A new tech company called Fungible officially launched in early February. The company is targeting the data center market with a product that will combine custom silicon with integrated compute, storage, and networking.
Cofounded by Pradeep Sindhu, who also founded Juniper Networks; and Bertrand Serlet, a software engineering lead at Apple, the startup has raised $32.5 million.
At present there’s not a lot known about this venture, but I had the chance to speak with Mr. Sindhu to get a little more detail about what the company’s up to.
It sounds to me like the company will develop a scale-out, hyperconverged platform that targets data-intensive workloads. As with other hyperconverged platforms, one of Fungible’s goals is to bring data as close to the compute layer as possible.
“There are computational problems at the intersection of big data, scale-out architecture, and artificial intelligence,” said Sindhu in an interview. “Those are the problems we’re trying to address.”
The platform will be built around a small number of compute/storage/networking building blocks. Customers can add blocks to ramp up capacity, and the system is designed for resilience—that is, if one or more elements in a block fails, the application or service will still continue to function.
Custom Silicon As A Differentiator
Unlike most hyperconverged offerings, which run on commodity x86 processors, Fungible is developing custom silicon.
This is a potentially risky move. The x86 processor brings significant benefits to vendors, including falling costs, a proven platform, a robust developer ecosystem, and a regular cycle of performance increases.
However, Sindhu is betting that as Moore’s Law runs up against physical limitations on the number of transistors that can fit on a chip, future performance gains on x86 will flatline.
“Because of the scarcity in the automatic improvement of performance, we’ll need new architectures for storage, compute, and networking,” he said.
(For a detailed view of his thoughts on the end of Moore’s Law, here’s a Packet Pushers podcast recorded with Sindhu in 2016.)
Rather than try to squeeze more performance out of the x86 platform, Fungible’s goal is to develop silicon that is optimized for specific domains.
“We are optimizing our silicon for data movement and data storage,” he said.
The trick for Fungible will be to demonstrate a significant enough increase in performance and other capabilities to justify the price premium that custom silicon demands.
But Not Just Silicon
In our interview, Mr. Sindhu wanted to make clear that Fungible isn’t just a chip company. While the silicon will enable advances over general-purpose chips, he also stressed that how Fungible assembles the other infrastructure components is also key to the product’s value.
“Packaging things together is relevant when you have problems of heat, power, and so on, which is very important, especially in data centers,” he said.
The Fungible system will also be programmable, both at the silicon level and for applications. APIs will enable customers to write software to run on top of the infrastructure.
“Just because we are building a system and embedding our silicon and software doesn’t mean other people won’t be able to program it,” he said.
I asked about possible use cases. Mr. Sindhu said it was too early to get into specifics, but Fungible will target opportunities where there are large amounts of data and data movement, as well as customers who could benefit from real-time analysis, such as the financial services industry, or government agencies that deal with national security threats.
He did outline three general customer buckets: hyperscale Web players, enterprises building out cloud architectures, and service providers.
Not A Spin-In
Because Juniper has invested in the startup, I asked if Fungible was positioned as a spin-in along the lines of Insieme and Cisco. Mr. Sindhu’s response was direct:
“No. This is an independent startup.”
That said, he did note that the new company will work closely with Juniper around networking. And while Sindhu is CEO of Fungible, he’s also staying on at Juniper in the role of Chief Scientist. Given this relationship, it seems logical to assume that, should Fungible be ready for some kind of exit in the future, Juniper will have an inside line.
As for when more details about the product and company will become available, Mr Sindhu said early 2018.
As mentioned, the company has raised $32.5 million in a Series A round. In addition to Juniper Networks, investors include Mayfield, Walden Riverwood Ventures (WRV), and Battery Ventures.