EdgeQ is a chip design firm that covers a trifecta of my nerdiest interests. It’s using the open source RISC-V architecture to build a 5G chip that can also handle edge AI tasks, and it’s doing so in a way that will disrupt the business model of proprietary network equipment vendors. As an added bonus, this week it started offering a pretty innovative business model that will lower the up-front costs of providing network connectivity.
EdgeQ is now letting customers buy its feature-rich chips for a set price and will charge additional fees to customers as they decide they want to unlock more of the hardware’s capabilities. For example, if you want to add more AI capabilities, your licensing fee will go up. But for customers who don’t need that much intelligence, they won’t have to pay for it, either.
Vinay Ravuri founded EdgeQ after working at Qualcomm and trying to build ARM-based servers. He had been excited about the opportunities presented by what the networking industry called software-defined networking and the separation of the control plane from the data plane.
When that shift took place, companies that needed networking gear no longer had to have both the proprietary networking chips and the proprietary software from the same vendor in one box. Rather, they could separate the intelligence required in networking from the effort of moving the packets around. The results were hugely significant for the buyers of networking gear such as Amazon, Facebook, Microsoft, etc., and equally significant (albeit in a negative way) for the sellers of that gear, such as Cisco and Juniper.
The same trend hit the cellular industry to some extent, and as a result the networking gear purchased by telcos experienced a reduction in complexity and price. However, the separation isn’t entirely complete, especially when it comes to 5G, where the standard implementation is so complex that network operators really do need to buy their gear from a chip vendor and a software provider that work closely together.
Ravuri wants to change that. With EdgeQ, he sees several trends coming together that will substantially lower the cost of networking gear and give more flexibility to companies that want to build private 5G networks using unlicensed spectrum or perhaps private Citizens Broadband Radio Service (CBRS) spectrum. To that end, the three-year-old business has raised $51 million to create a chip that companies can use to build their own networks.
This chip allows businesses to create generic network boxes for enterprise customers and other startups, or software to handle the intelligence associated with the networks. It also makes networking hardware much cheaper and allows for far more flexibility and features on the software side. Ultimately, Ravuri would like for EdgeQ silicon to become the merchant silicon for the 5G networking era.
The chip doesn’t only handle networking, though. It also can tackle machine learning (ML) tasks at the edge. Edge-based ML is important because many tasks on 5G networks will require super-fast turnaround times. When a car is talking to another car on the road, for example, there’s no time to send data to the cloud and back again to make a decision. Sending that data to the local 5G tower or access point is much faster.
The chip was also designed using RISC-V, an open source instruction set that rivals those of ARM and Intel’s x86. The benefit of using something like RISC-V is its flexibility and the potential to build a chip without as much upfront investment in licenses.
Ruvari is also thinking about lower up-front costs for end customers of the 5G network, which is why EdgeQ has adopted a pay-as-you-go model for pricing. Each chip has basic 5G capabilities when shipped, for which customers will pay a set price. But each chip also has additional features — such as the ML features or more complicated 5G support — that could get turned on with a software update as well.
All of which means companies that are trying to build out 5G network services on demand could use EdgeQ silicon and pay for features only when they can charge for them. The flip side of this pricing is that there’s no chance to amortize higher capital expenses (CapEx) over the lifetime of the hardware. Everything becomes part of the operating expenses (OpEx).
Buyers of cloud computing have been making this OpEx vs. CapEx calculation for years, but EdgeQ’s chips and business model mean companies offering network connectivity as a service may soon face an adjustment in their own accounting. Add in the relatively new phenomenon of private networks owned by companies and new entrants to the connectivity market, and it’s going to become a lot tougher to be a carrier.