What makes MicroIP's design-less model so different from traditional semiconductor players?
In semiconductors, the winner is determined by business model. The industry moved from IDM to fabless, and now we are introducing the design-less model. Fabless means a chip company without a foundry; design-less means a high-tech company without an in-house IC design team.
Instead, these companies retain control of their algorithm, architecture, and product spec, but outsource the chip design and manufacturing.
Today’s market is fragmented. Previously, a company might ship 1 million chips per month. Now, many applications only require 1,000 units. To lead, you would need 10 chip products—and 10 design teams, which is too costly. That is why many IC design companies no longer go public. Their design departments are the major cost. At MicroIP, clients provide the architecture and spec, and we deliver the chip. Qualcomm, for example, sent their AI chip to us—we handled the design and tape-out.
In this example, is MicroIP replacing Qualcomm's design department or enabling something new?
Qualcomm was not sure the model would work, so they ran a dual-track: one chip team in-house, and one with us. Our design succeeded on the first tape-out. Afterwards, they laid off their internal AI chip team. This is happening more broadly—companies realise that outsourcing design reduces costs. While that displaces internal design teams, many of those engineers can join companies like MicroIP.
Is this model spreading among other major chipmakers?
Yes, it is a growing trend. One client, a software company with no hardware engineers, needed a custom NFC chip. They chose to outsource rather than build a team. This is no longer just IC companies outsourcing—software and platform providers are also turning to design-less services like ours.
ARM recently announced it will start manufacturing its own chips. What is your take on that?
That strategy has risks. By becoming a chipmaker, ARM may alienate its customers—who now become competitors. Historically, IP vendors that tried becoming chip vendors failed. I do not think it is a good move. In fact, we are already seeing companies leave ARM for RISC-V. Some of our clients originally used ARM CPU IPs but now want us to use RISC-V instead.
What is MicroIP doing in AI and AIoT to support this next wave?
We have two core platforms: CAPS (Closed Platform AI Power Solution) and CATS (Customisation AI ASIC Technology in the Solution). CAPS helps allocate the right AI compute power—high, medium, or low—based on the application. Many companies use high-power chips for all tasks, which drives up cost and energy use. We optimise this.
Our AI software stack supports multiple chipmakers: AMD, MediaTek, Rockchip, NERA, etc. For example, a client can assign AMD for high-performance tasks, Rockchip for mid-level, and NERA for low-power applications. This strategic allocation can reduce costs by 40% or more. For three chips, instead of spending $3,000, you spend $1,750.
How does using MicroIP’s platform impact product time-to-market, compared to using platforms by competitors like NVIDIA?
If a company uses our CAPS platform, their product can hit the market in 1–2 months. That is much faster than building software stacks from scratch. MediaTek, despite having their own hardware, uses our AI software to provide end-to-end solutions. Some media call us "Tiny NVIDIA" because we do both AI hardware and software co-design.
Is this level of AI chip expertise common outside the U.S. and Taiwan? What about Europe?
Europe is making progress. We joined the Taiwan-Czech Advanced Chip Design Research Center (ADCRC) to help train local IC designers. Different countries are focusing on distinct areas: France on biochips, Spain on communications chips, Czech Republic on design services, and Germany on manufacturing. We have trained 30 Czech graduates already, who now design chips locally. This is both an R&D partnership and strategic expansion.
Unlike countries trying to dominate every aspect, Europe is distributing responsibilities. Germany, for instance, was chosen for manufacturing due to its culture of precision and discipline. It is not about doing everything—it is about doing what you are best at. Europe’s is a realistic, structured approach. It may take 10 or 20 years, but they have started.
When you met with U.S. officials, what advice did you give on maintaining an edge in AI and semiconductors?
I spoke with Don Graves about the wrong strategy of blocking equipment exports to China. China can still integrate mature equipment effectively. What should be restricted is EDA software. Without EDA tools, chip design reverts 20 years.
I introduced two of our tools: Architecture Compiler and iProfiler. These tools drastically cut design time and improve chip performance before tape-out. We used them for Qualcomm’s AI chip—and succeeded on the first try.
What is your view on the development progress of new global semiconductor hubs?
Chips are now seen as national security assets. Countries are waking up to the fact that without local design and production, they are vulnerable. But developing a full chip ecosystem takes decades. Taiwan has spent nearly 50 years on this. Even if TSMC were to build a foundry in the UK, it would still need to ship chips back to Taiwan for packaging and testing. That is not a complete industry.
The US also cannot build the whole supply chain alone. Trump knew that—hence efforts to protect Taiwan. If Taiwan’s industry were lost, the global tech sector would collapse. Every country wants autonomy, but no one can recreate Taiwan’s ecosystem overnight.
What are your thoughts on Taiwan’s position in all this?
Taiwan did not “steal” chip tech from the U.S. As someone who studied in the U.S., I can say that 20 years ago, no American VC wanted to fund chip startups—it was all about software and biotech. The U.S. moved away from semiconductors. Taiwan never stopped investing. That is why we are ahead. As one executive put it: “We did not take the jobs. The U.S. did not want them.”
