With 52% Of AT&T’s Traffic, DriveNets Aims At AI Networking

ChatGPT has ignited fear of missing out in the hearts of investors.

That FOMO was sparked by the unexpectedly large second quarter demand for Nvidia chips that power the machines that train and operate such large language models.

Most of the publicly-traded companies that stand to profit from generative AI — including Microsoft MSFT , Google GOOG , Amazon, Meta Platforms, Nvidia, Arista Networks ANET — are now well known.

Ra'anana, Israel-based DriveNets — a provider of cloud networking software — is a less well-known beneficiary of generative AI.

DriveNets — which has helped build cloud-based networks that now handle 52% of AT&T’s core production traffic, according to Communications Today — expects its technology to be in strong demand from companies like Microsoft and Google that are building and training LLMs.

Sadly for investors, DriveNets is still privately held; however, its CEO told me an IPO could be an option for raising growth capital.

(I have no financial interest in the securities mentioned in this post).

Why Generative AI Needs Cloud-Based Networks

To design, train, and operate LLMs, ChatGPT and other LLMs require substantial processing power. This represents a substantial market opportunity for providers of networking technology. Winners in the competition for this growing opportunity must satisfy demanding purchase criteria.

The market for cloud AI networking is growing rapidly. Between 2022 and 2027, industry analyst firm 650 Group, forecasts 38% average annual growth in the AI cluster connectivity market from $2 billion to “more than $10 billion.”

Due to the high cost of building and operating the networks needed to design, train, and operate LLMs, Microsoft and other hyperscalers are eager to maximize their processing power.

To do that, hyperscalers seek to process their AI workloads on networks with no idle time. As Ido Susan, DriveNets co-founder and CEO, explained, “AI compute resources are extremely costly and must be fully utilized to avoid 'idle cycles' as they await networking tasks.”

DriveNet deploys Ethernet communication technology which is more efficient than InfiniBand and occupies most of the market. DriveNets’ Network Cloud-AI boosts the utilization of AI compute resources — achieving up to 30% reduction in idle time, according to Susan.

One analyst is bullish on DriveNets’ ability to gain market share in cloud AI networking. According to Alan Weckel, founder and technology analyst at 650 Group, “DriveNets is an innovator that disrupted the traditional high-scale networking market, showing that a dis aggregated, white-box-based solution [— which saves money by running bespoke software on commodity hardware rather than expensive, proprietary switches —] can deliver greater network scale at a lower cost.”

Weckel sees DriveNets being a leader in AI networking as well. “They are now ready to do it again in AI networking, participating in early high-scale AI trials and [tapping into] the experience they acquired by building the largest core network in the world,” he said.

To be sure, DriveNets faces competition — including Arista Networks — which intends to provide customers a hybrid approach depending on the type of application.

As Arista CEO Jayshree Ullal told me in May 2023, “We are comfortable with the idea of a white box since that is how we started as software running on everyone else’s hardware. We want to provide customers what they need. For less mission-critical applications, they can use white boxes. For specialized applications — such as turnkey healthcare or financial solutions — they need Arista.”

Ullal agrees Ethernet is a crucial technology for generative AI as the amount of traffic that such networks must carry has exploded from the 1970s, when the maximum bandwidth envisioned was 2.95 megabits per second — well-below the more than 800 gigabits per second hyperscalers currently require.

Why DriveNets Was Started And How It Grew

DriveNets was started by Susan who sold his first company to Cisco Systems CSCO and took a leadership position there where he learned valuable business lessons.

In 2015, he co-founded DriveNets to do the same thing for networking as the cloud did for computing and storage. DriveNets has helped companies such as AT&T to lower their capital and operating expenses while accelerating their ability to innovate.

As Susan told me in a May 25 interview, before starting his first company he was in the Israeli army focused on technology. He co-founded Intucell, a networking technology company, which Cisco acquired in 2013 for $475 million.

He appreciates what he learned from Cisco. “Now we are competing with Cisco. I still have friends there. It is the best school for entrepreneurship. It sells $1 billion a week and you learn how the channel works and how to package products. I started studying computer science and got a degree from Cisco,” he said.

DriveNets was founded to solve the high costs and innovation-impeding nature of network equipment that locked customers into a single vendor. As Susan said, “While software, computing, and data storage have moved to the cloud, networking still locks in customers to a single vendor through proprietary hardware — making it more difficult for companies to move fast and innovate.”

DriveNets aimed to solve the problem by building “a next-generation software-centric service that would enable 24 x 7 deployment, be convenient to buy, and change how they manage networks. We help the world's largest network owners to detach a network’s growth from its cost,” he explained.

DriveNets has landed a plum client. “Over 50% of AT&T’s T network is running on our technology. They are enjoying 10-fold savings on capital expenditures and operating expenditures. We convinced AT&T to buy from us as a startup by building trust between people. My team is strong and comes from top industry leaders. We take the time, we go to the lab, we show it’s working,” Susan said.

AT&T said that DriveNets has helped it to carry more than half of its traffic via an “open disaggregated core routing platform.” Mike Satterlee, VP of AT&T’s Network Core Infrastructure Services, said the telecom giant began running some of its network backbone on DriveNets core routing software beginning in September 2020. While DriveNets expected to support all of AT&T’s traffic by the end of 2022, it ended the year with 52%, according to sdxcentral.

DriveNets has raised a total of $587 million in three funding rounds — most recently a $262 million financing in March 2022 that valued the company at nearly $1.6 billion, according to Pitchbook.

Susan is proud of DriveNets’ accomplishments. “We have tens of millions in revenue, we are doubling year over year. Our clients are top tier service providers and hyperscalers. We reduce the cost per bit and enable customers to innovate faster,” he said.

DriveNets Opportunity In AI Networking

DriveNets anticipates a significant opportunity in AI networking. As Susan explained, “We see an opportunity in generative AI. Our fabric uses Ethernet to connect tens of thousands of GPUs to train large language models. Our cloud network enables 32,000 Nvidia GPUs to handle 800 gigabnits. It is better than InfiniBand networks that can only handle hundreds of GPUs.”

He also aspires to take the company public. “My goal is for DriveNets to be a huge, unique company like VMWare that is bigger than HP HPQ or Dell. As far as exits are concerned — whether M&A or IPO — they are tools to raise money. Investors should provide us capital because we are in a big market controlled by incumbents with no innovation. It is time to disrupt them.”

If DriveNets chooses the IPO route, investors may want to take a close look.