Let's see what's left in Alphabet that isn't Google.
There's Calico, CapitalG, Fiber, GV, Intrinsic, Isomorphic, Verily, Waymo, Wing, X. Nest and DeepMind were absorbed into Google, and many of the previous Alphabet mini-companies are already gone (Makani, etc).
My personal expectation is that Calico will eventually wind down, as their research has been pretty unexceptional and I don't think they've managed to find any real revenue sources.
Intrinsic is probably on borrowed time, looking for revenue, and if they don't get significant amounts in the next two years, will be shut down.
Waymo isn't profitable yet but it looks like the execs have decided to give it a very, very long runway, and they are the technical leader in no-driver cars. It's still at risk if they don't find a good revenue model.
Wing has no real revenue to support their costs, and drone deliveries just aren't going to be a great business, so I expect it to shut down in the next two years.
Fiber- it's hard to tell why Google has allowed this to continue for so long. The only thing I can imagine is that they are somehow using the usage data to generate revenue or ML training data.
Intrinsic is really new so they probably have a couple years runway, but pharma sales aren't a great business to be in if your only product is a protein structure predictor and binding predictor. I work in this field and I still see this as an unrealistic bet on Alphabet's part.
That leaves Verily. It's clear Alphabet has downsized it significantly and scoped the projects down significantly. They have turned away from every single early idea and pivoted to a few remaining areas where they have some strength. I work in this field and it's amazing to me they keep it around- so I expect that unless they sell it in the next year or two, it will be shut down.
CapitalG, GV, and Google's real estate ventures will probably all continue because they have generated a fair amount of revenue.
Google Cloud is part of Google, not Alphabet, which IMHO is an insult to the people who run it. It could be spun off as an all-in Alphabet company but it's so closely tied to the rest of the Google infrastructure it will never live on its own.
Truly impressive how little Alphabet has achieved.
> Fiber- it's hard to tell why Google has allowed this to continue for so long. The only thing I can imagine is that they are somehow using the usage data to generate revenue or ML training data.
I'd guess they signed long term agreements with the municipalities where they offer service and they can't find a good way out. That or they have acccess to some resource related to being a residential ISP that they want to use for other things (cable TV feeds?).
Fiber sort of fulfilled its mission of getting more people connected at high speeds; the problem is the best way to do that is to annouce a build out in 20 cities, and look busy for 18 months while the telco and cableco upgrade their existing customers to something similar. I don't think you can do that more than once though. Actually building a network isn't fiscally responsible, since they won't see many takers when the incumbents can upgrade faster than Google can overbuild and most people are fine with ~50/5 service anyway.
As someone who previously did research in MD and simulating protein kinematics and binding, may I ask what you who are in the field today think about the complement so to speak - doing synthetic structures from 3D templates? Like David Baker's lab (I haven't been keeping track, to me it seemed like he was the only one in the world going all-in on this).
I think MD is a big waste of time and resources for nearly all problems, even free energy prediction. Maybe useful for studying kinetics of sub forms, ala Pande. Baker and deep mind lead the field with many ankle biters nipping at their heels. It's al proceeding roughly as I expected when I did casp in 03-04
Waymo has a huge potential totally addressable market. The possibility to earn money on every mile driven. Hardware costs are high now, but are on a very favorable curve for cost and capability.
Potential for Wing also seems good. Any delivery under 5 pounds an under 10 miles is going to be much faster/cheaper by air.
Man it’s so frustrating to see another failure to gain revenue from x labs. I really wish I had the resources to take one of these out and find product market fit. I just don’t understand why product market fit isn’t established and revenue found before launching — these failures seem to elementary to avoid.
It just seems like they flail when they launch because they are science projects and don’t have established and verified customer pain.
I really think that one exercise could reshape the whole effort. Just work on problems that also have some element of customer traction — it doesn’t mean you can’t moonshot. But it does mean you can stay in business long enough to have a chance at a moonshot which is the really thought part usually.
> I just don’t understand why product market fit isn’t established and revenue found before launching — these failures seem to elementary to avoid.
You can ask that question of the entire tech industry.
The answer is pretty straightforward: Because being a normal financially solvent business is not good enough. A simple technical improvement is not good enough.
