The data center network, for most of its history, has been a place of quiet, expensive rigidity. You bought a switch chip from a handful of vendors, and you got the protocols they decided to bake into silicon. Barefoot Networks, founded in 2013, looked at that lockstep and saw a software problem waiting for a hardware answer. Their solution was Tofino, a family of Ethernet switch ASICs that could be reprogrammed on the fly using the open P4 language, promising the performance of fixed-function silicon with the agility of software [Wikipedia].
For the cloud giants and telecoms building the planet's largest networks, that promise translated into a simple economic proposition: control. It was a bet that the industry's most demanding customers would pay a premium to escape the one-size-fits-all model, and it attracted over $150 million from investors who believed the same [SDxCentral].
The programmable wedge
The core of Barefoot's bet was the P4 programming language. Before Tofino, you could write a P4 program, but you couldn't run it at line-rate on cutting-edge silicon. Barefoot's Tofino chips were the first to execute P4 programs directly in hardware, at speeds up to 6.5 terabits per second [Yahoo Finance, Feb 2017].
This created a powerful wedge. For customers like AT&T, it meant being able to implement custom network telemetry. For Alibaba, Tencent, and Baidu, it allowed the creation of custom protocols tailored to their specific data center architectures [GlobeNewswire, 2017].
A founding team built for the fight
Building a new Ethernet switch ASIC from scratch is a capital-intensive, credibility-dependent endeavor. Barefoot's founding team was anchored by repeat founder Nick McKeown, a Stanford professor and co-founder of Nicira [Wikipedia]. He was joined by Martin Izzard, an early CEO with a deep background in Ethernet switch silicon from Broadcom, and Pat Bosshart as CTO [Lightspeed Venture Partners].
In 2017, the company brought in Craig Barratt as President and CEO. Barratt, a former Google SVP and CEO of Atheros Communications, added heavyweight operating experience [Yahoo Finance, Feb 2017]. This blend of academic vision, silicon expertise, and enterprise-scale operating chops gave investors confidence to write checks that totaled an estimated $155 million [SVVoice].
Traction with the hyperscale tier
Barefoot didn't chase the broad enterprise market. Its early and public traction came from the most performance-sensitive, scale-intensive customers on earth.
- AT&T. The telecom giant deployed Tofino in a full production environment carrying live customer traffic on a link between Washington D.C. and San Francisco [GlobeNewswire, 2017].
- Alibaba, Baidu, Tencent. All three Chinese internet giants completed evaluations and deployed Tofino [GlobeNewswire, 2017].
- Microsoft. Barefoot worked with Microsoft to expand the capabilities of the open-source SONiC network operating system on Tofino hardware [Business Wire, 2019].
The Intel exit and the integration challenge
In June 2019, Intel acquired Barefoot Networks for an undisclosed sum [SDxCentral, 2019]. For Intel, the acquisition was a strategic move to bolster its networking portfolio. For Barefoot, it was a logical endgame, providing the vast sales channels and manufacturing scale needed to push Tofino beyond early adopters.
Intel's task became to fold Tofino's capabilities into its Ethernet switch line, now marketed as the Intel Ethernet 800 Series and Intelligent Fabric Processors.
The unit economics of flexibility
Barefoot's bet was that the upfront premium for a Tofino chip would be offset by the long-term savings in network agility and simplified operations. In the end, Barefoot Networks wasn't just selling a faster chip. It was selling an escape from predetermined roadmaps. By getting acquired by Intel, Barefoot ensured its technology would get a lasting shot at that fight, moving the industry one step closer to a network that can be rewritten as easily as the code that runs on top of it.