Sam Altman: Why Hardware Could Yield the Next $10B Startups
blogs.wsj.com
blogs.wsj.com
The complexity of developing hardware is driven by the number of discrete components used. Every additional resistor is a resistor which can be fitted wrong. Every additional analogue circuit built using discrete components is a huge risk. To build cheap hardware, you avoid all that as much as possible by building PCBs which just wire up highly functional ICs.
In essence your hardware company then becomes a software company.
Sorry, that's not even close to being true.
Interesting how on HN "hardware" seems to mean "a board with chips". You know, like a Raspberri Pi. That is one type of hardware but not, I dare say, where the $10B (and maybe even $10M) startups would exist.
We are developing two hardware products right now. One is an innovative fire fighting system and the other a CNC machine. The complexity in these designs has very little to do with what chips you place on a board. For example, we expect to have to run three to six months of FEA simulations to validate aspects of the fire fighting system and then have to iterate through several design candidates over several months.
Beyond that there are regulatory issues that could up-end any design if you don't know what you are doing.
Packaging. I almost forgot packaging. Man, that can consume months of testing. You can design a beautiful product only to have it damaged during transportation due to incorrect packaging.
Oh, yeah, you also have to have good software.
Nah, harware isn't just about chips on a board.
Very beefy modular tabletop gantry; AC servos; Automatic Tool Changer; probably 3HP CAT 30 or 40 spindle; coolant; Rigid tapping; weight in the 500 to 1000 lbs range.
No, it's not your typical DIY 80-20 t-slot machine on eBay. This is intended as a serious step up from something like a Tormach without having to spend $100K+.
No ETA yet.
You will almost certainly find the mentality of people making industrial equipment (or airplanes) different than people who write software where it's just about accepted practice at this point to push products into the channel and fix them later. In a sense your customers become your beta users.
Wasn't always like this (with software). I am getting software updates from Apple it seems every few days. Back in the days of floppy disks and prior to the Internet they had to spend more Q&A time because it would be rather expensive to fix problems with buggy code.
What kind of company developes both fire fighting machines and CNC machines?
Are you employed by a prototyping company or something like that?
The only other set-up I can think of where such a combination could occur is something like Mitsubishi, very interesting!
> Are you employed by a prototyping company or something like that?
Yeah. Prototyping company immediately came to my mind as well. It could also be a private R&D institute that does projects on (e.g. EU) grants. I have a friend working in the second, and apart from doing various weird hardware for EU money, they do some batshit crazy stuff when they're bored at work. There's a guy there that routinely does things like driving around the parking lot with a pulsejet-powered warehouse trolley, drawing things on concrete with a DIY flamethrower or spending few months of free time on doing a beautiful, steampunk-style Geiger counter/night lamp.
The idea that software is eating hardware, while true on some level for some class of products, is also what leads to lots of startups with solid software backgrounds offering a beautifully designed product that suffers from cost overruns, delays, mechanical and firmware bugs (you think having to test an app across a couple of phones is a pain, try field testing your smart lock). "Move fast and break things" is very different from the mindset needed to build hardware.
Be mindful of the perspective of the authors when reading these tragedies: https://medium.com/@stevekreyos/the-rise-and-fall-of-kreyos-... https://medium.com/@Haje/how-a-half-million-dollar-kickstart...
The list is huge and highly dependent on the product being designed.
About ten years ago a company owned by a friend of mine shipped a machine worth $800,000 to Japan. The packaging/crate was not designed correctly. The machine suffered irreparable damage during transportation. No, there was no penetration of the crate. It was purely inertial loads on components that were not supported for transportation. They bent. It was probably a combination of forklifts, trucking, port container handling, etc. The machine was a total loss. The insurance company was not happy.
On shipping: these days it's pretty common to ship everything with mechanical accelerometers in the packaging that trigger on high loads so you can tell when the crate has been dropped. Crazy how often it happens.
I think really the lesson here is that you really, really, really shouldn't try to build an electronics or software product without in-house knowledge of electronics and software.
Which leads to increased replacement costs. Other fun costs to account for in the physical world: loss rate in transit, defect rate in manufacturing, inventory costs.
