Moises is the creative suite for musicians trusted by Grammy-winning artists, prestigious institutions, and more than 73 million artists and producers across the world. Moises runs on several proprietary models that the company trains in-house using licensed materials. By working with Shadeform, Moises is able to consistently secure compute for training with unified access to 30+ providers, reduce their infrastructure complexity with a single consolidated platform, and accelerate training timelines with access to better hardware at lower prices.

“Training proprietary models is a big part of our competitive edge. The GPU availability we get with Shadeform allows us to keep our training pipelines running, even during a global compute shortage. They’ve been critical to our business.”

Fernando Grunevald, Software Architect, Moises

Challenge:

Scaling training workloads without increasing engineering complexity

Moises’ internal research team has been a core differentiator for the company’s AI-powered creative suite since the early days. What started as a single PhD researcher building on open-source foundations has grown into a data science team that ships best-in-class proprietary models across stem separation, chord detection, lyrics transcription, and stem generation.

Keeping that team productive required consistent and affordable access to GPU compute, but that became harder as training demand grew. Moises ran most of their workloads with traditional hyperscale clouds, but as the research team scaled up training runs, the economics stopped working. On-demand pricing was expensive, commitment terms were inefficient, spot instances weren’t a viable option, and egress charges added additional overhead to plan around.

“The only way we could bring costs down was to make a very long commitment, and that didn’t feel like the right move.”

Looking outside of hyperscale options came with its own friction. Neoclouds couldn’t support the capacity requirements of the team individually, and working with multiple vendors to satisfy those requirements wasn’t sustainable. The team would have to manage separate accounts, separate threads, and unique APIs that would have meant rebuilding their training workflow to fit each provider.

“Every provider has its own platform and its own way of doing things. Our data scientists would have had to rebuild their workflows for each, which we couldn’t justify.”

Moises needed a way to give the research team consistent access to affordable compute, without introducing infrastructure and operational complexity that would slow their work.

Solution:

Unified GPU access with no operational overhead

Moises turned to Shadeform to deliver high GPU availability at lower cost, without added complexity. Shadeform’s unified API and platform let the team access on-demand, affordable compute across 30+ vetted providers, without needing to change their existing Kubernetes setup and training scripts.

“Having the ability to keep our setup was the major thing. We didn’t have to conform to one vendor or another. That made Shadeform a no-brainer for us.”

Better pricing also unlocked hardware that wasn’t within budget before. With their cost savings from Shadeform, Moises moved from training on H100 GPUs to H200s, simplifying their training architecture in the process. Additionally, because none of the providers in Shadeform’s network charge for egress, the team stopped having to budget for or worry about data transfer costs.

Once the team had enough training demand to justify long-term commitments, Shadeform handled the operational work of communicating with cloud partners, comparing terms, negotiating pricing, and providing expert technical support. All vendor overhead was consolidated into a single point of contact.

“We’ve talked to other providers directly, and Shadeform is still the best partner for us. It really feels like they work with us, and their operational execution has been excellent.”

All of this matters more as the GPU market faces growing supply shortages in 2026. With compute capacity becoming harder to secure across the industry, Shadeform’s unified platform has kept Moises’ research team unblocked. While other teams scramble for GPU capacity, Moises continues to ship better models and outpace competitors in their space.

“Without the GPU availability we get from Shadeform, we would be severely impacted by the GPU shortage. Having that baseline of capacity has been crucial for the company.”

Results:

Reduced costs, shorter training cycles, and reliable access to compute

By moving their training workloads to Shadeform, Moises’ research team gained the compute access they needed to scale efficiently and affordably. Today, the partnership protects their training pipelines through a tightening GPU market and keeps the team focused on shipping better models.

• Switched from H100s to H200s at 4.9x reduced cost
• Unlocked larger training batches and shortened timelines
• Zero capacity requests unfulfilled

Looking forward

Moises continues to invest in their consumer products as well as technology integrations with major brands. Both directions depend on the same thing: a research team that can keep training, keep iterating, and keep pushing model quality faster than competitors can.

With Shadeform managing the GPU layer, the team stays focused on the work that matters most.

To learn more about how leading research teams stay ahead with Shadeform, book a call with our team.