AWS introduces nova forge for tailored AI models
There’s a shift happening in enterprise AI, and it’s overdue. Most large language models (LLMs) in the wild are trained on public datasets that lack enterprise context. That’s a problem. These models are smart, but they don’t know your business. They don’t understand your processes, your metrics, or your customers. And the typical “solutions”, prompt engineering, fine-tuning, and retrieval-augmented generation (RAG)—are good for quick patches, but not for deep, sustained performance. They remain surface-level hacks, stuck running on top of someone else’s foundation.
AWS is changing that dynamic with Nova Forge. It gives companies the tools to build deeply customized models by embedding proprietary data directly into model training, not as an afterthought, but as part of the core process. That distinction matters. When your AI learns business logic during training, you don’t need to continually feed it external context at runtime. The knowledge is already part of the system.
Instead of training a new model from scratch, Nova Forge leverages what AWS calls “training checkpoints”—pre-trained model snapshots at various stages. You can inject your domain-specific data during early-phase training, mid-stage, or even just before deployment. You maintain control over how deeply your data shapes your model, depending on what you need. This process not only saves enormous amounts of time and money but also results in a model that’s fundamentally more aligned with business operations.
From a leadership standpoint, this is critical. Building your own LLM used to be a billion-dollar affair. That cost barrier kept the majority of companies locked out of custom AI. With Nova Forge, that gate is now open. You’re getting strategic autonomy at a fraction of the cost and without waiting years.
Stephanie Walter, AI practice leader at HyperFRAME Research, highlighted why current fixes fall short: “Prompt engineering, RAG, and even standard supervised fine-tuning are powerful, but they sit on top of a fully trained model and are inherently constrained.” She’s right. These methods introduce complexity with too many moving parts, context window limits, orchestration issues, and high risk of error. This is why embedding context directly into the foundation is a smarter bet.
David Menninger, executive director at ISG, echoed this idea, stating that Nova Forge’s approach makes inference easier to manage and maintain long term. Models aren’t constantly going back to source databases for clarification, they already know enough to make good decisions, faster.
We’re going to see a lot of movement here. Enterprise AI is no longer about tweaking someone else’s model. It’s about owning intelligence that’s specific to your business. Nova Forge makes that real.
Infrastructure-centric AI strategy differentiates AWS from microsoft
There’s a clear divergence in AI strategy between the major players. AWS and Microsoft aren’t solving for the same layer of the AI stack, and that’s intentional. Microsoft is focused on owning the user-facing experience. Everything it builds, from Fabric IQ to Work IQ, is designed to deliver compact, pre-integrated AI models that work within its productivity ecosystem. It wants businesses to depend on its tools to get faster results, without thinking much about what’s underneath.
AWS is taking a different approach. Its value proposition isn’t in owning the end-use experience. It’s about giving enterprises control over the foundation. Nova Forge is designed for companies that want to build intelligence that’s theirs, custom logic, proprietary data, private infrastructure. AWS enables that by focusing on scalable infrastructure and developer autonomy. That’s a big differentiator for enterprises looking for long-term flexibility.
Akshat Tyagi, Associate Practice Leader at HFS Research, breaks the contrast down clearly: “Microsoft wants to own the AI experience. AWS wants to own the AI factory.” Microsoft is streamlining access to intelligence through its environment. AWS is equipping businesses to develop intelligence on their terms and run it privately, securely, and in ways that scale beyond the vendor-provided interface.
For executives, this distinction matters. Where Microsoft optimizes for speed-to-productivity within its ecosystem, AWS is investing in enterprise sovereignty and customization. Companies with strict compliance needs, proprietary operating models, or unique service pipelines aren’t going to be fully supported by out-of-box productivity AI. They need tools like Nova Forge to create capabilities that elevate their own systems, not replicate someone else’s workflow.
AWS also understands its strength lies in infrastructure. It’s not fighting for the assistant or productivity layer, where Microsoft is already entrenched. Instead, it’s doubling down on what it does best: scalable compute, flexible platforms, and enterprise-grade development environments. That’s how it’s positioning Nova Forge, and that’s why it’s resonating with C-suites that think about AI as core infrastructure, not just a feature.
The industry is choosing sides. One is offering packaged intelligence. The other is offering the means to build your own. Executives choosing platforms need to be clear about what they’re optimizing for, control, capability, or convenience. The answer determines which partner is aligned with your ambitions.
Nova forge reduces costs and complexity in custom LLM training
Training a large language model from scratch has historically been out of reach for most companies. The cost alone, time, compute, expertise, can reach into the hundreds of millions, even billions. Add to that the engineering overhead and long development cycles, and it becomes obvious why only a few hyperscalers have attempted it. AWS knows this, and Nova Forge is designed to remove that barrier.