Because that's how they get the money. On the cloud it's easier to keep monetizing your customers, and more importantly, you can literally just point out that you have "A platform! That grows!" and get investor money. They all are software companies - not only because it's easier than hardware, but also because it's easier to get funding.
Some of the companies you'd thought of as hardware ones actually admit it. I recall hearing Estimote guys repeatedly telling that they are not a hardware company - they are a software one, that just happens to make trendy beacons.
The big problem is that IoT isn't really a hockey-stick business: the configuration of even a modest whole-home setup is so complicated that consumers can't do it themselves. You won't see the valuations that accompany software startups because user acquisition is so much harder that they will never give the growth returns that a software company can. Contrast an IoT platform that you have to buy a $50 device to use with a service like Dropbox where there is little to no barrier to becoming a customer.
Ultimately, I think IoT will enable spiffy home automation systems to come down to the price range of mere mortals. But I don't think it's the next $10B market simply because it doesn't solve any problems that most people have.
A need for A/C optimization doesn't extend to there being a need for IoT/home automation as a whole. What benefit exists for a "smart" house? Drastic energy reduction can be achieved through simple use of CFL/LED light bulbs; they don't need to be Wi-Fi connected. What else would you even want automated in your house?
EDIT: Also, my original point stands: Nest is not a hockey-stick business. They get revenue when a customer buys a product, yet their platform has to support all of the thermostats they've ever sold.
I think a lot of people (myself included) think there are lots of potential benefits to a smart house. I don't think we're there yet. Most of what exists now isn't very user friendly, and most don't have easy setups so the things that are out now are unlikely to have huge hockey stick growth.
However, that doesn't mean their are no benefits, and it doesn't mean the industry as a whole won't see large growth in the future. Google and Apple are both investing somewhat heavily in the area, along with a huge host of startups. Clearly there are people who think there are possibilities in the space.
As a small list of examples for myself: 1. nest is obvious
2. locks are key feature for me. if i can have my door unlock when i get to it with my phone or watch or whatever that is a win.
2.b. remotely unlocking things could also be nice, if i'm meeting someone and they get there just ahead of me. Alternately giving friends temporary access if they are staying for a weekend.
3. garage doors, see 2 and 2b.
4. remote control blinds (perhaps a bit more than is necessary, but i could see it being nice in my bedroom
5. lights, having them come on in the room i'm coming into with my auto-unlocking door, so i don't have to fumble with them when my hands are full of groceries.
without trying, off the top of my head i came up with 5 things i'd like to have in the next house i build. I suspect Apple and Google and the startups working the area for a few years might come up with 1 or two more that turn out to be really awesome.
1) There are real benefits with Nest since it can reduce energy consumption. Also, Nest isn't really a whole lot more expensive than other thermostats, and there's very little additional complexity.
2) There are lots of problems with locks: you're taking a $15 item, adding some wireless gear and servos to it, and adding significant installation cost and complexity (since you have to somehow power the thing). The alternative is you spend a few bucks getting keys cut to give your friends temporary access. Also there are security concerns here.
3) This one is pretty low effort (since there is already a lot of radio equipment in a garage door opener) but also pretty low benefit: 90% of the time you open the garage door you're in your car, where the opener is within easy reach. I don't think the value of this feature is very high.
4) They have these today. Less useful than you might think, and like anything mechanical, they break. Also, since blinds are typically not powered, this adds cost and complexity since you have to run electricity to your windows. These are a niche product in my mind; if you couldn't justify them 10 years ago, you won't when they're connected to an IoT platform either.
5) While cool, it's still hard to justify the additional cost and complexity. There's not really a driving need other than "oh that's cool". Until the cost comes down significantly, it's still a niche product for people who want to geek out their home. At today's prices, you're looking at $40 for a single light switch compared to $5 for a non-connected switch, plus you have to configure it to do what you want.
I'm all for building cool new products, but consumers want products that reduce cost and complexity, not ones that increase them.
I also suspect you enjoy configuring this stuff, which isn't a surprise considering this is HN :)
But I think lighting will be different. My personal bet is that for lights, the driving benefit is health. I automated my lights this fall [1] and I've been very happy with it. I have a substantially more regular sleep schedule, and bad sleep has a number of bad health impacts. We're now seeing more research on how screens are disruptive, too. My basic conclusion is that humans aren't really responsible light-switch users. Now instead of my controlling my lights, thelighting is controlling me. Which is how it worked for most of human history.