Pichai needs another trillion dollars added to Google's market cap, or it's not good enough.
(And VC has the same problem, all of tech is broken.)
Building technology might be actually kind of easy compared to finding product-market fit. The latter might be better described as "sales team - network - existing practices - existing monopolies - adoptation rate - product iteration speed - right region" fit.
Yeah. That’s fundamentally why we prototype and get to market fast and learn what we don’t yet know. And a perfectly cromulent outcome is, “this feels awesome but nobody wants it. Sunk cost. Shutting down.”
Google X is a failed experiment. Waymo is the only worthwhile thing they ever did and it predates X (as Chauffeur).
Just look at their "Projects" page here: https://x.company/projects/ Still touting public failures like Everyday Robots, Loon, Glass, Makani, now Mineral. Fail fast shouldn't mean fail always...
'Fail' in this context mostly means 'choose to use resources to do other things'. Alphabet has certainly learned a huge amount from all of these ventures, and that doesn't go away if they close them down. There'll be people at Alphabet looking for uses of what came out of these companies in the future. The point of Google X was to try projects that had very little chance of success but that would have been world-changing if they pulled it off. The probability of 'fail always' was always quite high.
The same lesson applies to everyone. Your startup might fail but you take the learning and experience with you, and use it to do other things in the future. When you're setting goals there should be a reasonably high probability of failing because otherwise you're just not aiming high enough.
> Alphabet has certainly learned a huge amount from all of these ventures, and that doesn't go away if they close them down.
This is a fantasy, and I know first hand.
I worked for a startup that was acquired by Google. They shut us down after a year. I am here to tell you that they "learned" absolutely nothing. The archived technology was never touched. Years later there were multiple teams doing stuff redundant to what we did. They didn't know we ever existed and their projects will probably never be released.
> The archived technology was never touched. Years later there were multiple teams doing stuff redundant to what we did. They didn't know we ever existed and their projects will probably never be released.
I've worked there for 12 years and this is 100% accurate. It's all one big director and vp level circle jerk.
> There'll be people at Alphabet looking for uses of what came out of these companies in the future
Do you have any evidence for this? Because at a certain point, the framework almost defines failure out of existence. Which is very much in line with X’s philosophy.
I think the point of moonshots is that most are expected to fail. But the technology that is developed in many cases lives in, just not necessarily in the form of a standalone company.
I see that as a good thing. I'd much rather Google pour $ billions of its profit into moonshot R&D than simply returning it to shareholders or growing its ad business.
> Google X is a failure. Waymo is the only worthwhile thing they ever did and it predates X (as Chauffeur).
A failure in what sense? It's their research arm.
It feels crazy to bash down what they're doing even if it led nowhere. I'd prefer the money goes toward funding "failed research" than sit in their bank accounts.
Do you feel the same way about NASA, CERN, scientific research in general? There are many areas that receive significant funding for decades and lead nowhere by the looks of it.
I don't feel that way about the part of NASA that awarded contracts to SpaceX, but I do feel that way about the part of NASA that wastes billions on SLS (I know it's Congress's fault).
It's about efficiency. Even for research that doesn't make revenue, you can do it efficiently or inefficiently. SLS and Google X are way down on the inefficient end of the spectrum.
What are you expectations of Google X based on? in other words, what are you measuring against? Are there similar research ventures pursuing moonshots that are doing better?
Your comment was pretty harsh. Google X has been around for what? only 14 years? The number of projects they funded and the researchers they employed during that time frame is great initiative and admirable on its own.
The criticism of SLS is beat to death. The reason it exists is to ensure a viable alternative launch vehicle to space and beyond. It’s good to have redundancy in certain things. The USG has determined space access is one of them.
Yeah it costs more. But it costs less than if you ever needed it suddenly and it didn’t exist.
The question is whether or not Google, either in its bank accounts or as X, is the best steward for funding innovation. One could argue that it isn’t, and they should’ve paid that cash out in dividends and leave their shareholders to do such work.
Right, which is why I said “pay out their cash in dividends”, which go to any shareholder proportionally.
Tech stocks are weird compared to most company stocks because many of them have never issued dividends to shareholders, which is what companies traditionally do with excess cash.
