AI Chip Startups Seek An Edge By Enlisting The Channel

Solution providers are finding new partnership opportunities as a flood of AI chip startups enter the market—but there is risk involved.

Fabrizio Del Maffeo was thousands of miles from his company’s headquarters in the Netherlands. It was March at a trade show in Las Vegas, and his team was there to take on the hottest, most dominant AI chip provider in the world.

Axelera AI wasn’t looking to show how its AI chips could rival Nvidia’s most powerful and expensive GPUs along with its bespoke, liquid-cooled server racks for the most demanding AI workloads in the data center.

Instead, Del Maffeo’s team was there to show how the company’s Metis accelerator cards can run a computer vision model detecting hundreds of people and objects using an 8K screen resolution at a fraction of the cost of alternative offerings.

“There are many things you can do with just a few boards, which cost a few hundred dollars. You don’t have to spend $10,000 on the problem,” said Del Maffeo, founder and CEO of Axelera AI.

But Del Maffeo’s team hit a snag: The equipment they had shipped for the event didn’t arrive on time, so they scrambled to buy an off-the-shelf desktop computer from a local store and borrowed an 8K camera from Axis, another company at the trade show.

This did not present an issue, however, since Axelera AI was able to plug the Metis cards into the PC, hook up the camera and get the software running with ease.

For Del Maffeo, the rush to integrate everything together and the resulting demonstration reflects the big opportunity he sees in the channel: equipping solution providers with a plug-and-play product that can run a variety of high-performance AI workloads and is more affordable than other silicon offerings.

“We want to sell cards that integrators can plug inside their computer, install the software and just run or develop their algorithm,” said Del Maffeo, an Advantech and Aaeon veteran who got his first taste of the channel working for Italian distributor Esprinet.

To seize on that opportunity, Axelera AI recently launched its global Partner Accelerator Network program with more than 15 partners and promoted its CMO, Alexis Crowell, to grow the startup’s channel business in North and South America as general manager.

“Just given the size and reach of my company, the only way we can fulfill the global demand is through these channel partners.

I can’t do it otherwise,” Crowell said. Axelera AI is among several venture-backed startups selling AI chips that hope to leverage the expertise and reach of channel partners to land new customers in a variety of verticals and grow revenue in a world dominated by Nvidia.

One U.S.-based system manufacturer that sees big channel potential for chip startups like Axelera AI is Miami-based mini PC maker Azulle, which found the startup’s Metis chip to be the most powerful among Nvidia alternatives for edge computing and designed a special motherboard to maximize the chip’s performance.

Alex Rodriguez, CEO of Azulle, said Axelera AI is “going to do very well” if the company is able to market the Metis platform properly, mainly because of how the chips are much less expensive than what Nvidia is providing, which could play well with smart city applications.

“We can come in and say, ‘We can give you a device that’s 65 percent cheaper and it runs just as well if not better.’ That helps [customers sell their] product more,” he said.

But while these chip startups are creating greater competition and giving solution providers more options for how they build computers and servers for customers, there is a risk that any startup could retrench from the channel or shut down altogether.

In April, the CEO of Toronto, Ontario-based Untether AI told CRN his startup was beginning to recruit for new channel partners after striking relationships with companies such as Asa Computers, Avnet, Boston and System 76.

The CEO, Chris Walker, who was previously at Intel for 29 years, said his company was focusing on partnerships with solution providers and other companies in the tech ecosystem to build vertical solutions, like video analytics for retail.

“Everybody’s adding their own value to it. That’s always been the value of working with channel partners and [bringing] vertical solutions to the market,” he said.

Within several weeks, however, Untether AI had shut down, with Walker leaving the startup shortly before and AMD subsequently acquiring its engineering team.

The CEO of a U.S. systems integrator who partnered with Untether AI said there is a general desire in the channel to have multiple suppliers in the burgeoning AI computing space, even if they are startups.

“Just like how customers want to diversify away from completely being dependent on Nvidia, so do partners,” said the systems integrator leader, who asked to not be named to speak candidly about such dynamics in the industry.

Nevertheless, the systems integration CEO is skeptical of how successful startups can be in the face of Nvidia’s dominance, reflected by how much the tech industry has optimized and built software around the AI computing giant’s chips and code base.

“Unless something actually happens to Nvidia, I don’t see any realistic chance that people will ever be able to make even one dent in their business,” he said.

Even if a startup doesn’t completely shut down, there is still a risk that it could change direction and reduce its investments in the channel.

This is what happened with Graphcore, a Bristol, U.K.-based AI chip designer that launched a partner program in 2020 and then terminated it a few years later. While the startup said last year that it still worked with channel partners, it was putting greater emphasis on its business with cloud service providers. And that was before the company got acquired by Japanese investment giant SoftBank in pursuit of the firm’s AI goals.

One challenge with chip startups is that businesses, particularly enterprises and hyperscalers, are reticent about investing in technology that may not have long-term viability, according to Matt Fornito, who was the global head of AI for Trace3 when Graphcore named the Irvine, Calif.-based solution provider as an inaugural partner.

“They want to go with the most stable, most secure [technology that is] most likely to be there five years from now, and so you don’t know what’s going to happen with something like a Graphcore,” said Fornito, who is now CEO of AI Advisory Group, a Denver-based firm that advises businesses on how to use AI effectively.

But solution provider and chip startup executives believe there is still plenty of opportunity to build a meaningful business, particularly at the edge, which Del Maffeo considers “highly fragmented” and for which there is no clear winner right now.

The Axelera AI leader is also making a bet on enterprise data centers because he believes there are many businesses that need a cost-effective option.

“Not everyone can afford this, and not everyone needs this, but there is more and more need to run AI within enterprises on-premises,” he said.

Another chip startup betting on the potential of AI computing at the edge is San Jose, Calif.-based SiMa.ai, which runs a global partner program and works with a variety of systems integrators, VARs, distributors and other types of partners.

Krishna Rangasayee, SiMa.ai’s founder and CEO, sees a few key advantages with his company’s MLSoC system-on-chips, particularly the silicon’s performance and power efficiency as well as the startup’s open-source software.

But Rangasayee, who was previously executive vice president of global sales at programmable chip designer Xilinx, believes SiMa.ai’s investments in the ecosystem—which also include ISVs and ODMs—could be its “largest differentiation” over time.

“At the edge, ecosystem is everything. Partners are everything. So we are more partner-focused people than most companies would think of. But if they’re going to be 60 percent, 70 percent of my revenue, that’s definitely what we want to invest in,” he said.

Like Axelera AI, another chip startup called NeuReality sees ample opportunities in enterprise data centers with channel partners.

But instead of providing an accelerator chip to compete with GPUs, the Israel-based company is selling what it calls the “first true AI CPU,” which is designed to increase utilization for clusters of GPUs and similar chips.

NeuReality co-founder and CEO Moshe Tanach said low utilization for AI chip clusters has been a significant problem in the industry.

“Instead of this diminishing return, we can boost it five or 10 times depending on the use case and the type of processor that we’re attached to in terms of queries per second, which, at the end of the day, lowers the cost of ownership,” he said.

To get customers started, NeuReality has developed a reference server called the NR1 Appliance that contains its namesake NR1 chip. And Tanach said the plan is to have OEMs white-label the system so that they can then sell it through channel partners.

Tanach, who previously worked at semiconductor firm Marvell Technology, said NeuReality is making a channel push because he knows that’s how many businesses buy technology.

“For us, channels are super important,” he said.