Hacker News Reader

Comment by tech-historian | Hacker News Reader

This is an interesting purchase for Amazon on a number of levels. Obviously they have a huge delivery fleet that could benefit from self driving vehicles. And a giant company like Amazon must love an opportunity with an enormous addressable market such as this one.

But the timing is interesting too IMHO. Amazon played the long game a little bit here and waited until self-driving technology was in the "trough of disillusionment" [1] to pounce on a promising startup. It saved itself billions of dollars in the process, let some of the pretenders fade away, and gave itself better odds of success.

Self-driving tech is still a slog though, and success is not guaranteed by any stretch. But I give them kudos for this.

[1] https://www.gartner.com/en/research/methodologies/gartner-hy...

View on HN

Considering how far behind they are from competitors like Google, I wouldn't exactly call it a winning strategy. And they are going to spend billions on the acquisition anyways, regardless of how good or bad the technology is. To me this move seems more like desperation than brilliance on Amazon's part.

> how far behind they are from competitors like Google

Zoox run their tests in San Francisco, a city known for chaotic traffic conditions. They do so rather successfully (see their YouTube videos). This accomplishment alone means they are not really "far behind".

>They do so rather successfully (see their YouTube videos).

Well I wouldn't expect them to post Car Crash Compilations of their own making.

I live in San Francisco in a particular neighborhood where there is a large amount of self driving cars being tested. I assure you, none are ready for prime time. I had one almost drive directly into me last week but I saw the operator take control of the wheel urgently.

Doesn't SF have wide, neatly planned streets and great weather all year? I can't imagine these cars driving on some narrow European roads from the medieval times in the winter

No. and No. Unless you include fog and rain as great weather.

San Francisco is one of the oldest cities in California, as such, the city layout is largely from the pre-automobile days.

https://www.google.com/maps/place/San+Francisco,+CA,+USA/@37...

A look at google maps shows the city is splits in large perfect squares of roads themselves split in smaller squares. It's typical US modern layout, the antithesis of European layout.

Most of the US is definitely very regular, the exception being Boston. There was an interesting article about city street orientations.

https://geoffboeing.com/2018/07/comparing-city-street-orient...

I would consider it somewhere in between the grid of Manhattan and the winding chaos of Boston. There are definitely fewer weather events. I don't any single self tech could successfully navigate snow.

Chaotic??? You haven't traveled much I guess.

I'd love to see a self driving car attempt India or Viet Nam.

It would hilarious to see the first cases of PTSD in AI.

I think far behind is in reference to Amazon.

In raw tech for "self-driving car" Amazon does not have the expertise or personnel, but they have wider transferable experience in cultivating the R&D of such a project what with their dabbling with SLAM-driven drones and their work in robotics.

So they're not starting from zero + a speculative acquihire if that makes sense, at least in my opinion

Disagree. The future of self driving is not in self driving cars, but rather in self driving deliveries.

https://techcrunch.com/2020/04/07/nuro-gets-ok-to-test-its-d...

Zoox is in prime position to be a better nuro

I don't know about the last mile in deliveries, but self-driving logistics seems like a sure bet.

Interesting, rather than 1 employee driving 1 van, the same employee could command/oversee N number of self-driving deliveries. Self-sanitizing machine would also ensure "contact-less" delivery.

Certainly nothing brilliant about the acquisition. I'm with you there. It's just an M&A deal. But considering the (relatively) cheap price, there's a chance it eventually pays off to be a transformative catalyst for the business.

Amazon's market cap is currently $1.2 trillion. If they spent $2.5B to acquire zoox, that is only 0.2% of market cap. Not a bad risk/reward ratio.

> Not a bad risk/reward ratio.

Hard to tell since you haven’t defined any particular reward. Keep in mind the risk/reward ratio for a particular investment would be the same regardless of the percentage it made up of the larger portfolio.

Self driving software will be winner takes all, or a maybe two companies will sell software. Pretty easy to see 20M units/year at $5000 each, so call it 100B in revenue per year with profit margins of 50%. The winners will likely see close to a trillion increased valuation. There is a natural monopoly for the first successful movers.

As a side note, risk/reward ratio is the same, but when making bets with positive expected return, you must consider bankroll size and risk of ruin. The VCs are realizing they can’t play a $300M/year game where the probability of winning is 20%, even though the expected return is 100x.

Why is it winner take all?

First mover advantage snowballs: real-life client deployments increase the miles driven hundredfold and the revenue increases the resources devoted to mapping and to fix rare situations with manual fixes to either maps or AI.

The behemoth company with custom mapping of every road in your country and manual remapping of construction sites will provide a far better experience than the scrappy AI start-up. (And they can probably leverage that to gain better insurance rates, or to get driving licences in places where second movers can't.)

This doesn't mean that self driving is a winner take all market.

It does mean that buying the alternative will usually be a "crappy" choice. Leaving the incumbent with the ability to charge a margin and hire better devs. There will be other players who are competitive locally, or provide "good enough" solutions for some customers.