I don't think many people will retrofit lighting systems right away; it's painful and expensive. But I can certainly see high-end homes selling smart lighting as an add-on. $40/switch seems like a lot, but $275k vs $276k seems pretty small. I expect that plus the 5% of people crazy enough to do it themselves will be enough to keep companies like Philips producing and experimenting.
1. Nest is a lot more expensive
2. Not worried about the security. If someone wants in its just not that hard. Locks can be picked, doors can be broken down. Adding hackers to the list isn't meaningfully different.
3. I'm not talking about opening it for me, I'm talking about opening it for other people like they showed in the apple watch keynote.
4. All the things on this list can be done, and obviously they cost more money. For the most part, these are luxury things. Whether they are "worth" it depends on how much money you have to spend. As someone else noted, it might sound expensive, but 275k vs 276k isn't a big deal. 575k vs 580k is even less important.
5.see #4
There are clearly lots of tech products that have nothing to do with reducing cost. They exist only to delight users. With more work in this area the complexity will come down as people make better products.
The fact that you don't have any (and anyone I know of either, to be fair, apart from lights in a few places), speaks a lot about how much the market is ready for this.
Why do I need lighting controls dependent on a cloud service?
No its not. Most components are practically free #1, placing them is also very cheap #2. Hardware is not expensive because of _hardware_ costs. Its expensive because of engineering, prototyping, and latency of development.
/1 single/two digit cents. Except for specialized parts. For example hiend precision resistors, supercaps, weird connectors, non commodity microcontrollers, fpgas, fast ADCs, or precision opamps/voltage references. Nevertheless you will largely use jelly bean components with only few expensive parts on the board.
/2 again except for the very hiend small footprint ones. Super small components targeting lightweight products (cellphones, ultra light laptops) require professional millions of dollars machines and experience. Here is a story of one company almost tanking because CEO is a Jobs wannabe and decided to make his product "sexy" using same components and techniques as Macbooks:
http://www.theamphour.com/237-an-interview-with-joe-and-mark...
Then came Dropcam with sexy packaging, a simple price, and ... well-made software with a subscription model. Next, they're snapped up by Google for $555 million. Would they have been acquired if they only did the hardware and nothing else?
The point is that the tools for prototyping, simulating, assembly, and manufacturing are becoming much cheaper. Along with the fact that people are sharing more.
I agree that there is no substitute for RTFM. That will always be required. Although, it's layers of abstraction in hardware too. Just like software libraries and frameworks: if someone has published a circuit that includes some subsystem I need, I can very likely ctl+v much of it into my own design without having a full understanding of it. Which is the entire reason why an industry around ICs can exist at all. It's just that more people are sharing now than ever before.
Pick and place machines are now becoming accessible at low cost (see Fire Pick Delta/LitePlacer). This addresses many of your assembly errors and mitigates costs.
I can't construct a reasonable argument against your points E, F, and G, which seem to all equate to "something unlucky and magical happened that broke your prototype".
If it's a hot hardware area (past hot areas: graphics chips, microprocessors, WiFi chips, OC-768 networking chips, ...), nobody will wait for the FPGA prototype. You get funding, you go to VLSI and maybe you produce an FPGA board for validation along the way. Probably not since what you're doing is probably too complex for a single FPGA anyway.
Otherwise, your volume is too small and the FPGA IS your product.
And that's only if you can't somehow make it run on a microprocessor.
The only place I've ever seen an FPGA validation board was with RF startups in the cellular space. And, even then, it was way back when the protocols were much simpler.
tl;dr If your demo doesn't run on a microprocessor (or SoC), it doesn't exist.
I agree that open-source hardware is not nearly as easy as open-source software. But I think it's definitely going to be both more open and less expensive.
Who, in their right mind, would invest in hardware? NRE kills--especially in mechanical engineering which actually seems to be regressing. Machining companies now charge a fortune for things which my grandfather used to do by hand in his garage in about an hour. Not to mention how long it takes to get an injection mold (14 weeks is not uncommon).
