20VC Newsletter - May 5th 2024
Here are the transcripts and top takeaways from 20VC episodes released this last week!
Monday’s episode with Arthur Mensch, Co-Founder & CEO at Mistral:
Download the full transcript here:
My 7 key takeaways:
Biggest Takeaway from DeepMind
A team of 5 is faster than a team of 50.
Unless you organize the 50 into 10 uncoupled teams of 5.
I learned it the hard way & built smaller teams at Mistral.
How Investors Should Be Thinking About AI Investing Today?
There will be 2 opposing directions:
The application layer will get thinner
The models will get thinner.
We believe the model will still be big enough & needs a platform for vertical applications.
How Does the Cost of Compute Come Down Over Time?
Two ways 👇
Reducing hardware cost: We get around 30% reduction every 2 years
Increasing algorithm efficiency: We’ve made 100x improvement vs. 3 years ago.
Our bet is on efficiency & we think there are more improvements to be made.
Why Did Mistral Go Closed Source
There was an opportunity to grow our business with some commercial models.
It was a good way to cement strategic relationships with cloud providers.
We want to be an open-source leader with assets we can license.
Are We Still in the Land of Experimental Enterprise Budgets?
Obvious AI applications, like customer support, are moving into core budgets.
Industries like telecom & healthcare are still experimental.
It will likely evolve next year.
Why Does Mistral Not Need to Scale Compute in the Same Way as Everyone Else?
We’re convinced we don’t need to scale as fast.
There are a lot of new barriers that aren’t compute-related.
We are leaders in model efficiency & we’re strengthening that position.
How Mistral Can Train a Silicon Valley Quality Engineer in Europe in 3-4 Months
We hire 23-24 year olds & onboard them in 4 months.
They operate just as well as any other Silicon Valley engineer.
Europe has great talent & opportunities if we convince them not to go to the US.
Wednesday’s episode with Mark Suster, GP at Upfront Ventures:
Download the full transcript here:
My 7 key takeaways:
How to Make Money as an Investor
You have to believe in something others don’t – and you have to be right.
Betting on gen AI & crypto is too late for returns now.
We’re doing things LPs aren’t hearing from other people right now.
Everyone is a Sheep When Investing
People invest most when markets are about to hit their peak.
People sell most when markets start plummeting.
I started buying in Q4 2008 and I’ve never been more stressed.
Three Criteria for Reserves Deployment
We look at 3 things:
Do I still believe in the market?
Do I still believe in the team?
Do I still believe in the valuation?
One of our best companies went from $15M to $500M valuation
But it didn’t have enough traction so we didn’t write a check.
Venture Returns Will Come Down Because PE is the Only Exit Now
IPO markets are shut.
M&A markets are shut.
Returns will come from PE & they will buy from venture funds.
BUT they will be paying reasonable prices.
60% of 1200 Unicorns were Funded by Four People; Tiger, Coatue, Insight and Softbank
1,000 of 1,200 Unicorns funded in 2021/22 won’t hit $1BN exit value.
60% of these are marked up by investors.
This creates zombie companies.
Why Now is the Best Time to Be Investing in Space
The cost of launching rockets in space has fallen by 90%.
There are more than 100 spin outs from SpaceX & over $10BN raised.
A majority of them are based in LA.
Why the Woke Left Worries Me
The far left are radicalizing against Jews.
There are only 15m Jewish people in the world.
Anti Semitism is rising in the US and we need to stop it.
Friday’s episode with Keith Rabois, Managing Director at Khosla Ventures, and Eric Glyman, Co-Founder and CEO of Ramp:
Download the full transcript here:
My 6 key takeaways:
How Ramp is the Fastest Executing Company on the Planet
Time is the most limited asset in early stage startups.
We count our days at Ramp & do a calendar audit.
This forces us to think about what creates the most leverage & prioritize those.
Micromanage or Hire and Get out of the Way?
It depends on the task relevant maturity of the person.
I use a 2x2 grid on conviction and consequences:
High Conviction, High Consequences: “Micromanage” closely.
High Conviction, Low Consequences: Guide first, then delegate.
Low Conviction, High Consequences: Delegate to experts.
Low Conviction, Low Consequences: Fully delegate.
Lesson From Brian Chesky
Most leaders should be thinking six months ahead.
Not every lever can be tuned in a week or a month.
Thinking six months ahead prepares for those levers.
Challenges for Ramp - What Could Go Wrong?
Hiring: Hiring the wrong people for the job.
Risk: Being too conservative & getting it wrong.
Prioritization: Lose focus on making our products best in class.
Lesson From Jack Dorsey
CEOs can “write” or “edit” people’s work.
They should always try and do the latter.
If you’re constantly using a lot of red lines, that’s a bad signal.
How Many Direct Reports is Too Much?
There is no one size fits all formula.
Andy Grove recommends 5-7, Jensen Huang’s doing fine with 60.
It depends on your level of depth as an executive.
Let us know what your big takeaways from this week’s shows were in the comments below!
Thank you for reading, and don’t miss the great guests we have next week:
Sarah Tavel, General Partner at Benchmark.
Tom Blomfield, GP at YC.
Thank you for reading 20VC.
This post is public so feel free to share it.