Welcome Back to XcessAI
When ChatGPT launched, many assumed the AI race would be won by whoever built the smartest model.
That assumption shaped almost every conversation over the last two years. Companies raced to train larger models, investors rewarded frontier labs with enormous valuations, and enterprises searched for the model that could outperform every competitor.
But something has quietly begun to change.
Recent reports suggest that a growing share of enterprise AI workloads is already running on open-weight models, including several developed in China. At the same time, models like Qwen, Kimi and GLM are closing the performance gap while remaining dramatically cheaper to deploy.
Most headlines frame this as a geopolitical story, but it could be signalling something much bigger.
It suggests that intelligence itself is becoming increasingly accessible. And when that happens, competitive advantage has to move somewhere else.
The First Era of AI
The first generation of enterprise AI was remarkably simple.
Need intelligence? Rent it. A company subscribed to a frontier model through an API and paid for every token consumed.
The value proposition was obvious. The better the model, the higher the price. The smarter the intelligence, the stronger the competitive advantage.
It was an attractive business model because intelligence itself was scarce. Only a handful of companies could train frontier models, and everyone else became customers.
That scarcity justified enormous valuations and equally enormous investments in compute infrastructure.
Intelligence Is Becoming More Accessible
Open-weight models are changing that equation.
Instead of renting intelligence from a single provider, organizations can increasingly download it, customize it, and run it inside their own infrastructure.
That changes far more than licensing costs. It changes who owns the technology.
A company can fine-tune a model using its own proprietary data. It can deploy it inside private environments. It can optimize it for specific workflows. It can reduce inference costs dramatically. And perhaps most importantly, it becomes less dependent on any single AI provider.
The conversation starts moving away from "Which model should we buy?" toward "Which architecture should we build?". That is a much more strategic question.
The AI Moat Is Moving
This is something we've been gradually exploring throughout XcessAI.
In The Commoditization of Intelligence, we argued that model selection was becoming less strategic than system design.
In Operating System, we explored how AI is evolving into the interface through which work gets done.
In The Missing Layer, we looked at how context is becoming the key ingredient that turns generic intelligence into useful intelligence.
Open-weight models reinforce all of those ideas. If multiple organizations have access to similar levels of intelligence, then intelligence itself becomes less of a differentiator.
The advantage shifts toward everything built around it.
Context
Workflow design
Proprietary data
Governance
Integration
Execution
Those are much harder to copy than another language model.
From Model Companies to Platform Companies
This shift resembles what happened during previous technology cycles.
Owning the operating system became more valuable than owning individual applications. Owning the cloud platform became more valuable than owning individual servers.
AI may be entering a similar phase. The model increasingly becomes infrastructure. The platform built around it becomes the product.
That platform includes how information flows through an organization, how AI interacts with enterprise systems, how context is preserved, how costs are managed, and how outputs are verified before they influence real decisions.
The intelligence is only one component. The surrounding architecture creates the value.
Why Enterprises May Benefit
For many organizations, this is encouraging. Competition among models is pushing costs lower while improving performance.
Open models provide greater flexibility. Closed models continue to push the frontier of capability.
Enterprises increasingly have a choice rather than a dependency. Instead of committing entirely to one provider, many organizations are beginning to adopt hybrid architectures.
Some workloads continue using frontier models. Others migrate to open-weight models running privately. Smaller models handle routine tasks. Larger models are reserved for complex reasoning.
The result is a more efficient AI stack. Not because any individual model is dramatically better, but because each one is used where it creates the most value.
The Strategic Question
None of this means frontier AI companies become irrelevant.
Building the most capable models will remain extraordinarily difficult. Frontier research will continue to push the boundaries of reasoning, science, coding and multimodal intelligence.
But the competitive landscape is broadening. Success may no longer depend solely on creating the smartest model. It may increasingly depend on creating the best ecosystem around it.
That ecosystem includes infrastructure, developer tools, enterprise integration, orchestration, security, governance and distribution.
In other words, the moat is moving.
Final Thoughts
AI has spent the last two years competing on intelligence. The next phase may compete on architecture.
As intelligence becomes more accessible, value migrates toward the systems that organize, deploy and govern it.
The winners may not simply build the smartest models. They may build the environments where those models become genuinely useful.
The race is no longer just about making AI more intelligent. It is increasingly about making intelligence work.
Until next time,
Stay adaptive. Stay strategic.
And keep exploring the frontier of AI.
Fabio Lopes
XcessAI
💡Next week: I’m breaking down one of the most misunderstood AI shifts happening right now. Stay tuned. Subscribe above.
Read our previous episodes online!
