For decades, intelligence was the exclusive province of the human brain — expensive to cultivate, impossible to replicate, and inherently scarce. That era is ending. As artificial intelligence models grow more powerful and more accessible, a provocative thesis is gaining traction among technologists and investors alike: intelligence itself is becoming a commodity. And if that’s true, the strategic implications for the technology industry are profound.
The argument was laid out with striking clarity by Alfonso de la Rocha in his newsletter adlrocha, where he contends that the commoditization of intelligence will reshape not just the AI sector but the entire architecture of how software is built, deployed, and monetized. His central claim: when intelligence is cheap and abundant, the value shifts away from the models themselves and toward the infrastructure that delivers them — the data pipelines, the orchestration layers, the specialized hardware, and the application-specific integrations that make AI useful in practice.
The Model Layer Is Flattening — and That Changes Everything
De la Rocha’s thesis rests on an observable trend. The gap between the best large language models and the second-best is narrowing rapidly. OpenAI’s GPT-4 was once considered leagues ahead of the competition. Today, Anthropic’s Claude, Google’s Gemini, Meta’s open-source LLaMA family, and a growing roster of Chinese models from companies like DeepSeek are all converging on similar capability benchmarks. When multiple providers can offer comparable intelligence, the intelligence itself ceases to be a differentiator. It becomes, in economic terms, a commodity — like electricity, bandwidth, or compute cycles.
This is not a hypothetical scenario. It is already playing out in pricing. OpenAI has slashed API costs repeatedly over the past 18 months. Google offers generous free tiers for Gemini access. Meta gives away its model weights entirely. As de la Rocha writes in his newsletter, the race to the bottom in model pricing mirrors what happened with cloud computing a decade ago: once Amazon Web Services, Microsoft Azure, and Google Cloud were all offering similar infrastructure, the competition shifted to managed services, developer experience, and vertical specialization.
The Infrastructure Thesis: Where Value Migrates When Models Are Cheap
If intelligence is cheap, what becomes expensive? According to de la Rocha, the answer is everything around the model. Data quality and curation, fine-tuning for domain-specific tasks, retrieval-augmented generation (RAG) architectures, inference optimization, and the ability to orchestrate multiple AI agents in concert — these are the layers where competitive advantage will accumulate. The model is the engine, but without the chassis, the fuel system, and the driver, the engine is just a block of metal.
This view has found increasing support in the venture capital community. Recent reporting from Reuters has highlighted a shift in AI investment away from foundation model companies and toward infrastructure startups — firms building tools for model evaluation, data labeling, AI observability, and deployment orchestration. The logic is straightforward: if every company will soon have access to roughly equivalent AI capabilities, the winners will be those who can deploy those capabilities most effectively within specific business contexts.
The Open-Source Accelerant
Meta’s decision to open-source its LLaMA models has been one of the most consequential strategic moves in the AI industry. By making high-quality models freely available, Meta effectively commoditized the model layer for everyone — including its competitors. But as de la Rocha argues, this was not an act of altruism. It was a calculated bet that Meta would benefit more from a world where intelligence is abundant than from one where it is scarce. If every developer can build on top of LLaMA, the demand for the infrastructure that supports those applications — much of which runs on Meta’s preferred hardware and software stack — increases enormously.
The open-source dynamic has also empowered a new class of smaller, more specialized models. Companies like Mistral AI in France and various Chinese AI labs have demonstrated that you don’t need tens of billions of dollars to produce models that are competitive for specific use cases. Fine-tuned models with 7 billion or 13 billion parameters can outperform much larger general-purpose models on narrow tasks, from legal document analysis to medical diagnosis to code generation. This proliferation of capable smaller models further erodes the moat around any single large model provider.
DeepSeek and the Global Commoditization Pressure
The emergence of DeepSeek, a Chinese AI lab that has produced models rivaling Western counterparts at a fraction of the reported cost, has added urgency to the commoditization thesis. DeepSeek’s R1 model, released earlier this year, demonstrated reasoning capabilities that surprised many industry observers, and it did so with what the company claimed were significantly lower training costs than comparable Western models. The implication was clear: if a relatively unknown Chinese lab can produce frontier-quality intelligence cheaply, the idea that any single company can maintain a durable advantage through model quality alone becomes increasingly difficult to defend.
This competitive pressure from China has also raised questions about the long-term viability of the massive capital expenditures that U.S. tech giants are pouring into AI infrastructure. Microsoft, Google, Amazon, and Meta have collectively committed hundreds of billions of dollars to building data centers and acquiring GPUs. If intelligence becomes cheap enough that smaller players can compete effectively, the return on those investments becomes less certain. As de la Rocha notes, the commoditization of intelligence doesn’t mean AI is less valuable — it means the value is distributed differently across the stack.
The Application Layer: Where the Money Actually Gets Made
History offers instructive parallels. When cloud computing commoditized raw infrastructure, the value migrated upward to SaaS applications. Salesforce, Workday, and ServiceNow built enormous businesses not by running better servers but by solving specific business problems on top of commodity infrastructure. De la Rocha suggests a similar pattern will emerge in AI: the most valuable companies will not be those that train the best models but those that build the best applications on top of commodity intelligence.
This is already visible in the market. Companies like Harvey (legal AI), Abridge (medical documentation), and Glean (enterprise search) are building vertically focused products that use foundation models as components rather than as the core product. Their value proposition lies not in the intelligence itself but in how that intelligence is integrated with domain-specific data, workflows, and compliance requirements. A general-purpose chatbot can draft a legal brief, but a purpose-built legal AI tool that understands case law, jurisdiction-specific rules, and a firm’s particular style guide is worth far more to a paying customer.
The Counterargument: Intelligence May Not Commoditize as Fast as Expected
Not everyone agrees with the commoditization thesis. OpenAI CEO Sam Altman has repeatedly argued that the next generation of AI models will be so significantly more capable than current ones that the gap between leaders and followers will widen, not narrow. If GPT-5 or its successors achieve genuine breakthroughs in reasoning, planning, or agentic behavior, the model layer could re-differentiate itself. In this view, the current convergence is temporary — a plateau before the next leap.
There is also the question of data moats. Companies with proprietary access to large, high-quality datasets — think Bloomberg in finance, Epic in healthcare, or Thomson Reuters in legal — may be able to fine-tune models in ways that are difficult for competitors to replicate, even if the base models are equivalent. The intelligence may be commodity, but the data that shapes it for specific purposes is not. This creates a more nuanced picture than a simple commoditization narrative suggests.
What Happens When Everyone Has a Brain
De la Rocha’s framing raises a philosophical question as much as an economic one. If intelligence can be provisioned on demand, like electricity from a socket, what does that mean for the humans and organizations that have traditionally derived their value from being smart? The answer, he suggests, is that execution, creativity, and judgment become the scarce resources. Knowing the answer is less valuable when everyone can ask a model for the answer. Knowing which question to ask, and what to do with the response, is where human value persists.
For the technology industry, the practical takeaway is that the current gold rush around training ever-larger models may be approaching its point of diminishing returns — not because the models aren’t getting better, but because the marginal improvement matters less when the baseline is already very high. The companies that will define the next era of AI are likely not the ones building the biggest models but the ones building the most effective systems around those models. Intelligence is becoming a commodity. The question now is who will build the best products with it.
WebProNews is an iEntry Publication