Cloud-Prem: $100B AI Opportunity Hiding in Plain Sight

Cloud-Prem: $100B AI Opportunity Hiding in Plain Sight

In 2006, Salesforce defined what software delivery should look like.

A single-tenant CRM evolved into a fully multi-tenant SaaS platform. Low upfront costs. Centralized updates. Predictable revenue.

It was the beginning of a wave.

Over the next 15 years, SaaS became the dominant distribution model for enterprise software. Infrastructure scaled horizontally. Customers moved to the cloud. Everything seemed to converge on one truth: the future of software is delivered, not installed.

Until AI broke that model.


The Modern AI Stack Isn’t Your Good Old SaaS 

The SaaS stack of the last 20 years was built for lightweight apps: CRM systems, email automation, dashboards. It worked because:

  • Data volumes were small and structured.
  • Performance didn’t depend on milliseconds.
  • Security was met with compliance checklists.
  • Everything could live in the vendor’s cloud.

But AI changed everything.

Today’s AI stack is different in kind, not degree:

Article content
Traditional SaaS vs. Modern AI Stack

Modern AI applications don’t just store data—they perceive, reason, and act.

To do that securely, at scale, and in compliance, they can’t run in someone else’s cloud.

And the cloud itself has changed.

It’s no longer cheap, commoditized compute—it’s a scarce, high-performance utility. GPU clouds like CoreWeave, Lambda, and Together AI offer access to the world’s most valuable silicon. Model hosting and AI PaaS players are building atop them.

But enterprises aren’t willing to ship petabytes of proprietary data into someone else’s runtime.

They’re not going back to classic on-prem. And they’ll continue adopting newer GPU-rich clouds. What’s needed now is a delivery model that brings AI to the data—not the other way around.

That model is Cloud-Prem.

Cloud-Prem: The Inevitable Architecture for Enterprise AI

Cloud-Prem isn’t your old on-prem. And it’s not traditional SaaS either.

It’s a new category: Vendor-managed software, deployed inside the customer’s cloud—their VPC, their infra, their rules.

An architectural shift that blends:

  • The agility of cloud-native SaaS
  • The control of on-prem deployment
  • The trust boundaries modern AI demands

This isn’t a future trend. It’s already happening.

Why Cloud-Prem Is a $100B Opportunity Hiding in Plain Sight

Jensen Huang often describes “zero-billion-dollar markets” as spaces where demand hasn’t fully surfaced—yet the signals are already there. It’s a lesson I learned firsthand while working closely with him to build NVIDIA’s CUDA business.

Cloud-Prem is one of those markets.

Below are the leading indicators—framed as real enterprise problems—that are driving growing demand for Cloud-Prem architecture.

1. AI Model Customization Every enterprise wants its own LLM—fine-tuned on private, domain-specific data. But they can’t move that data to a vendor’s cloud. Cloud-Prem brings the model to the data.

2. Inference Close to the Data Inference is latency-sensitive—and data egress is expensive. Running AI next to the data (not 100ms away across clouds) improves both performance and economics. Cloud-Prem minimizes latency and avoids egress drag.

3. Sensitive Workflows Agentic AI that takes action—modifying infrastructure, accessing HR systems—can’t safely operate from a multi-tenant SaaS. Cloud-Prem keeps control in the customer’s hands, without losing SaaS agility.

4. Regulatory Compliance & Data Sovereignty Laws like GDPR, DORA, HIPAA, and India’s DPDP Act demand control over data location, access, and handling. Cloud-Prem bakes sovereignty into the architecture.

The Cloud-Prem Stack is Emerging

Cloud-Prem isn’t just a deployment tweak—it demands a rethinking of the entire software delivery stack. A new ecosystem is forming, built to serve enterprise AI from inside the customer’s infrastructure.

1. Full App Delivery

Tensor9 , Nuon , Ryvn (YC F24)

End-to-end platforms that handle packaging, deployment, upgrades, and operations of AI and SaaS apps inside customer environments.

Core Capabilities:

  • App packaging, orchestration, and lifecycle management
  • Control plane / data plane separation
  • Built-in observability, upgrades, and version control
  • Developer-first UX with SaaS-like simplicity

Why it matters: These platforms abstract the full delivery pipeline—helping vendors go Cloud-Prem from day one without reinventing their deployment architecture.

2. Kubernetes Distribution

Replicated , Northflank , Glasskube

Tools that simplify distributing and managing Kubernetes-native applications inside customer VPCs or on-prem clusters.

Core Capabilities:

  • Helm/Kubernetes packaging and multi-cluster install support
  • Secure delivery and versioned updates
  • Customer licensing and entitlements
  • Integration with enterprise IAM and audit systems

Why it matters: Kubernetes is the foundation of Cloud-Prem. These tools make it feasible to ship and manage K8s apps at scale—without giving up security or UX.

3. Infrastructure & IaC

HashiCorp Terraform, Pulumi , Amazon Web Services (AWS) , Ansible by Red Hat , SaltStack Enterprise , Crossplane project

Tools to define, provision, and manage infrastructure via code—across multi-cloud and hybrid environments.

Core Capabilities:

  • Declarative provisioning of VPCs, K8s clusters, services
  • GitOps and CI/CD-native workflows
  • Policy and compliance integration
  • Support for multi-cloud and hybrid deployment

Why it matters: Cloud-Prem architectures must be reproducible, secure, and auditable. IaC is the foundation for scalable, compliant deployments.

