Author:Li Jia, Wall Street News
AI is entering a new stage, moving from 'knowing how to answer' to 'being able to execute'.
According to the Wind Trader platform, Goldman Sachs' latest report shows that AI commercialization is shifting from 'per-seat subscription' to charging based on consumption, transaction volume, and results; meanwhile, Agents are evolving from assistive tools to workflow executors, and industrial value is migrating from the models themselves to proprietary data, business context, and domain-specific expertise.
This means the AI industry competition is shifting from 'whose model is stronger' to 'who can truly master workflows'. Model capabilities remain important, but the ability to enter enterprise production environments, understand business context, and reliably complete tasks will become a more critical competitive barrier.
This judgment comes from Goldman Sachs' recent on-the-ground investigation of the Silicon Valley AI industry chain. From August 18th to 19th, Goldman Sachs conducted its third consecutive year of visits to AI startups, top venture capital institutions, and researchers from Stanford University, UC Berkeley, and UC San Francisco. Goldman Sachs believes that as Agents accelerate their implementation, the value distribution among frontier models, open-source models, world models, enterprise software, and proprietary data will change.
Agent Implementation: What Enterprises Truly Lack is Not Capability, but 'Controllability'
If the past AI solved the problem of 'helping people complete tasks', Agents are now trying to solve 'completing tasks on their own'. However, during large-scale enterprise deployment, the biggest obstacle may no longer be model capability, but how responsibility is divided, and whether the entire execution process can be controlled.
The report cites Stanford researchers, noting that most enterprises are still in a manual supervision mode. Especially in legal, risk control, insurance, and audit fields, once a model makes an error, questions of who bears responsibility, how to trace the process, and whether it can be corrected promptly may be as important as the model's capability itself.
Therefore, workflows most likely to be automated first typically have three characteristics: clear decision boundaries, verifiable results, and errors that can be rolled back. Invoice processing is a classic case. AI is responsible for extracting fields and performing checks, low-confidence cases are handed over for human review, and then accounted for through reversible ERP processes.
This also means that information service providers with trusted content, validated domain-specific models, and mature regulatory relationships are more likely to be the first to enter enterprise production environments.
Model Competition: Frontier Models and Open-Source Models Move Towards Specialization
Regarding the debate around 'open-source or closed-source', the signal released by Goldman Sachs' investigation is not an either-or choice, but that different models may correspond to different layers of workflows.
The frontier model camp argues that enterprise benchmark tests often underestimate model capabilities. In real production environments, the business losses caused by a decline in model precision may far exceed the savings in inference costs. Therefore, although several AI-native companies claim to adopt a multi-model strategy, they still heavily rely on frontier models in their core production environments.
Another view holds that the vast majority of enterprise workflows do not require frontier-level intelligence. As open-source model performance continues to improve, customers are increasingly willing to accept limited performance loss in exchange for lower inference costs. One venture capital firm estimates that within the next 12 to 18 months, about 90% of inference tokens will flow to open-source models.
This implies that the future AI model market may form a clearer division of labor: frontier models handle high-value, high-reliability complex tasks, while open-source models undertake larger-scale standardized tasks and the majority of token consumption.
World Models: AI Computing Power May See a Second Growth Curve
Over the past 18 months, researchers have increasingly shifted their focus from LLMs to 'world models'.
Unlike LLMs primarily trained on internet data, world models need to understand the environment, causal relationships, physical laws, and dynamic interactions in the real world; their data comes more from physical systems, specific industries, and actual operational scenarios. This means the importance of proprietary data may rise further.
Goldman Sachs believes that the problem space corresponding to fields such as physics, industry, science, and robotics is far larger than pure text generation, and these workflows often require higher computing power investment. As AI further moves from the digital world into the physical world, the computing power demands for model training, simulation, and inference may also see a new growth curve.
Goldman Sachs expects that computing power demand may increase approximately 24-fold over the next 5 years, and supply-demand tightness is expected to last longer, directly benefiting cloud computing and computing infrastructure companies like Microsoft, Oracle, and CoreWeave.
