Over the past month I have focused on long term opportunities in Infrastructure Investment. The areas of interest to me outlined here are an invitation for founder-conversations for a venture adventure.
A good qualifier I use is “Is this a 10 year market?” – putting this simple question at the beginning of a discussion and gently raising it during founder pitches for their vision has served to separate chaff from meaningful kernels.
Another good filter is ‘enterprise‘ as the end-customer. Even if a startup is selling to intermediaries like service providers, at some point, most infrastructure “solutions” are consumed by, and hence defined-by enterprise customers.
Following this criteria, I find the following four areas highly interesting:
- Cloud & Control
- Custom Hardware
- Data Intelligence
We’re now definitively entering the era of smartphones, tablets, and other screens enabling mobility for just about every experience. Historically treated via ‘gateways’ or addressed by merely reformatting pixels for the mobile screen, mobility is an aspect that is as much about context as it is about the specific features/information selected for display/interaction on a mobile screen.
In a world of touch interfaces, ‘swipe’ gestures, and easy eye-tracking for display control, the nature of information presentation and interaction changes in a fundamental way. Database, partitioning, load-balancing, caching, front-end serving choices, etc are all up for optimization once you factor in mobility. Unless addressed at every layer of the solution, mobility will break more applications than it enhances and deserves to be treated as a fundamental factor in engineering infrastructure for any service/application.
Thinking further, advances in mobile frameworks like Musubi (Mobisocial Group, Stanford) permit serverless app development by leveraging the smarts in any smartphone while honoring user privacy for all personal/social data. Efforts like these have the potential to reinvent mobility for the next decade.
Cloud & Control
I believe the ‘control’ layer for enterprise, data-center, and service provider infrastructure (private, public, hybrid) is up for grabs for the first time in infrastructure history. None of the large technology vendors: Cisco, EMC/Vmware, Citrix, HP, IBM,… have a defensible position or a compelling solution here. More than merely an OS (open) or resource-aware hypervisor, this layer needs to address policy and compliance as much as it needs to deliver service supervision, quality at the right cost and performance across sub-clouds.
As enterprises begin to consume a rapidly increasing set of software services hosted elsewhere, these services will be required to exchange information between themselves. This exchange of information also needs to occur with compliance and following customer policies. At some point, a TIBCO of cloud(s) also needs to exist. The control layer I envision needs to interact with all the constituent parts of a cloud based, cloud-delivered set of services. Nicira and other’s vision of SDNs are early pointers to the kind of protocol work yet to be done to support service-aware traffic engineering goals over data center and core networks.
In physical infrastructure, at some point, a simple and stupid-fast optical layer in the service of flexible IP networks will eventually happen without jurassic interface definitions like OTN. Physical layer innovations that lead to simplicity will scale – and win.
In the end, the cloud is less about technology – it is about providing the right business solution. Yes, innovative (and open) technology is required but in the end, technology is the easiest part of the stack.
Custom Hardware (Workload optimized Computing)
The most meaningful advances in computing, networking, and storage have come from consumer focused webscale companies in the past decade: Yahoo, Google, Amazon, Facebook, Twitter. Most of these innovations will have applications directly in the enterprise world over the next ten years.
Innovations like the low latency, throughput optimized hi1.4xlarge instance announced by AWS and benchmarked by Netflix in July 2012 are now available on demand without spending a dime on custom engineering, maintenance, or revisions. The software (application) layer needs to do precisely nothing to leverage this hardware innovation – it just runs (faster)!
This is the eventual promise of cloud – ideally it should know no bounds and permit a variety of specialized hardware to be utilized without requiring a change in *any* software. We are approaching the end of commodity hardware machines and entering the era of custom (re)configurable hardware subservient to software.
Data Intelligence (DI)
Moving beyond the first generation of large scale data processing tools (GFS, MapReduce, Hadoop, Pig, Hive,…) we’re now entering phase-2 of wringing intelligent insights from Big Data interactively, and in near real time.
BigQuery (Dremel) at Google, Cassovary at Twitter are examples of easy to use (for programmers) services that scale with ease from tens to thousands of servers and return instantaneous results across diverse data sets. The old/conventional paradigm of batch-processing required Business Analysts, the eventual users of these tools, to work with programmers/IT or learn frameworks like SQL, SAS, R,.. in order to run, modify, and redefine queries of interest. While it won’t turn mere mortal Business Analysts in to data scientists, it meaningfully reduces the ‘process’ friction and conventional-IT cooperation required in order to get the desired intelligence out of data. Suitable only for developers today given their data models and querying languages, the interactive nature of these tools will make it attractive for startups to build data visualization, manipulation, analytics, and reporting tools on top of these magnificent data processing frameworks. The resultant Agility in business processes directly and positively enables business model innovation.
Thank you for reading this far. Some working notes for this post are here.