Can custom AI outperform the tech giants?

Thinking Machines, an AI startup, has released its first software model named Inkling. Unlike the major tools you are likely familiar with—where you communicate with an AI via a website or an application that the creator controls—Inkling is what experts call open-weight. This means the engine behind the AI is freely available for anyone to download, inspect, and modify. Companies can take this base model and fine-tune it, which is the process of training the AI further on their own private records and data to make it an expert in their specific field.

To understand why this shift matters, imagine the difference between buying a pre-written cookbook and hiring a private chef. Public AI models like ChatGPT are the cookbook; they are excellent and broad, but they were trained on general information. They give everyone the same answers. A company using an open-weight model like Inkling, however, is like hiring a chef. They can take that base intelligence and train it on their unique, private internal knowledge—such as legal records, proprietary financial formulas, or custom coding styles. The model becomes a specialist in things that don't exist on the public internet. This process is called fine-tuning, where you give the AI extra practice in a specialized domain so it understands the specific language and needs of your organization. Because the company hosts the model themselves, they don't have to send their sensitive data to a third party to get an answer, keeping their business secrets secure.

The industry is beginning to move toward a split. For general tasks, we will likely keep using the big, all-purpose AI tools provided by tech giants. But for high-value work—like managing a hedge fund or running complex enterprise software—companies are realizing that a one-size-fits-all model leaves money on the table. By owning their own AI infrastructure, businesses can avoid subscription fees and, more importantly, ensure their private knowledge helps their own models improve rather than feeding into a competitor's system. Ultimately, this approach turns AI from a service you rent into an asset you own.