For the price of a couple injection molds, I can flog a gaggle of stupid 20-somethings living in squalor for about a year. No contest as to which has a better probability of cashing out.
In addition, who is going to buy out a hardware startup? Hardware really doesn't fit the Yahooglezoft strategy, so now your exits are very limited. Even worse, hardware acquirers tend to be rational about things like costs and profit.
Finally, everybody who does hardware tries to create the cloud back end and then get anybody else to do the hardware while they extract the rent.
So, longer timelines to cash out with higher risk. Yeah, don't think so.
Now, the upside is that if you do manage to get a product out, you have an actual barrier to entry. 2 guys and a dog in the Ukraine can't just clone you.
While I love the manufacturing incubators, they don't solve the fundamental problem of not knowing how to run machines. The existence of a CNC machine or a TIG welder doesn't magically make you a machinist or a welder. Being able to run a CNC machine doesn't mean you know how to make an injection mold.
Most people fail horribly simply at making a PCB board, and those are totally straightforward.
Read Andrew "bunnie" Huang's blog for a taste of all the little sh*t that can go wrong in manufacturing: http://www.bunniestudios.com/
I heard a quote from someone discussing war.
When a war is about to break out 1) amateurs start talking about strategy. 2) Professional start talking about logistics.
Google can be a milti billion firm because they can go from 1 to 10 (in customer #, capacity) easily. With manufacturing, that's just physically impossible.
And physics doesn't lie.
[0] https://news.ycombinator.com/item?id=9238839 [1] http://blog.semiconductors.org/blog/what-end-use-application...
What about MRR, YRR, profits, number of employees, revenue/profit per employee?
The only ones profiting from high valuations are VCs' themselves. Therefor one should be really skeptical when reading such articles.
When there is a high valuation (and ultimately market cap) the employees who own shares do well too.
When trying to express the long-term potential of a business, a factor of 2 is noise. You're lucky if you can get the number of digits right.
MRR/YRR measures the business today, but the long-term prospects are more interesting.
Number of employees or revenue/profit per employee don't measure business potential. Some huge businesses (Wal*mart) have low revenue/employee. Some small businesses (specialty consulting) have high revenue/employee.
[1] Fitbit did movement tracking well, and now there's Microsoft Band, Nike Fuelband, etc. Pebble popularised smartwatches and now there's Moto 360 and Apple Watch. And so on. And Fitbit and Pebble weren't the first movers in their markets, there'll be smaller companies who failed by being too early.
This is becoming more and more common model in many industries, and is a much harder business to copy.
The software-defined radio/network/storage etc are happening quickly now.
For the low-end hardware, e.g. the IoT field, the wearable market etc, the software is also key, for instance, they need the cloud platform to be effective and useful these days.
We just need more workshop/factories to turn the (low quantity) hardware-design into real products with high quality, which is still very challenging and a bottleneck.
There are sooo many tools, let alone basic productivity apps that just don't exist yet. And here we are, breathless and panting, moving onto the next big thing - hardware, biotech... Hey its all good stuff. Except for the part where I can't create a simple DB-like table on my phone and have that sync (securely!) with my other devices - without making me create a Dropbox account, or wanting to up-sell me in the app, or get all social on me. Basic, simple and honest apps. Like the Code Vault app I had on my Nokia Communicator in 2001.
I think you have to anticipate a gradual movement towards total programabiltiy of reality.
Solving a problem with hardware either solves a problem or replaces the thing that fixed the problem before. Solving a problem with software is usually (if not always) making computers better or easier to use. The underlying difference is how clear the change is in the user's life.
The much larger cost to develop and produce hardware like some in the comments have been discussing leads us to value hardware much more significantly than software. It might be due to iTunes, Java, and Windows update making people mad when they waste their time, or have their computer shut down on them in the middle of their work, but there is a significant bias.
My point was: if $50M is your maximum potential - if everything goes perfect - it'd be pretty hard to get some interest from VCs.
For their math to work out, they need at least the potential for an exit a few orders of magnitude more than their investment. If VC rounds start at $1M, a max potential of $50M will not be taken into consideration so easily ?