I mean, NASA and CERN have consistently delivered results over the past few decades. Research fund granting involves significant amount of evaluating the prospect of the project and how successful the PI did historically. It doesn't mean "pour money into something and never expect to get anything out of it" which is what Google X is doing.
And not everything it gives is easy to see on the bottom line.
I know of some great people (like "Best In Their Field" great) attracted to google to work on X projects. Then while they're there, they may as well spread some of their knowledge to other google projects. Or the tech developed there failed as a standalone product, but the knowledge gained folded in to another project.
Even if every "moonshot" project failed, it may have given benefits to the company as a whole. Though at some point I do wonder if the intent of google X projects is to fail as a standalone product, and just be folded into more core projects....
They criticism of x labs is not that they were insufficiently ambitious, its that they don't execute to achieve moonshots, they play the game to lose. Even if they had a good idea, they would not be able to take it to market, it was all a waste.
Is "AI-fuelled smart agriculture" a "moonshot" at this point?
I thought there were several companies selling fully working systems for camera-targeted weed/fertiliser spraying, right now? Including John Deere? And Trimble and others have things like precision GPS auto-steer for tractors for over a decade?
Seems to me Google has unlimited ad money, so people working there can make the big bucks without needing to deliver sales or profits or build a sales/dealer/repair network or set up a farm equipment manufacturing operation or anything else they don't feel like doing.
Stock buyback money goes to investors who can reallocate it to companies actually providing value to customers instead of wasting it on boondoggles.
Waymo hasn't succeeded yet. Their excessively slow scaling may yet be their downfall. And again, the project predates X. It's more like X came out of Waymo than the other way around.
A lot of projects like X were really about keeping people off the job market.
edit: or they were about finding a way to marginalize competitors for senior roles. Give them a cushy job that plays to their ego, so they don't fight you for the key positions of power in the company.
google X was basically hiring standford ingroup folks from the D school and other PHD programs and giving them a lot of funding with zero accountability.
Why do you expect better lower failure rate for *moonshots* than for startups? That's entirely unreasonable.
It's very interesting that you're focusing on failures rather than on graduated projects. Do you really expect 100% success rate for moonshots?
It might actually be hard to estimate success rate as best ideas got absorbed into Google itself. E.g. Google Brain originated at X. Wikipedia has other examples: https://en.wikipedia.org/wiki/X_Development#Others
I'm agreeing with you but going even further. X-nee-twitter is not only primarily a social media site, that AI is not part of it. Musk touching something doesn't make it a part of that X, and if he names a dozen irrelevant startups 'X' that's not enough to change anything.
This is very unfair. If you look at how competitors are doing, Waymo is already very successful. Even if they can never replace human drivers in all scenarios, they have already set new standards for automated driving.
It's as unfair as complaining that an airplane company's airplanes lose a wing when their competitors often lose both wings.
The standard isn't good enough and probably never will be, they've been restricted to super-well-mapped, mostly-good-weather cities without a lot of adverse real-world conditions they're going to encounter.
In fact, performance may not improve, it may worsen as they expand geographically, and/or encounter "new" conditions (other robot drivers, who knows?)
Humans are remarkable at coping with new conditions, machines are not.
My point is that Waymo isn't a good example of Google's success considering that after decades of development it still can't drive on the right side of the road in broad daylight on a clear day.
Which is no different to a whole host of bad human driving as well. Mistakes are inevitable in something like driving, they happen to even the best human drivers. All that matters is that over time the self driving mistakes decrease to the point they are far below the human average.
I don't see the point of picking one example of a mistake and pretending it's indicative of a failing endeavour when literally no other company is even close to what Waymo is doing right now.
Several different comments in this sub-thread call this a "mistake".
Machines don't make mistakes.
If a crane collapses, it didn't make a mistake.
If a robot car drives into oncoming traffic, it's not a mistake, it's an engineering failure, a bad design.
This wasn't a fuckup in a prototype, this was a life-critical fuckup 15-20 years into the project which is still confined to a small number of well-mapped cities with good weather.
So my point stands, major fuckups decades into a project when conditions are fairly close to ideal are not acceptable and are indicative of a failed project.
