Last week, while quoting a smart device, I got stuck again on whether it would actually run
At ten on Thursday night, I was in Hangzhou talking with a client about a small temperature-and-humidity device. What they cared about most was not how it looked, but whether power consumption would suddenly jump when they switched to a different chip. The supplier said it would take two days to answer. I told them it was no problem, while quietly knowing that if this surfaced after prototyping, the rework fee could exceed my consulting fee.
That moment when the design does not exist yet, but the team is already mentally simulating failure, is exhausting. I used to gather parameters on the spot and compare each option. It was slow, and I often missed edge cases. I got stuck this way several times too: I made a call based on experience, only to have the test data overturn me later.
OpenAI is bringing models into the chip-design workflow
OpenAI recently introduced model capabilities for chip design. A designer provides requirements in text, and the model helps generate circuits, simulation results, and checks whether the design meets requirements such as power consumption and performance. It is not directly replacing chip engineers with a chatbot; it gives engineers a faster way to compare several options.
The article mentioned that the engineering team at chip company Cerebras is already using these models to optimize processor designs. Research institutions are also using them to explore different circuit structures. In our language: this is not about letting someone without electronics knowledge send a design directly into production. It gives people who know the field another assistant for catching mistakes early.
If I were building a hardware prototype, an embedded product, or evaluating a supplier's proposal, I would keep an eye on this development. Still, the results in the press release do not mean I can hand over a live order today. Professional tools, engineering judgment, and physical testing are still required.
How much of this can I replicate today?
I did a rough calculation based on public information. Beyond model usage, integrating this into a complete chip-design workflow requires electronic design tools, supplier documentation, and someone who understands circuits. With an outsourced project, a preliminary design assessment typically costs RMB 5,000–20,000. Organizing the information and running basic comparisons myself takes about 3–7 days. The technical barrier is not low: I need at least a working understanding of power consumption, clocking, and signal latency, plus a sense of which conclusions must be tested on real hardware.
The first step is not buying an enterprise package. I would open OpenAI's relevant official entry point and check whether it currently offers model access, research documentation, or a partnership application. I would also pull up a completed project from the past as a practice case. The technical barrier, in one line: without basic circuit knowledge, I cannot even judge whether its “optimization recommendations” are real.
There is no urgency to try this now. Chip design itself is a long way from most side-project operators. If I focus on consulting, content, or software products, I would note this direction and wait until I have a concrete hardware requirement. That would be cheaper.
Here is how I would arrange it by stage
If I were just starting and serving only software or content clients, I would not invest yet. If I had one or two hardware clients, I would ask suppliers for complete parameters first, then use the models to organize and compare them, while keeping final component selection out of the model's hands. If I were scaling and managing several product lines, I would look into enterprise partnerships and an internal workflow, with engineers reviewing every critical conclusion.
For a small team like ours, the realistic value may not be “letting AI build chips.” It may be turning expensive physical trial and error into cheaper paper-based simulation earlier. Not everyone needs this tool. But if I am being worn down by prototype lead times and repeated design changes, it is worth running one small validation.