AI is moving from experimentation into production. The first phase of enterprise adoption was largely defined by access to frontier models via APIs – which has been effective for this period. Companies could move quickly, test use cases, and put powerful models in the hands of builders without having to worry about or manage the underlying infrastructure.
However, the next phase will look markedly different as enterprises begin to grapple with the costs inherent to consuming AI at scale. As workloads scale, inference becomes one of the largest and most important production layers in the enterprise. Cost, latency, performance, data control, and model choice all begin to matter more. Enterprises will still use the leading proprietary frontier models for use cases that require them (R&D, innovation, product development, etc.). However, we will also see them increasingly move toward a blended / multi-model architecture that combines frontier APIs, open source models, fine tuned models, and dedicated inference infrastructure.
This shift however, creates a new security dilemma. The inference layer is where prompts, customer data, model responses, tools, agents, retrieval systems, and enterprise workflows all come together. It is also where new risks concentrate, from data leakage to unsafe outputs, model abuse, and infrastructure exposure. The more model choice expands, and the fewer the guardrails, the more security becomes a critical part of inference architecture.
That is why S Ventures is excited to announce our investment in Together AI, forging a partnership that will help drive the future of secure enterprise inference.
Together AI + SentinelOne: Securing the Inference Cloud
SentinelOne is an AI Security leader delivering AI driven autonomous cybersecurity to reduce risk, eliminate noise, and give defenders the advantage. Together AI is building the AI Native Cloud, a full stack platform for production AI spanning inference, fine tuning, model shaping, compute, and AI infrastructure.
Together AI and SentinelOne will allow enterprises to aggressively adopt AI infrastructure, where speed, flexibility, and security are non-negotiable foundations.
The coming partnership will help us establish the leading Secure Enterprise Inference Cloud, redefining the trade-off between model flexibility and control. Together brings the infrastructure layer for production AI. SentinelOne brings deep enterprise security expertise across endpoint, cloud, data, AI applications, agents, and runtime protection.
As enterprises move toward multi-model deployment, the security dilemma intensifies across two key areas: visibility and model risk. Security teams require comprehensive visibility to understand which models, applications, and agents are being used, what data is flowing through prompts, and where sensitive data may be exposed. SentinelOne’s Prompt Security and Purple AI provide a critical foundation, securing AI usage by protecting against prompt injection, preventing data exposure, and enforcing policy. Additionally, the adoption of open source and specialized models (while delivering better economics and performance) introduces new model risks around provenance, usage, behavior, runtime controls, and exposure. Security must be applied consistently across this blended environment of models from various providers and open source ecosystems.
Finally, securing the underlying AI infrastructure is paramount and requires a level of autonomous defense. As inference workloads transition onto dedicated infrastructure, enterprises demand robust protection across cloud posture, workload behavior, access, data movement, and runtime activity. This demanding and rapidly evolving environment (where speed increases and complexity rises) is precisely the kind of transition SentinelOne was engineered for, delivering the intelligent security layer required to keep pace and protect the enterprise’s mission-critical AI control plane.
Inference Is the New Enterprise AI Control Plane
For the last several years, the AI market has been dominated by training scale, model capability, and access to the most powerful frontier systems. Those remain critical. But as enterprises move from experimentation to deployment, the center of gravity shifts toward inference.
Inference is where AI becomes operational. It is where a model reasons over a customer request, summarizes sensitive information, generates code, retrieves internal context, invokes tools, or supports an employee workflow. It is also where cost becomes persistent. A model that is called millions or billions of times becomes an infrastructure decision, not just a product decision.
In many ways, this will resemble the move to multi-cloud, but for models. Enterprises will not standardize on one model provider for every use case. They will want the ability to route workloads based on cost, latency, accuracy, security, and control. That is the market Together AI is building for.
But multi-model architecture also expands the attack surface. Every prompt becomes a potential input vector. Every retrieval workflow becomes a potential path for indirect prompt injection. Every agent action becomes a policy and identity question. Every model endpoint becomes a production system that needs monitoring, governance, and protection.
Security cannot be added after the fact. It needs to sit alongside the inference layer itself.
Why Together AI
Together AI has established itself as one of the leading infrastructure companies for the open source and enterprise AI ecosystem. The company provides a full stack AI platform that helps teams run inference at scale, fine tune open source models, scale compute, and move production AI workloads from experimentation to deployment.
Together sits in a strategically important position in the AI stack. It is not just helping developers access models. It is providing the core infrastructure layer for a more open, flexible, and production oriented AI future. That includes serverless inference, dedicated model inference, fine tuning, GPU clusters, managed storage, and the broader infrastructure needed to support AI applications at scale.
The company’s momentum reflects the market need. Together has already built a large developer and enterprise customer ecosystem, serving AI-native companies and global enterprises that need high performance infrastructure for modern AI applications. As more traditional enterprises move from AI experimentation into scaled deployment, we believe Together is poised to become the definitive platform for production inference.
We invest in companies that align closely with SentinelOne’s mission to define the future of autonomous, intelligent security. Together represents a critical infrastructure layer for that future, the place where enterprise AI moves from model access into production execution.
Securing Our AI Led Future
Enterprise AI will be multi-model, multi-cloud, and increasingly inference driven. That architecture will unlock enormous productivity, but it will also create new security requirements that legacy approaches were not built to address.
The winners in this next phase will be the platforms that help enterprises move faster without losing control. They will give customers flexibility without opacity, performance without unmanaged risk, and cost optimization without weakening security.
We are aligned and excited to invest in and partner with Vipul, Ce, and the entire Together AI team as they build the AI Native Cloud. And together, we will actively secure the next phase of enterprise AI.