The funding landscape for AI chip startups, once sunny as a mid-July day, is starting to look shady as Nvidia asserts its dominance.
US chip companies raised just $881 million in funding from January 2023 to September 2023, down from $1.79 billion in the first three quarters of 2022, according to a recent report. . Mythic, an AI chip company, ran out of cash in 2022 and was close to being forced to shut down operations. Meanwhile, rival Graphcore, once well-capitalized, is now facing mounting losses.
But one startup appears to have found success in the ultra-competitive and increasingly crowded field of AI chips.
Hailo was co-founded in 2017 by Orr Danon and Avi Baum, previously CTO of wireless connectivity at microprocessor company Texas Instruments, to design purpose-built chips for running AI workloads on edge devices. I am. Hailo's chips perform AI tasks with less memory and power consumption than typical processors, making them great candidates for compact, offline, battery-powered devices such as cars, smart cameras, and robotics. .
“I co-founded Hailo with the mission of making high-performance AI available at scale outside the confines of the data center,” Danon told TechCrunch. “Our processors are used for tasks such as object detection and semantic segmentation, as well as AI-powered image and video enhancement. used to run large-scale language models (LLMs) in
Many AI chip startups have yet to land a single large contract, let alone dozens or hundreds. But Danone says Hailo currently has more than 300 customers in industries such as automotive, security, retail, industrial automation, medical devices and defense.
In a bet on Hylo's future, financial backers including Israeli businessman Alfred Akilov, car importer Derek Motors and venture capital platform OurCloud this week announced a $120 million extension to the company's Series C. Danone said he invested $10,000 in Hailo. This will “enable Hailo to take advantage of all opportunities in the pipeline” while “setting the stage for long-term growth.”
“We are in a strategic position to bring AI to edge devices in a way that significantly expands the reach and impact of this remarkable new technology,” Danone said.
Now, you may be wondering if a startup like Hailo can really compete against big chip companies like Nvidia and, to a lesser extent, Arm, Intel, and AMD. One expert, Christos Kozirakis, a professor of electrical engineering and computer science at Stanford University, thinks so. He believes accelerator chips like Hailo will become “absolutely necessary” as AI becomes more widespread.
“The difference in energy efficiency between CPUs and accelerators is too large to ignore,” Kozyrakis told TechCrunch. “We use accelerators to streamline important tasks (such as AI), and we set aside one or two processors for programmability.”
Kozirakis believes longevity poses a challenge for Hailo's leadership. For example, chips whose AI model architecture is designed to operate efficiently will become obsolete. Kozirakis said software support could also be an issue if a significant number of developers aren't willing to learn how to use the tools built around Hailo's chips.
“Most of the challenges with custom chips are in the software ecosystem,” Kozirakis said. “For example, this is where Nvidia has a big advantage over other companies in his AI space, because Nvidia has invested in software for his architecture for over 15 years.”
But with $340 million in the bank and about 250 employees, Danone is confident in Jairo's path forward, at least in the short term. He sees the startup's technology as able to address many of the challenges enterprises face with his cloud-based AI inference, particularly latency, cost, and scalability.
“Traditional AI models rely on cloud-based infrastructure and often suffer from latency issues and other challenges,” Danone said. “Real-time insights and alerts do not work, and dependence on the network compromises reliability and cloud integration, raising data privacy concerns. Hailo provides a solution that operates independently of the cloud. We are addressing these challenges and making it possible to handle larger amounts of AI processing.”
Curious about Danone's perspective, I asked about generative AI and its heavy reliance on the cloud and remote data centers. Indeed, does Hailo believe that the current top-down, cloud-centric model (such as OpenAI's modus operandi) is an existential threat?
On the contrary, Danone said generative AI is driving new demand for Hailo's hardware.
“Demand for edge AI applications has skyrocketed in recent years across most industries, from airport security to food packaging,” he said. “The new surge in generative AI is driving this demand even further, as we are seeing a desire to process LLM locally from customers in industries such as industrial automation and security, as well as the computing and automotive industries. ”
How about that.