The funding landscape for AI chip startups, which was once extremely favorable, is now facing some challenges as Nvidia solidifies its position as the industry leader.
In the first three quarters of 2023, U.S. chip firms managed to raise only $881 million, which is a significant decline from the $1.79 billion they raised during the same period in 2022. In 2022, Mythic, an AI chip company, faced a financial crisis and came close to suspending operations. Meanwhile, Graphcore, a previously well-funded competitor, is currently grappling with increasing losses.
However, a certain startup seems to have achieved remarkable success in the highly competitive and rapidly expanding field of AI chip technology.
Orr Danon and Avi Baum, who have a background in wireless connectivity at Texas Instruments, founded Hailo in 2017 and develop custom chips for running AI tasks on edge devices. Hailo’s chips are highly efficient in executing AI tasks, consuming less memory and power compared to regular processors. This makes them an excellent choice for small, portable devices like cars, smart cameras, and robotics that operate offline and rely on battery power.
“I co-founded Hailo with the goal of making high-performance AI accessible on a large scale beyond the confines of data centers,” Danon shared with Eltrys. Various tasks, such as object detection, semantic segmentation, and other similar applications, utilize our processors. Additionally, we use AI technology to enhance images and videos. Recently, we have deployed extensive language models (LLMs) on edge devices like personal computers and infotainment electronic control units, among others.
Several AI chip startups are still struggling to secure a significant contract, let alone multiple ones. Hailo currently boasts over 300 customers across a wide range of industries, including automotive, security, retail, industrial automation, medical devices, and defense, according to Danon.
This week, Hailo received a significant boost in funding with a $120 million investment from a group of financial backers. Investors view this investment as a promising sign for Hailo’s future prospects. Danon expressed that the new capital will allow Hailo to take advantage of all upcoming opportunities and lay the foundation for sustained growth.
“We have a strategic advantage in bringing AI to edge devices, which will greatly enhance the reach and impact of this incredible new technology,” stated Danon.
Now, you may be curious if a startup such as Hailo can truly compete with established chip giants like Nvidia and, to a lesser extent, Arm, Intel, and AMD. An expert, Christos Kozyrakis, a professor of electrical engineering and computer science at Stanford, believes that accelerator chips, such as Hailo’s, will be essential as AI continues to expand.
Kozyrakis told Eltrys, “We cannot overlook the disparity in energy efficiency between CPUs and accelerators.” Using accelerators for critical tasks like AI and keeping a processor or two available for programmability maximizes efficiency.
Kozyrakis acknowledges that the long-term success of Hailo’s leadership could be challenged if the AI model architectures that their chips are optimized for become outdated. According to Kozyrakis, the potential challenge lies in software support. If a sufficient number of developers are not willing to familiarize themselves with the tooling designed for Hailo’s chips, it could pose a problem.
“The software ecosystem poses the majority of challenges when it comes to custom chips,” stated Kozyrakis. It’s clear that Nvidia has a significant edge in the field of AI compared to other companies. Their extensive investment in software for their architectures over the past 15 years has given them a substantial advantage.
However, with a substantial $340 million in the bank and a workforce of approximately 250, Danon is filled with confidence regarding Hailo’s future trajectory, at least for the immediate future. He recognizes the startup’s technology as a solution to the common obstacles faced by companies utilizing cloud-based AI inference, such as latency, cost, and scalability.
“Conventional AI models heavily depend on cloud-based infrastructure, which can sometimes lead to problems like latency and other obstacles,” stated Danon. They lack the ability to provide instant insights and notifications, and their reliance on networks puts reliability and integration with the cloud at risk, raising concerns about data privacy. Hailo is tackling these challenges by providing solutions that function autonomously from the cloud, enabling them to handle significantly larger volumes of AI processing.
Curious about Danon’s perspective, I inquired about generative AI and its significant reliance on the cloud and remote data centers. Hailo seems to understand the potential risks associated with the dominant top-down, cloud-centric approach, as exemplified by OpenAI’s operational strategy.
According to Danon, generative AI is actually creating a surge in demand for Hailo’s hardware.
“There has been a significant increase in the demand for edge AI applications across various industries, including airport security and food packaging,” he stated. “The increasing popularity of generative AI is fueling the demand for processing LLMs locally.” Customers from various industries, including compute, automotive, industrial automation, security, and more, are now requesting this capability.”
