The Futuriom Interview: Sudip Chakrabarti

Sudip Chakrabarti

By: R. Scott Raynovich

Sudip Chakrabarti is a partner at Lightspeed Venture Partners, one of Silicon Valley’s hottest investment hands. He has an affinity with and deep knowledge of networking and cloud infrastructure companies -- as well as expertise in the game of cricket. He joined Lightspeed in 2016 as a partner focused on enterprise infrastructure software. Prior to joining Lightspeed, Chakrabarti was a partner at Andreessen Horowitz, where he worked on investments in networking and the Internet of Things (IoT), including Forward Networks, Tachyon Nexus,  Distelli, Samsara, and some stealth companies. He was also a board observer for a dozen portfolio companies including Actifio, Databricks, DigitalOcean, and Mesosphere. He started his venture career at Osage University Partners, where he invested in university spinouts such as Menlo Security, Infinio, and Skytree.

Chakrabarti has PhD (Computer Engineering) from Georgia Tech, an MBA from Wharton, and a B.Tech from IIT Kharagpur. In addition to loving cricket, he is also an accomplished soccer player. Other experiences include cofounding two technology startups, operating roles in two silicon valley software companies, and brief stints at McKinsey and Internet Capital Group, a VC firm specializing in SaaS and cloud computing investments.

Q: What opportunities are you looking at in IoT?

I focus on infrastructure. Internet of Things (IoT) kind of falls into that bucket. I've never been interested in consumer. I've been focused on industrial. In the past, I had worked on a deal called Samsara -- the idea was bringing networking to industrial IoT. 

As a VC (venture capitalist), I believe consumer IoT to be a pretty hard space to crack. You have to come up with the smartest door lock, and then at the next CES, you see 20 Chinese knockoffs. It's a whole different ball game. There's a little bit of fatigue in consumer IoT. 

Let me describe IoT at a very high level. In the industrial, there is the hardware -- sensors. I have been spending some of my time on this hardware space. All of this IoT hardware could be really interesting from a functionality standpoint if you could do custom processor design. Today, people slap together components, but the primary challenges are: It's not small enough, it's not low-power enough, and then there is the cost. So there's room to design your own custom processor. So ARM is there. The challenge is that ARM is much more expensive to use when you are designing devices that cost a few cents. There are some challenges to using ARM as your vehicle to design. 

I have been spending some time on the alternative to ARM. For example, there is a project called RISC-5. The original RISC idea was conceived at UC Berkeley. Their whole goal is to take RISC to IoT. It's a pretty interesting idea. They have open-sourced their ISA (instruction set architecture). It's more like what open-source has done for software, the RISC 5 project wants to do for hardware. There are other ways of building the custom chips. RISC 5 has a shot at it.

Q: That's interesting. What's the next part of IIoT? 

The other part is the infrastructure for IoT. The network is a big component of that. You have sensors either embedded in the industrial refrigerator, or you might have standalone sensors. But how do you connect a fleet of sensors? With a very intuitive interface without exposing the messiness of networking. There is this whole notion of reliable, seamless network. Samsara is that story. The team was the former Meraki. They understand networking very well. 

Once you have the networking, there is this whole management layer on top of it. How do you provide a cloud based management layer on top of it? The IoT things are fairly dumb. They are mostly data collectors. There has to be a cloud-managed back end. There are a few companies trying to be more. 

Then you have the analytics part of it. There are a gazillion analytics companies. There are so many companies out there. They are all trying to provide analytics. It's not a whitespace, but at the same time I haven't seen companies cracking it or coming out as the leader. It's still a space that can be interesting.

You can collect the data and send it to the cloud level. Or you can do a chunk of processing at the edge. Sending it back requires power. More importantly, latency is an issue, in self-driving cars, for example. There are emerging technologies to process and manage the data. 

Let's say I'm a developer for the Internet of things. I make a distinction between dumb sensors and smart sensors. I need the whole infrastructure deploying on the edge devices, managing the devices -- monitoring, troubleshooting, and monetizing. is a company like that. There are a few others. Lots happening at an open-source project level.

Q: Security will be a huge issue, no? 

Security is a big part of it. What I'm trying to say is there is this dumb sensor and you need an infrastructure stack. Data, networking, analytics, and then in parallel you need this other stack for much smarter things. It's basically apps instead of dumb data collection. In both cases, security is a big thing. 

We have been looking at IoT security, but it's a very hard problem. It's not easy to secure a large number of devices when they're really dumb. 

Q: What does datacenter infrastructure look like for a whole IoT infrastructure?

Given the volume, velocity, and variety of the data that IoT devices are expected to send, I am not sure whether existing datacenter architectures will be able to keep up. 

Maybe the datacenter will look somewhat different, maybe you do more stuff on the edge. Think of a gateway connected to sensors: You do some computing on the sensors, some on the gateways, and some on the cloud. I don't think it's that much of a stretch to think of datacenters going that way. The next-gen datacenter will need to support several orders of magnitude more devices and data, securely and with low latency and great user experience. There's a real reason for the infrastructure to look different. 

From a technology point, one big open question is monetization. In the server world, or the mobile world, monetization was a lot different because the end devices were more expensive. In the IoT world, the devices cost a few cents. How much is the customer going to pay for infrastructure running on such inexpensive devices? Unless you have a huge number of devices, you can't make much money by charging purely on a per-devices basis.

When IoT companies start, they are only managing a few hundred devices, so the monetization looks pretty anemic. 

Q: But there are still many use cases for IIoT that have been making a good business case. 

Yes, but the use cases themselves are highly fragmented. That could be an issue. Think about it, things like temperature and moisture sensing. Are any of those use cases huge? I just don't know. There are bunch of areas within (industrial) IoT, each of which is probably a couple-of-hundred-million dollar market, but not a billion dollar market. The winner will likely be able to build a solution that addresses multiple use cases and hence take on a large enough market.

Q: What do you think of Amazon and Microsoft and what they are doing in IoT with their cloud services?

I was at AWS:Reinvent last November when they announced their IoT platform. They are very well placed. There is this notion of serverless computing. Amazon calls it the Lambda service. You can run your code as a function. If I can run a service without taking any headaches about which servers to run on, how to scale, etc., then that's a powerful concept. 

Why is it important? A lot of the IoT use cases tend to be event driven. The serverless computing paradigm is perfect for that. One of the challenges that Amazon has is the lock-in. With Microsoft and Google catching up, can you take your serverless computing across clouds.

Think of it this way, in the ideal world. Let's say you have a bunch of IoTs. The devices collect some data and need to process that data somewhere through a service. Think of it as a microservice. That service might be running anywhere. The device needs to trigger that service and get that service to react to the data. The whole infrastructure has to do that, and if you depend on virtual machines, you are toast. You cannot react at that speed, efficiency, and -- ultimately -- cost.

Q: What are some other interesting approaches you have seen in IIoT? 

There are companies that are working on stream processing. Why is that important to IoT? Analytics is a big part of all of this. All of the big data techniques that have been created have been based on batch processing. Hadoop and Spark -- batch processing engines. Spark is a very fast batch processing system, but it's micro batches.

To handle the data and latency for IoT, you need a true stream processing engine. You need a true stream processor. Google, Twitter, and LinkedIn have built similar engines internally. They needed a stream processing engine. For example, Twitter built Heron. IoT will definitely need that stream processing. There's that angle, too. 

For a more detailed, professional analysis of the IIoT market, purchase our 50-page Ultimate Industrial Internet of Things (IIoT) Report, which covers a wide range of communications and cloud technologies that are being applied to businesses around the world to provide connectivity, analysis, automation, and optimization of a range of industrial applications.