Software is finishing up the appetiser, before the the main course arrives.
Happy for you all to quote me on this one 10 years from now.
If one day my VC ever gets to the levels of @a16z - this would probably be the reason.
Their founder's "software is eating the wold" classic quote underpinned their "contrarian" investment thesis at a time when the world was recovering from the dot-com crash.
They were right and the VC world today sees them as one of the legends alongside the early Silicon Valley pioneers still around, like @sequoia
Today they think the future is the metaverse. I.e. digital-2.0
What I'm saying is that all the innovation that's powered technology today. The "digital" revolution. That THING is moving from the digital realm (back) to the physical realm.
The silicon revolution gave us Venture Capital 1.0
That empowered the digital revolution: VC 2.0
The physical became digital thanks to the software revolution. And now we have digital banks, AI systems, and virtual meetings.
Now we start to see the digital going back to the physical.
Don't believe me?
Take AI.
Traditional computers allowed us to build AI models. They are energy expensive as you know and (in my view) unluckily going to keep achieving fundamental breakthroughs on a traditional (von Neumann) architecture.
But now we are starting to build AI models into the architectures of new chips.
We start to see #neuromorphic computing applications.
This week I've been reading a paper from last August about a compute-in-memory chip that's competing with digital AI systems at a fraction of the energy cost (link below).
#Quantum computing went from pure speculation to an active field of investigation with specific commercial targets.
In my deep tech focused startup community I'm seeing startups looking at physics based sensors for their potential to beat best in class AI in healthcare.
I'm seen at new kinds of antennas to change the semiconductor industry in telecoms. And light-based technologies applied to cybersecurity and computation.
This wave is not crashing away the digital world any time soon, and in fact there will be a need for new kinds of software to go with new hardware.
But one thing is for sure:
A new technology infrastructure is coming.
And the startups building it are for the large part not going to manipulate bits.
They will be manipulating atoms, crystals and polymers.
In fact, some are doing it now.
Want to take a contrarian Venture Capital bet in the 2020s?
Back these folks.
Links you can check out:
- A compute-in-memory chip based on resistive random-access memory nature.com/articles/s4158β¦
1. Follow me @francesco_srv for more of these 2. RT the tweet below to share this thread with your audience 3. Join our Deep Tech community: siliconroundabout.tech
"We have only just started to scratch the surface of what we can do with data and new algorithms" - according to @mickhalsband from Lunar Ventures
π°οΈ Space infrastructure
"Space and space-related technologies can help humans evolve over the next 20-50 years and beyond" - according to Avantika Gupta from Type One Ventures
#cybersecurity was a recurring theme throughout the 1st day in a way:
It was mentioned during the #quantumcomputing panel I chaired yesterday, where I was joined by ORCA Computing's Richard Murray , @QuantinuumQC 's Waseem Shiraz and @aegiq_tech 's Max Sich
It was of course the key theme of Rand's talk on homomorphic encryption
And it came back right at the end when Roderick Rodenburg snatched victory of the startup competition for @RosemanLabs - awarder by the 3 VC judges: Obinna Nnewuihe, @SpaceSaxena and @qualleja
What I learned from this weekβ: most startup investors forget the 1st VC principle βthe Power Law.
Especially the "outsider money".
Outsider money is growing in the industry and peaked in 2021.
Now it seems to have come down, somewhat, but it's still going strong.
Outsider capital is pretty much any investor who is not a VC or has not been a professional angel for 5 years or more. Most high net worths and Family Office investors fall into this category (but not all!)
Having money to invest + Knowing what it takes to build a business β being able to generate high returns.
The missing piece is superior access to deals.
At least having access to some of the KEY deals.
It may seem silly.
Startup founders need money. Investors have money.
To all my fellow Europeans who moan about big tech regulations all the time. Sometimes it's actually GOOD. And the US is copying it.
The US is crawling closer to passing the Journalism Competition and Protection Act: which would allow publishers to bargain with Meta and Google and force them to pay for news.
Similar legislation is being discussed in the EU and in particular in France.
But the real pioneer was Australia, where early sign of success (i.e. redistribution of some revenue back to newspapers and reversing the industry downfall) is bringing back new life to real news scouts, rather than fake-news prone internet media platforms.
Is it just me or the made up title "serial entrepreneur" mostly attracts HNWs and repels VCs?
Logic:
π
Assumption 1) unsophisticated rich people trying their hand with venture may love the idea of a "safe bet" and quick returns.
Assumption 2) VC mechanics really requires them to hunt for outlier dragon returners (that normally take 5-10 years to exit) and the 1-3X quick exits are what happens when startups fail to get THERE... but at least get SOMEWHERE
Did you know that a machine learning model trained without human labels can predict physics even when information is lacking?
I was geeking around to discover examples of self-supervised learning (i.e. not using labelled data for AI training) on recurrent neural networks.
Why?
Because if you think about it, the brain is full of feedback connections.
Training a recurrent neural net (i.e. a system with loops) is not always easy (feedback tends to blow up stuff like when microphones and speakers feed back on each other).
Especially if you want to do away with a controlled environment to train the model with.
But since biology often taught us how to build better machine learning models, you can argue there is probably potential in exploring such systems.