* Assuming any player finds a viable business model, and that our current understanding of ML, Computer Vision, and control systems is sufficient to produce a commercially viable product.

Why? I am sure self-driving will have regulations around it for specific performance guarantees.

I think that's where "good enough" comes in. It's likely that some companies will be just compliant, whereas the market leader is well ahead of compliance. e.g. a car that safely comes to a stop in an unhandled situation vs. a car that can handle the situation. It's also possible that the space also reaches commodity status where no differentiators emerge ( as one would expect with a big S curve in performance ).

Then Tesla already won because they have the most miles/data? Doesn’t seem sound.

Will a good demo of self-driving increase Amazon's stock by more than 0.2% ?

That's enough to justify the purchase.

"transformative catalyst for the business" is the reward, for a tiny risk of 0.2% market cap.

> Considering how far behind they are from competitors like Google

Why do you say that?

Yeah... except this is more complex than Gartner's hype cycle captures. This isn't k8s or openstack.

There are real technical challenges to making this work, not that it doesn't magically solve all things - it can't solve anything yet. The long tail still appears quite long.

Agreed, the tech will take time to become viable. But Bezos is patient, is he not?

I mean, he founded Blue Origin in 2000 and it finally emerged from its self-imposed silence in 2015.

The hard thing to do, which Bezos does very well, is to stay focused, patient and stay invested. Now it’s even easier to be even more patient as amazon has a very deep pocket.

Well, I mean, back in 2006, Bezos' company was saying that they could be launching into space as early as 2010, so I don't think it was self imposed silence for that long. (Source: https://www.wsj.com/articles/SB116312683235519444)

It seems to me that we don't need to improve the tech at all. We just need some pragmatic infrastructure to support the existing tech.

There are lots of tech improvements, but many of them are public, so even just waiting helps.

Using EfficientNet instead of ResNet, depth detection algorithms, better object classifiers are available than just a few years ago.

I believe that there's less than 5 years between the first company achieving self driving and many other companies.

AI algorithmic progress is outpacing Moore’s Law at it best.

Compared to 2012, it now takes 44 times less compute to train a neural network to the level of AlexNet.

How that affect self-driven car is still to be determine, but progress is very real

https://openai.com/blog/ai-and-efficiency/

This is only in terms of effeciency, not in terms of accuracy of overall performance.

Of course, but it's a great example - there is certainly a lot of progress in the other dimensions as well.

However this gain also directly translates to a real-world result improvement: You can now train a lot more at the same cost.

The government budget crisis that the pandemic will cause as tax receipts decrease and outflows increase might be a catalyst for this.

I could see a big company or conglomerate of companies being willing to take over part or all of certain parts of road development and maintenance as long as they can have one or more dedicated autonomous lanes.

I imagine that there will be a lot of room for creative solutions over the next 10 years.

That’s frankly dystopian. It’s much worse than deregulation; it’s the dissolution of a uniform state. A return to the Middle Ages, its warlords and its robber barons.

> That’s frankly dystopian. It’s much worse than deregulation; it’s the dissolution of a uniform state. A return to the Middle Ages, it’s warlords and it’s robber barons.

You just described my best guess of the future (at least of the US).

In fact, you might be able to say that we are already moving that direction given how much influence large corporations and industries have over our government.

Note that I am not happy about my prediction, but my eyes are open to what is happening around me.

Doubtful. Taxes ain’t going away.

> Doubtful. Taxes ain’t going away.

Which part is doubtful?

I didn’t suggest that taxes would go away.

I suggested that tax receipts (the amount received by government) will decrease due to reduced economic activity during the pandemic. This is a fact (it has already happened).

This would give the tech companies an opportunity to solve some governmental budget shortfalls by relieving some of the burden in exchange for some access rights.

Sadly Detroit's Big 3 announced that they believe electric cars are the next big thing so they're cutting back on research for autonomous cars. They're blaming it all on the pandemic. I said publicly that Tesla and Comma.ai would eat their lunch. Now I can add Amazon to the list!

https://www.detroitnews.com/story/business/autos/mobility/20...

There are numerous Tier 1s (Bosch, Continental, ZF) working on providing autopilot like tech to auto manufacturers. Additionally companies have there own efforts (GM super Cruise for example which btw has nothing to do with Cruise and is developed separately). Comma has open sourced all there code, so good chance the OEMs have someone looking at it.

Apart from GM's supercruise which is only avaialable on a top end Cadillac CT6, there is nothing that rivals OP or AP. Telluride/Pallisade are good but not close to either of above products. Not sure whats holding manufacturers back

I find it weird that they'd pose those two directions as opposites rather than as complementary tech.

Why will Zoox under Amazon eat their lunch, but not as an independent company?

Because under Amazon funding will no longer be an issue. They will be able to install devices in Amazon's rather large fleet and gain millions of miles of experience on roads all over America. It will accelerate their development process significantly.

Reply on news.ycombinator.com