As an aside, it's also not the case that "better than human" is all that matters. The kind of fuckup also matters because people vote, and votes change regulations. If people don't like it, it won't be legal. Driving into oncoming traffic in broad daylight with 20 years of engineering progress is not something people are going to say "oh okay well sometimes my grandma does that too" to.
The car almost certainly made a reasonable decision based on the information available. They have all the technology and logic required to make decisions based on ambiguous and vague signage, and if they hadn't done this, we'd be reading articles about how Waymos would stop in situations where humans would figure out what to do.
There is a lot of press trying to "gotcha" Waymo but from what I can tell, their systems are actually quite good.
I'd guess that the giant agricultural firms have already built quite accurate models using decades of data and expertise, and the addition of fancy new AI techniques didn't offer large improvements.
As someone who worked at Google X robotics until leaving to start an open source farming robotics project very similar to Mineral (see my profile) I can say that there’s so, so much opportunity for new ideas in this field, but Google X is not very good at launching actual companies. So much of what’s done at Google X is canceled (I worked on what became Everyday Robots, now canceled), that you can’t take “project at X was canceled” as a sign of anything at all really.
There are very weird forces going on at Google X. I don’t think I’m equipped to provide a thorough analysis, but I think there’s just too much easy money and the only people they really have to please are the guys at the top, who are deca billionaires who don’t even have to spend their own money to fund this stuff. It’s all a very fun and interesting R&D adventure, and those guys really have accomplished a lot in this life so I wouldn’t write off their abilities by any means, but the arrangement is very strange by traditional R&D standards.
And it really seemed to me like the people in charge of our project were more focused on getting massive quarterly milestone bonuses than they were actually delivering something that would work.
Honestly I really can’t say exactly what is going wrong, except that Google X is a very strange place.
Thank you so much! I made some good progress tonight. Finished bringing up the system with all of our new electronics, and in the next six months we’re going to start building a small fleet of them!
> While the company had forged partnerships with some leading players in food & ag, said the publication quoting an unnamed source, it had “struggled to translate such deals into durable revenue streams.”
Looking at their video[0], it seems clear the right place for this technology is mounted to existing tractors. There are good reasons why farm machinery is designed like it is: durability, crossing rough and uneven terrain, repairability, economies of size, etc.
All these sensors and monitoring systems are standard on a modern tractor or combine. For frequent close crop monitoring between drives, hyperspectral drones will probably win using a simpler software approach.
The hardware looks great for inexpensive early learning (which was probably the goal), but uncompetitive as a real product.
There's Calico, CapitalG, Fiber, GV, Intrinsic, Isomorphic, Verily, Waymo, Wing, X. Nest and DeepMind were absorbed into Google, and many of the previous Alphabet mini-companies are already gone (Makani, etc).
My personal expectation is that Calico will eventually wind down, as their research has been pretty unexceptional and I don't think they've managed to find any real revenue sources.
Intrinsic is probably on borrowed time, looking for revenue, and if they don't get significant amounts in the next two years, will be shut down.
Waymo isn't profitable yet but it looks like the execs have decided to give it a very, very long runway, and they are the technical leader in no-driver cars. It's still at risk if they don't find a good revenue model.
Wing has no real revenue to support their costs, and drone deliveries just aren't going to be a great business, so I expect it to shut down in the next two years.
Fiber- it's hard to tell why Google has allowed this to continue for so long. The only thing I can imagine is that they are somehow using the usage data to generate revenue or ML training data.
Intrinsic is really new so they probably have a couple years runway, but pharma sales aren't a great business to be in if your only product is a protein structure predictor and binding predictor. I work in this field and I still see this as an unrealistic bet on Alphabet's part.
That leaves Verily. It's clear Alphabet has downsized it significantly and scoped the projects down significantly. They have turned away from every single early idea and pivoted to a few remaining areas where they have some strength. I work in this field and it's amazing to me they keep it around- so I expect that unless they sell it in the next year or two, it will be shut down.
CapitalG, GV, and Google's real estate ventures will probably all continue because they have generated a fair amount of revenue.
Google Cloud is part of Google, not Alphabet, which IMHO is an insult to the people who run it. It could be spun off as an all-in Alphabet company but it's so closely tied to the rest of the Google infrastructure it will never live on its own.
Truly impressive how little Alphabet has achieved.