with Robotaxis, Waymo's lack of scaling aggression has been an ongoing puzzle.

Yes their solution has trouble expanding into new geographies, but why not scale assets aggressively in the geographies they have mastered.

The California CPUC dataset offers some clues. When you have very expensive hardware, lack of utilization kills your economics.

A $150,000 robotaxi that can only generate revenue 25% of the time has roughly $1 per mile in depreciation expense alone.

Waymo seems to do materially worse.

The AI-driven collapse in the cost to write software should simultaneously

-cause software demand to explode and
-radically change value chain structure

👇

How we think about demand:
AI promises radical productivity improvements for knowledge workers
At 4.5x productivity improvement against $30 trillion in 2030 knowledge work wages with 10% value capture into the tech stack suggests $13 trillion in 2030 AI software demand

people ask why google isn't shipping
isn't it obvious?
they are caught in the thorns of the innovator's dilemma how does google make money?
they charge a toll to deliver a user from the text prompt to another website.
language models take the information in *all* other websites, compress it, and deliver it directly to the user.
No transit required.
And no toll paid.

We open-sourced our 2027 Tesla model.

Expected value: $2000 per share.

ark-invest.com/articles/valua…

But cynics wonder:
1) What if robotaxi never happens?
2) What if production scaling is arbitrarily constrained?
3) What if Tesla has to keep pricing aggressively to move units?

🧵 The value of an open source model is that you can reasonably disagree with our assumptions and generate your own expected value.

People ask what happens to our investment case if robotaxi capability never deploys.

The short answer: expected value of $900 per share.

We published our annual Big Ideas report this week

ark-invest.com/big-ideas-2023/
 
This has been a year marked by convergence
 
Convergences that should yield unprecedented value accrual
 
We believe that disruptive innovation should command more than half of total market cap by 2030 We focus on 5 innovation platforms that all follow steep cost declines, cut across sectors and are themselves platforms of innovation.
 
Fascinating how technologies are not only cutting across sectors, but also serving as catalysts for each other
 
(fly)wheels within (fly)wheels

chatGPT at >10 million daily users in 40 days

Instagram took 355 days to get to 10 million registered users Note 10 million daily uniques implies 20+ million monthly

And I’m actually shocked it’s *only* 20 million and wonder if sample bias in these data under-represent penetration.

From a variety of people in a variety of industries: "chatGPT has really been useful for writing my grant/funding proposal"

I suspect that application volumes are going to spike and funding entities are going to be massively overwhelmed the adoption rate of disruptive technology is in part governed by its degree of backward compatibility with existing infrastructure and processes

that chatGPT is particularly useful for backsolving against laborious text-heavy processes is part of what's driving its rapid uptake

some real time reactions to the Tesla semi live stream
👇 The tri-motor configuration is clever.
@skorusARK modeled this in our estimates for the vehicle
The point is to mix and match power to try to keep every motor operating at most efficient rpm

Rather than contract graphic design a book publisher decides to generate cover art by simply feeding the first line and title of each book into an AI model

(In this case, #midjourney, which its developers expect to improve ~1000x over the next 6 months)

A few results

👇🏻 “As Gregor Samsa awoke one morning from uneasy dreams he found himself transformed in his bed into a gigantic insect.”

Metamorphosis

#midjourney

the tradeoff will look something like:

California
1,500 square feet

or

Montana
3,000 square feet + 3 acres, a cybertruck, starlink, battery backup, a bitcoin miner & solar The sacrifices required to go off-the-grid are evaporating: individuals can increasingly preserve their exit option while remaining a node

Technology pushes in two directions: it elevates the power law peak, and it extends the long tail.

apparent to me that on-chain transparency caused a quicker and more severe de-leveraging in crypto

in the non-crypto leverage cycle, the market is still groping blindly; we await the information that might pinpoint where the skeletons lie. (on-chain evidence of financial distress catalyzed customer withdraws which forced off-chain leverage into the open)

cross-sector technology improvements will be misread as inflation
as CPI quality-adjustments drift from the actual quality adjustments underlying production output Take new cars:
CPI quality-adjustments are made based on the manufacturing cost increment of adding new features;
a car that costs $1,000 more because a manufacturer invested $1,000 (plus allowable margin) into better airbags is assessed to cost the same amount on a real basis

you are distracted by noise

over the past 2 months,
the estimated time to Artificial General Intelligence
has fallen by between 60 and 70%

(find me a chart that is more macro-economically meaningful than this one; I’ll wait) You could reasonably conclude that these revisions in estimated time to AI imply a change in the net present value of real GDP of ~100x (at a 10% discount rate.)

throw a billion dollars at game development:

Odds that you yield a fun well-balanced game that persists?

If that billion dollars were crowdfunded does it meaningfully raise the odds of success?

What if that billion dollars were NFT-funded? Imagine that the game has 100,000 usable/tradeable objects. Further imagine that the utility value of the objects has to trend towards the entertainment-monetization rate of the experience. For US internet experiences this is ~$.25 per user hour.

love this data in Tesla’s impact report
Tesla is the only automaker in the world that is financial-incentive-aligned with delivering ongoing customer-safety improvements
it shows suspect that there is no other automaker in the world with the wherewithal (or financial incentive) to collect this sort of data and use it to develop preemptive safety response systems

everyone else is just trying to minimally clear the regulatory hurdle

Today we published our Big Ideas 2022 report, available here:

ark-bigideas.com/2022/en/pages/…

Some highlights 🧵 We forecast that the cost to train a neural network will halve every 9 months through 2030.

Roughly half of the anticipated cost declines will be due to compute hardware innovation and half due to software and model architecture improvements.

People say “follow the developers”

To win the money contract compete on characteristics that matter for the developers that matter.

For money, developers with ~♾ duration guide the most resource and so are most important; these are: sovereigns, legal systems, and institutions

https://twitter.com/yassineark/status/1475596881589936130

The introduction of the Bitcoin protocol has now catalyzed 3 revolutions:
A revolution in money;
A revolution in finance (defi); and
A revolution in media (web3)

Their economic importance is presented in descending order

The hosting platforms (marketplaces, game engines and virtual world sandboxes) become massive centralization vectors in the NFTivere
🧵

https://twitter.com/wintonARK/status/1430730151331991553

People want to buy .jpgs in order to social status signal.
They will naturally congregate on platforms that protect their status.
If the non-true owners of an avatar can similarly wear it on-platform, why buy an avatar at all (and what’s the fun in wearing it)?

/2

Continued thoughts

Why should Tesla’s rate of improvement in planning and control should exceed its competitors

🧵

https://twitter.com/wintonARK/status/1429904355931541521

One reason why Tesla's planning module should improve at a higher velocity than that of its competitors is due to an undeniable advantage in error-finding.

Tesla can measure simulated intervention rates on its human-driven fleet by running its software in shadow mode.

/1

Perhaps useful (at this point) to compare to Tesla Vision to HD-mapping strategies where an understanding of the road layout and intersection shapes, stop-signs and traffic flow lane-mappings are held in-memory.

🧵

https://twitter.com/wintonark/status/1428828394150502403

One can simultaneously understand why a Waymo or Cruise would pursue the HD-mapping strategy, and the vulnerabilities of such a rigid approach.

/1

some thoughts on tesla AI day



starting with Tesla Vision

🧵 The Tesla Vision work is extraordinary.

And vision is the right word for it.

(Massive kudos to @karpathy and team.)

/1