Three Takeaways From a16z’s “Software Programs the World”
Here are three takeaways from a16z’s podcast “Software Programs the World”, in which Marc Andreessen, Ben Horowitz, Scott Kupor, and Sonal Chokshi discuss the thesis behind a16z’s Fund V. This is a notable episode because it’s rare to have Marc Andreessen and Ben Horowitz on the stage together.
Cheaper Computing Enables Software to Program the World
Fund V’s thesis is a continuation of Fund I’s thesis, famously documented by the Wall Street Journal article “Marc Andreessen on Why Software Is Eating the World”. a16z started 7 years ago to invest in the simultaneous growth of mobile, social and cloud. Fund V focuses on 3 new trends that kickstart new businesses going forward:
- Cheaper hardware: instead of twice the performance every other year, we can have the same performance for half the cost. Existing technology can be applied to new trends, e.g., Nvidia becoming overnight leader in AI chips with its GPUs.
- All chips will be on the internet.
- Software can apply to the physical world. Entrepreneurs can now write software that applies to all cars, all money, everything that flies, etc.
VCs need to evaluate new business plans and new markets they wouldn’t have considered 7 years ago.
There will be new distributed platforms, such as Apache Spark, cryptocurrency and AI. They may be offered by new entrants or by established cloud players, such as AWS.
Even though we’ve been talking about IoT for years, it’s interesting to hear a16z is doubling down on the software side of IoT. As a consumer, the physical world certainly doesn’t feel very connected now. I can’t swallow the cost to replace every light bulb at home with a $50 Philips Hue. And do we really need refrigerators to talk to the internet? Current products seem more like experiments, similar to how companies experimented for 15 years before the iPhone showed up and smartphones took off. I think it’ll take another decade for the world to truly feel connected.
AI Will Be More Accessible and Will Apply to Everything
There are 3 advantages to large companies, such as Google and Facebook, implementing AI. These factors are democratizing, and smaller players will be able to play.
- Engineering resources: AI framework software is becoming open source (e.g., TensorFlow), i.e., free for anyone to build algorithms on top of. AI software and algorithms are also on the verge of going to the cloud. All major cloud providers will provide AI-as-a-service. This frees engineering resources to focus on the end product.
- Hardware: AI is computationally intensive. Hardware is becoming cheaper, more distributed and more accessible via the cloud. This trend is similar to how startups no longer have to run their own servers in a closet.
- Data: it takes a lot of data to train an AI algorithm. The popular belief is it’s hard to compete with Google because no one has access to datasets as vast. There have been a few rare cases of smaller companies assembling big datasets in certain markets. New generation of deep learning tries to solve problems either by using small dataset, or by using simulated data. Instead of computers learning how to drive on the street, imagine if they can learn how to drive in a simulated world, complete with pedestrians, natural disasters and anything else the real world has. a16z’s portfolio company Improbable is used as an example for simulation. Improbable creates simulated worlds using gaming tools and train AI within simulations in the cloud. Improbable’s SpatialOS operating system announcement at Slush 2016 goes further to discuss “simulations models as a service”, in which you can pay for a rain forest model and run your own simulations. Changing the real world is expensive, but simulation is cheap and scalable. This is the most interesting AI-related takeaway to me. The concept makes sense at a high level. It’s hard for me to envision exactly how much work it takes to simulate something open-ended and complex, such as the world economy. I’m excited to see some real-life examples.
It’s becoming obvious all new business ideas will likely have an AI component, which wasn’t the case even 12 months ago.
Entrepreneurs Need to Raise Prices
This is a smaller topic Marc Andreessen raised towards the end of the conversation. a16z believes companies tend to be weak at go-to-market, and charges too little. Charging higher prices means:
- Being able to afford a larger salesforce and more substantial marketing campaigns, or what Marc Andreessen calls the “too hungry to eat” problem.
- Delivering higher perceived value. Buyers take product more seriously. Product has higher engagement and stickiness.
As someone who have worked on multiple open source products and companies building on top of open source technologies, this subject is often on my mind — how do we give the product away for free and have a successful business? Is upselling a “pro” version, selling professional services, or partnerships sufficient to survive? a16z’s concerns match my experiences. More on this subject in a future essay.