4. Data Platforms (Cloud-Prem Native)

ClickHouse , Redpanda Data , WarpStream , Snowflake

Analytics databases, vector stores, and streaming platforms that run inside the customer’s environment—often with vendor-managed control planes.

Core Capabilities:

  • BYOC or self-hosted architectures
  • No data egress, full data locality
  • Compliance and telemetry built in
  • Elastic performance in customer VPCs

Why it matters: AI systems depend on low-latency access to proprietary data. These platforms bring compute to the data, not the other way around.

5. Monitoring & Observability

Datadog , Prometheus Group , Grafana Labs , Chronosphere , InfluxData

Systems to observe performance, behavior, and health of applications running inside customer environments.

Core Capabilities:

  • Metrics, logs, traces inside VPC
  • Prometheus-compatible agents
  • Dashboards, alerting, telemetry pipelines
  • BYOC or self-hosted modes

Why it matters: Cloud-Prem means vendors lose access to observability by default. These tools return visibility—without exporting sensitive data.

6. Security & Policy

Open Policy Agent by Styra , HashiCorp Vault, Aserto , StrongDM , Doppler

Access control, identity, policy enforcement, and secrets management—designed for zero-trust and hybrid environments.

Core Capabilities:

  • Identity-aware, least-privilege access
  • Secret rotation and key management
  • Policy-as-code (OPA/Rego)
  • Support for isolated, air-gapped and zero-trust setups

Why it matters: Trust is the linchpin of Cloud-Prem. These tools enforce the security posture that regulated industries demand by default.

7. Remote Access

Tailscale , Teleport , HashiCorp Boundary

Lightweight, identity-aware access tools that allow secure, audited remote access to infrastructure inside the customer’s network—without VPN overhead.

Core Capabilities:

  • Zero-trust networking built on WireGuard or SSH tunneling
  • Device- and identity-based access control
  • Session logging, policy enforcement, and MFA
  • NAT traversal and secure peer-to-peer connections

Why it matters: Cloud-Prem apps live in locked-down environments. These tools enable secure, policy-driven access—without the complexity or risk of traditional VPNs.

Article content
Cloud-Prem Landscape | Devang Sachdev | April 2025

What This Means for AI Founders

Cloud-Prem doesn’t just require new tooling—it enables a new delivery model. The next generation of AI-native SaaS companies won’t just be multi-tenant—they’ll be multi-environment.

Cloud-prem is not just a technical choice or late-stage compliance feature—it’s a go-to-market accelerant.

Choosing cloud-prem early:

  • Speeds up sales cycles in high-trust markets (healthcare, financial services, defense)
  • Avoids the compliance walls that slow traditional SaaS
  • Positions your product as future-proof in a post-SaaS world
  • Wins more deals, while still operating like SaaS.

If you are building a company at the intersection of AI and the enterprise, the delivery model is no longer a backend decision. It is part of your go-to-market, your product strategy, and your defensibility.

The Bottom Line

SaaS changed the way we delivered software. Cloud-prem will change the way we deliver AI.

It won’t replace SaaS. But for AI-native software touching sensitive data, real-time decisions, or privileged systems, Cloud-Prem is quietly eating SaaS—one AI workload at a time.

It’s a shift in architecture. A shift in trust. A shift in market access.

And for those who recognize it early, a shift in opportunity.

Tiffany Joy Basse

Fractional Chief Growth Officer, Investor, Entrepreneur, Author, Speaker

3w

Brilliant breakdown—this shift has been hiding in plain sight for too long. Cloud-Prem isn’t just a technical pivot; it’s a fundamental go-to-market unlock for AI founders trying to serve high-trust, regulated, or latency-sensitive markets. As someone who’s helped GTM strategies across SaaS and AI, I’ve seen firsthand how legacy deployment models stall deals in healthcare, finance, and public sector. Cloud-Prem finally gives enterprise buyers the control they demand without sacrificing the agility of modern software. Thanks for surfacing this trend—this is where the real AI infrastructure opportunity lies.

Beezar Sirini

Co-Founder Boost Express | Turning Event Data into Scalable Revenue | Data Monetization Expert

1mo

Niraj Shah you were talking about this a years ago already.

Simon West

Helping ISVs build amazing SaaS solutions at Amazon Web Services (AWS). Ex-Twilio

1mo

I’ve been calling this Hybrid-SaaS. The control plane and some functionality runs in the ISVs VPC, with things like inference running in the end customer VPC. It does address the regulation and latency concerns of the end customer; it also removes some of the costs for the ISV - as their customer is becomes responsible for the infrastructure costs for inference, for example. This does make pricing and packaging more complex for ISVs.

Naoures Ben Brahim

Customer Value & Strategy Director @ Temenos | Business Strategy

1mo

Insightful! Thanks

Arjun R Narayanaswamy

Entrepreneur | Explorer | MIT Alum

1mo

This is a nice article. It lays out all the ways that today's world is different from traditional SAAS, because of essentially one factor: Data Gravity. It's hard to move the data so the compute must come to the data. I've been wondering what the right word for this anti-SAAS movement is . I've heard "BYOC" a lot and now "cloud-prem" One question: Don't you think the final name for this movement should also include "edge" as a deployment option? Right now the edge does very little complex compute. However the same arguments for Data Gravity will mean that the edge is as viable a deployment location as cloud or on-prem. Cloud-prem-edge is a mouthful :)

To view or add a comment, sign in

More articles by Devang Sachdev

Insights from the community

Others also viewed

Explore topics