Building the cloud behind modern AI
Modal was founded on the idea that developers building AI should spend their time creating applications, not managing infrastructure. As powerful open models have become more accessible, the challenge has shifted from building AI models to operating them reliably at scale. Running modern AI applications requires enormous amounts of compute, specialized hardware, and infrastructure capable of adapting instantly to changing demand. Modal was built to remove that complexity.
The company’s cloud platform powers everything from low-latency inference controlling robotic systems to reinforcement learning, AI agents, weather forecasting, and massively parallel batch processing. Developers bring their code while Modal manages everything underneath, giving customers access to the compute they need without worrying about provisioning GPUs, orchestrating workloads, or managing distributed infrastructure. “What hasn’t changed over the last five years is the problem we’re solving,” says Akshat Bubna, CTO & Co-founder of Modal. “We’re rethinking the compute layer for these data- and compute-heavy applications. The difference is the size and scale we’re operating at today.”
That scale is possible because Modal built its platform differently. Rather than relying on Kubernetes or traditional container orchestration, the company developed a custom runtime, filesystem, and scheduling system purpose-built for AI workloads. Those investments enable fast startup times, intelligent autoscaling, GPU snapshotting, and a developer experience that feels as simple as running code locally. But as the platform expanded across more clouds, regions, hardware types, and products, operating that infrastructure became an observability challenge of its own.
Visibility across one of the industry’s largest AI platforms
Today, Modal operates across more than 20 cloud providers, supporting customers with a single, consistent platform regardless of where compute is running. Behind every customer deployment, Modal continuously schedules workloads, manages GPU capacity, balances traffic, and absorbs the differences between cloud providers so developers never have to think about the infrastructure powering their applications.
That complexity grows every day. Modal launches approximately 26 million containers daily while operating across diverse GPU architectures and constantly shifting AI workloads. Every provider introduces different networking behavior, hardware characteristics, and failure modes, making comprehensive visibility essential. “We operate across many different planes of complexity,” Bubna says. “We need to understand the health of all of them at once and pinpoint issues in seconds. It’s not just us that’s impacted. It’s our customers, and their customers.”
For Modal’s engineering team, observability isn’t simply about monitoring infrastructure. Engineers need to understand the health of the entire platform in real time while quickly determining whether an issue originates in customer code, Modal’s scheduler, or an underlying cloud provider. That level of visibility allows the team to isolate issues quickly, reduce blast radius, and maintain the reliability customers expect from a managed AI platform. “When you’re operating across more than 20 cloud providers, visibility isn’t optional,” Bubna says. “We need to understand what’s happening across every layer of the platform in real time so we can solve problems before our customers ever see them.”
“We operate across many different planes of complexity. We need to understand the health of all of them at once and pinpoint issues in seconds. It's not just us that's impacted. It's our customers, and their customers.”
Scaling observability alongside the business
Modal adopted Datadog early because it was easy to deploy and capable of supporting the scale the company expected to reach. As Modal expanded, Datadog expanded alongside it. What began as infrastructure monitoring evolved into a unified observability platform supporting Infrastructure Monitoring, Application Performance Monitoring (APM), Network Monitoring, CI Visibility, logs, metrics, and traces.
Today, Modal engineers use Datadog as a single platform to investigate every layer of their environment. Instead of moving between disconnected tools, Modal teams correlate infrastructure, networking, Kubernetes environments, applications, and cloud services from one place to quickly understand what’s happening and where action is needed. “We wanted something that could handle our scale and cardinality without us building around it,” Bubna says. “We didn’t want to stitch together multiple tools or slow our roadmap because observability couldn’t keep up. Datadog has simply grown alongside us.”
Observability has also become part of Modal’s engineering culture. Every new service is instrumented as it’s built, giving engineers confidence to release quickly while maintaining complete visibility into production. Rather than slowing innovation, observability has become one of the reasons engineers can move faster.
Modal has also embraced AI-assisted operations through Datadog MCP Server. AI agents are given access to Datadog telemetry so they can investigate incidents, correlate logs, traces, infrastructure metrics, and networking signals, then surface the most likely root causes for engineers. Instead of replacing engineers, the agents eliminate much of the manual investigation that traditionally slows incident response. “We use Datadog MCP Server as part of our engineering workflow,” Bubna says. “Giving our AI agents direct access to production telemetry helps them investigate issues faster and gives our engineers the context they need to reach root cause much more quickly.”
Modal is also collaborating with Datadog as an integration partner, helping developers combine AI infrastructure with Datadog’s observability and AI capabilities so AI agents have the production context they need to investigate issues and accelerate operational workflows.
As the platform has grown more than tenfold and expanded into entirely new product lines, observability has never become a bottleneck. “Observability hasn’t stopped us from scaling,” Bubna says. “It’s simply come along for the ride instead of becoming a rebuild project.”
“We use Datadog MCP Server as part of our engineering workflow. Giving our AI agents direct access to production telemetry helps them investigate issues faster and gives our engineers the context they need to reach root cause much more quickly.”
Helping every AI company move faster
Modal operates one of the industry’s largest AI platforms with Datadog providing unified visibility across its environment. Every month, Modal ingests approximately 116 billion log events, representing nearly 246 TB of log data, alongside 97 billion APM spans collected across 66 production services and approximately 1,300 hosts, while tracking 2.7K metric names. With this unified visibility, Modal engineers can understand the health of the platform at a glance, regardless of where workloads are running.
Those insights translate directly into reliability. Modal detects 99% of incidents within five minutes, and approximately 96% of issues are resolved before customers ever experience an impact. Instead of reacting to outages, engineering teams identify issues early, understand their root cause quickly, and restore service before developers are affected.
Datadog also helps Modal balance comprehensive observability with operational efficiency. The platform ingests approximately 120 billion APM spans every month, while indexing only about 2.3 billion. That approach gives engineers access to the telemetry they need to investigate complex issues while optimizing observability costs as the business continues to grow.
For Modal’s customers, those capabilities translate into confidence. Teams can launch AI products, scale through unpredictable traffic spikes, and run compute-intensive workloads without provisioning infrastructure or managing GPU fleets. Modal absorbs the operational complexity so developers can focus on building applications instead of operating cloud infrastructure.
Looking ahead, Bubna believes every company is becoming an AI company, creating entirely new demands on cloud infrastructure. With Datadog providing the visibility to operate one of the industry’s most complex AI platforms, Modal can continue expanding its cloud while giving engineers the confidence to keep building what’s next. “We know observability will never slow us down,” Bubna says. “Because we trust the visibility Datadog gives us, we can keep taking bigger swings, expanding the platform, and building the next generation of AI applications.”
“We know observability will never slow us down. Because we trust the visibility Datadog gives us, we can keep taking bigger swings, expanding the platform, and building the next generation of AI applications.”
