security, keeping only passing samples for training. Positive/negative reinforcement optimizes decisions to "learn correct code."
Currently, Poolside has processed over 200,000 code repos automatically, scalable. Monteiro: global high-quality code ~3 trillion tokens; top models need 1.5 trillion synthetic. Poolside is approaching with proprietary tech.
Model iteration: Experiments select "optimal brain"
In R&D, Poolside uses a highly experimental parallel training system. Small-scale experiments explore architectures/losses/strategies, scaling winners.
Monteiro: reduces failure costs, keeps agility. First core model in 2025, pretrained on RL and synthetic data, bridging gaps in reasoning/task complexity.
Pre-release, Poolside offers base models and Maidu via AWS Bedrock.
In deployment, unlike SaaS, customers deploy directly in their cloud (e.g., AWS VPC), data stays internal. Popular with finance/gov/enterprises for privacy/local optimization.
Features: chat, autocompletion, code explanation, auto-docs; supports VS/VS Code; open API. For legacy systems, fine-tunes on private repos with RAG-like mechanisms for precise context suggestions.
For secure sectors, multi-model partitioning and RBAC prevent leaks.
On measuring productivity, Monteiro: "Not a single number." Tracks usage, calls, feedback, adoption.
Optimizing complex tasks like migration/refactoring—where one subtask failure cascades—industry challenge.
Finally, Monteiro: Poolside aims to shift dev paradigms. "Future devs build systems with AI, not just write code."
From models to deployment, synthetic data to RL, Poolside carves a unique path in crowded AI coding with pure self-built tech. On AGI path, enabling true "software understanding," beyond "code generation."
References:
https://www.youtube.com/watch?v=wMQoDRTWwUg&t=9s
https://www.focal.vc/transcripts/5yf-episode-20-poolside-ceo-jason-warner
https://www.youtube.com/watch?v=0TCMv2oiw-E
https://poolside.ai/blog/announcing-our-500-million-fundraise-to-make-progress-towards-agi