Fluidstack helped poolside deploy 2,500+ GPUs within 48 hours

In 2023, Jason Warner and Eiso Kant founded poolside to build AI for software development, reshaping developers’ experience with both Assistant and frontier models built for one another. With over $600M in funding, poolside deploys AI systems within their customer’s infrastructure. No data or code leaves their security boundaries.

In Jason Warner’s own words, poolside is building a future-proof AI workflow for software engineers:

“Our vision for the future is one where developers can build software with AI as an intuitive, secure, and flexible collaborator.” - Jason Warner, CEO of poolside

poolside combines foundation models, an API, and an assistant designed for highly complex environment such as Financial Services, Defense, Technology as well as Systems Integrators, alongside any other verticals with strict security and compliance requirements.

Given their enterprise-focus, poolside needed a partner that could reliably deliver a secure architecture without compromising on availability and flexible and economical prices. These requirements and constraints made Fluidstack the ideal partner.

The Challenge: Slow, Expensive, and Inflexible Options

poolside had two competing goals.

The first was a hardware and performance requirement; poolside needed high-performance and reliable GPU compute. The second was agility; because poolside is a fast-moving company, it required flexible and immediate access to compute resources.

Unfortunately, the mainstream market was riddled with compromises. The lead times from legacy players would extend into 3+ months, with 3-year commitments at high rates. Additionally, existing cloud platform performance was not optimised for AI workloads. Worse, they featured old-fashioned support channels that weren’t able to support customers who needed to run on failure-prone GPUs.

Instead of using a legacy player, poolside decided to partner with Fluidstack to efficiently deploy a powerful and price-effective stack.

The Solution: Fluidstack

Just days after signing with Fluidstack, poolside had access to an enterprise-grade GPU compute stack with over 2,500+ GPUs. This cluster of A100s and H100s shipped with some noteworthy benefits: fantastic performance, agile deployment, and attractive pricing.

Performance: Optimised Workloads and 24/7 Support Coverage

Because of Fluidstack’s global engineering team, poolside enjoyed 24/7 support via Slack. Practically speaking, Fluidstack became an extension to poolside’s engineering team, providing support pre- and post-deployment.

“Fluidstack’s dedicated support is excellent. Day and night.” - Eiso Kant, CTO of poolside

Fluidstack’s engineers shared insights into the GPU cloud market, including the NVIDIA product roadmap. They studied poolside’s workloads and recommended appropriate cluster specifications. Once the cluster was live, Fluidstack’s software platform supported, maintained, and tuned poolside’s workloads for peak performance. This included managing common errors such as GPU ECC errors, and helped to automatically fix these issues for poolside.

Fluidstack empowered poolside to optimize their compute strategy, reducing costs and time-to-market, alongside ensuring high performance during operations.

Speed: 2,560 GPUs deployed within two days

Traditional hyperscalers require a 3-month wait time; Fluidstack, meanwhile, offers a lead time measured in days for multi-thousand GPU clusters. For poolside, 2,560 GPUs were available just 48 hours after contract-signing.

“Operationally, the pace at which the team moves is unprecedented – this led to the fastest lead times and support from any of the partners we worked with.” - Margarida Garcia, VP Operations of poolside

In short, by ensuring immediate access to high-performance and secure infrastructure, Fluidstack empowered poolside to iterate on model development without delay.

Pricing: All-Inclusive & Cost-Effective

Fluidstack’s pricing model was a game-changer for poolside. Fluidstack’s pricing bundled:

• 24/7 engineering support with 15-minute response times on Slack
• 99% uptime SLA at both node and InfiniBand networking levels
• Fully-managed Slurm or Kubernetes, including setup, maintenance, and frequent upgrades
• Proactive monitoring of all cluster components.

Fluidstack’s pricing was both market-competitive and transparent. For a predictable budget, poolside could run large AI workloads.

Impact: Faster Market Entry, Continuous Model Optimization, and Cost Management

Poolside is able to attribute certain highlights of its early success to the Fluidstack partnership.

“From a business perspective, Fluidstack allowed us to start training our models earlier than expected with a level of flexibility others couldn’t offer.” - Margarida Garcia, VP Operations of poolside

• Accelerated Market Entry: With Fluidstack’s prompt lead time, poolside was able to deploy its MVP ahead of schedule
• Enhanced Model Optimization: Poolside could scale its GPU cluster with ease, enabling fast iterations to improve the model’s accuracy
• Cost-Savings: Fluidstack’s transparent pricing provided poolside with flexibility and immediate savings, allowing it to scale-up without financial headaches.

Because of Fluidstack, poolside was able to place its focus on its product at a critical time in the journey.

Closing Thoughts

Poolside’s experience with Fluidstack underscores that a powerful, secure GPU cluster shouldn’t come at the price of slow lead times, expensive costs, and poor support.

“One of the most important aspects of running an AI company is access to compute. We all know that large-scale clusters are difficult to operate, but Fluidstack has been exceptional.” - Jason Warner, CEO of poolside

In addition to Poolside, Fluidstack works with top AI labs, building 10k+ clusters for the world’s leading AI companies. If you want to discuss your GPU needs, click here to speak with our engineers.