KR — ZDNet Korea
Microchip Bets on Turnkey Solutions Combining Hailo's AI Accelerators
Chips1 min read
Microchip Technology, a large U.S. chip company, bought Hailo, a startup that makes AI chips for devices at the edge. The deal shows how big chip makers now want to sell complete AI systems instead of single chips.
What happened
Mark Reiten, senior vice president at Microchip, explained in an interview why the company bought Hailo. Hailo makes AI chips that run on devices like robots and drones, not in big data centers. Microchip plans to combine Hailo's AI chips with its own wide range of chips to offer a complete, ready-to-use solution.
Key facts
- Hailo supports more than 180 AI models that can be used without extra work, according to Microchip.
- Hailo's chip design puts a large amount of memory inside the NPU (the chip part that runs AI).
- Other AI chip startups, such as Habana Labs and Groq, were also bought by big companies.
Background
AI chips need a lot of memory, called DRAM, and they need software that works well with the hardware. Both are expensive and hard to maintain. Small startups that sell only one chip often lack the money and staff to handle these problems. Their rivals, meanwhile, can offer many kinds of chips at once.
Why it matters
The article says the era of startups competing on chip speed alone is ending. Large companies that can supply every chip and software tool a customer needs may now win. Microchip also hopes Hailo's design will cut the cost of memory and power use, which could make AI devices cheaper and more energy-efficient.
What to watch
- Whether Microchip's combined chips and software sell well to robot and drone makers.
- How the price of DRAM memory changes and whether it affects AI device costs.
- Whether other small AI chip startups get bought by large companies.
Glossary
- Microchip Technology
- A U.S. company that makes many kinds of chips, such as analog, power, motor, and communication chips.
- Hailo
- A startup that makes AI chips for devices like robots and drones, which process data on the device itself.
- NPU
- A chip part built to run AI tasks quickly and with less power.
- DRAM
- A type of computer memory that stores data while devices are running. AI chips often need a lot of it.