Instead of requiring enterprises to build everything from the ground up, Nova Forge offers a smarter path: pre-trained model checkpoints. These are snapshots captured at various stages of the model’s training, early, mid, or post-development. Enterprises can jump in at the point that fits their needs and inject proprietary data right there. It’s a targeted way of shaping the model with domain-specific knowledge, without incurring full-stack R&D costs.
The flexibility this offers is key. Some businesses may only need minimal tuning. Others might want deep integration of domain vocabulary, workflows, or regulatory logic. Nova Forge supports both. Companies decide how much depth they need, and AWS provides the infrastructure to make it operational and repeatable.
Robert Kramer, Principal Analyst at Moor Strategy and Insights, put it clearly: “Enterprises choose how deeply they want their domain to shape the model.” That’s control. It shifts the conversation from what’s possible to what’s optimal.
From a business standpoint, predictability is just as valuable as capability. AWS will deliver Nova Forge as a subscription, avoiding the open-ended cost escalations that come with pure compute consumption models. That means companies know what they’re signing up for, both in budget terms and technical scope. CNBC reports pricing starts at $100,000 per year, still a significant investment, but compared to full training stacks, the cost is a fraction.
This is a straight path toward enterprise-grade AI that’s custom and portable. Models can be trained via SageMaker Studio and deployed through Bedrock, AWS’s hosted environment for scalable inference. That flow minimizes tool switching and ensures models move securely across development and production pipelines.
For C-suite executives under pressure to innovate fast, this setup matters. You’re not locked into a multiyear AI build just to get started. You’re not exposed to unpredictable cost spikes. And you’re not forced to compromise on domain accuracy. Nova Forge lets you start strong, scale quickly, and stay in control. That’s a better way forward.
Rising demand for domain-specific AI drives innovation beyond traditional fine-tuning
Across industries, enterprises are pushing AI into highly specialized areas, healthcare, financial services, manufacturing, industrial controls, and code generation. These domains demand high precision and strict compliance. General-purpose models can’t deliver that. They weren’t built with industry-specific data or rules in mind. Traditional customization methods like fine-tuning, prompt engineering, and retrieval-augmented generation (RAG) help, but they don’t go far enough.
These surface-level enhancements sit on top of the model. They react to queries but don’t fundamentally change how the model understands your business. That’s a problem when you’re dealing with sensitive processes that require detailed operational context or must adhere to legal and regulatory standards. High latency, context-window limitations, and orchestration complexity all create room for error, and in these sectors, error isn’t acceptable.
Nova Forge addresses this with a more integrated solution. By injecting domain knowledge during key stages of model training, enterprises can shape how the model thinks, not just how it responds. This means the model learns business rules and operational logic before it ever starts generating outputs. It becomes a first-party system that implicitly understands the environment in which it operates.
Stephanie Walter, Practice Leader at HyperFRAME Research, emphasized this shift, noting that embedding expertise into the foundation allows AWS services like Trainium (its AI-optimized chip), SageMaker (its ML development platform), and Bedrock (its deployment environment) to work together more effectively. These AWS services form a unified pipeline for building and deploying enterprise-grade language models with less cost and more precision.
This approach isn’t just about strong tech. It’s strategic. Companies in regulated or technical fields can now develop models without depending on external APIs that may expose data or limit control. That increases regulatory alignment, improves data security, and gives executives long-term flexibility.
For leadership teams planning multi-year AI adoption across diverse use cases, the implication is significant. You can now build models that don’t just process data, they understand it in your terms. And you don’t need to compromise between speed, cost, and compliance. Nova Forge offers a way to meet all three. That’s the kind of enterprise AI play that scales with real-world requirements.
Key executive takeaways
- Nova forge enables smarter AI integration: AWS’s new service lets enterprises embed proprietary data directly into model training, allowing businesses to internalize their logic within the AI. Leaders should consider this approach to reduce reliance on external prompts and improve accuracy at scale.
- AWS doubles down on infrastructure ownership: Unlike Microsoft’s ecosystem-driven AI strategy, AWS is handing businesses the tools to build custom intelligence privately. Executives prioritizing control, customization, and data privacy should align with AWS’s infrastructure-centric model.
- Custom AI at lower cost and complexity: Nova Forge offers pre-trained model checkpoints to reduce time, cost, and engineering demand, starting at $100K per year. Leaders aiming for bespoke AI without full-stack overhead now have a commercially viable path.
- Precision AI for regulated and technical use cases: For industries demanding security, compliance, or technical accuracy, Nova Forge enables domain-specific training beyond surface-level fine-tuning. Decision-makers should evaluate this for high-stakes environments like healthcare, finance, and industrial systems.
