The Tech Monopolies Go Vertical

There are tools though through which we can mitigate verticals and have in the past:<p>- regulated markets (eg limiting service bundling such as free email financed by targeted ads, tying devices to social logins, predatory social network effects, mandating availability of profile data for ad targeting to third parties, stopping ad platform Ponzi schemes eg without objective third-party efficacy measurement, create strong consumer laws against entertainment devices wanting to track and send ads to you, rethink 5G networking)<p>- mandate standard formats for your data&#x2F;text providers to export at your request for you to move to another provider<p>- stop hw&#x2F;sw bundling and&#x2F;or tax closed hw&#x2F;sw platforms different from general-purpose computers (such as has been done with Playstation)<p>- net neutrality<p>- return to public sponsorship of development for standards in information processing and demand alignment to such standards in at least public tenders or beyond (such as in the construction industry)<p>- mandate E-commerce transactions be represented through standard, signed order and billing manifests (rather than allowing proprietary ad-hoc&#x2F;pseudo currencies which would also be problematic for tax authorities) or create&#x2F;mandate digital currencies<p>The challenges we&#x27;re facing as civilization due to digitalisation aren&#x27;t entirely new and have been addressed for eg the telco and financial industry. It&#x27;s only in the last decade that we&#x27;re being brainwashed to become slaves to &quot;the Cloud&quot;.

As an example of this I do wonder if the PC industry realises the bind they’re now in with the Apple and the M1.<p>We don’t know the unit cost of those parts, but we know from past statements they are a very small proportion of what Intel is charging OEMs for their processors.<p>Given that the perception has always been that MacBooks are the best built laptops available, but priced at a substantial premium, it’s been fine for a Dell or HP to coast by on lower prices and thinner margins (as long as Apple wanted 30% or so in profit on each unit sold it was easy to compete on price and still make a living).<p>With the M1, Apple now has the option of competing directly on price while preserving their margin, with a substantially faster machine, that still has the other premium elements (e.g. the aluminium body, the massive trackpad).<p>I don’t know how any one of their competitors can compete with a similarly-priced device that is faster, better built, and commands much higher residual value down the road.<p>Maybe I’m being naive but I’d be very worried right now if I were Dell&#x2F;Lenovo&#x2F;HP, especially given that Intel doesn’t have a working 5nm process.<p>It’s even worse for companies like Hynix, given the performance gains from the on-package memory architecture in the M1. AMD are going to have to go down that route now to compete, and that doesn’t leave much room for direct-to-consumer DRAM vendors (although given the questionable histories in that market it may not be the worst thing for the consumer).

I&#x27;m reminded of how retailers once they cross a certain threshold of popularity and distribution penetration begin rolling out their own branded products. Amazon Basics is an example of that but is true of almost any medium&#x2F;large retailer. I often wondered when will that happen with Apple&#x2F;Google&#x2F;Amazon. It does seem inevitable, in the hindsight.<p>Beyond a certain scale middle men&#x27;s costs become non-trivial and hard to ignore. In case of Apple, though it is more than costs alone because they want end-to-end control of UX. By the looks of it processor had become a limiting factor in achieving that control.<p>Makes me wonder, processor is the lower layer in the stack. What about upper layers? Apple have already rolled out their credit card. Maybe their own cellular network? A new WiFi standard, perhaps?

Ruminating about the possibility that mainstream general computing is now coming to an end. I guess as long as we have Unix we will still have that.<p>But it looks like for the average consumer, they wouldn&#x27;t really care as long as their needs are met, even if those machines are becoming more and more like appliances.

For some reason the article&#x27;s explanation about &quot;software ate everything&quot; and jumping between each company isn&#x27;t resonating intuitively with me, so I offer my own take on the situation:<p>For a long time, the hardware&#x2F;data center offerings of Intel, AWS, etc. were close enough to what other companies needed to not make it worth their while to invest in inventing their own solutions to low % compute problems.<p>However, as:<p>-- compute loads and costs grew<p>-- types of compute became more specialized<p>-- design and build of one&#x27;s own silicon became more accessible&#x2F;differentiable<p>-- companies&#x27; needs and tolerances for paying a premium diverged enough from what Intel&#x2F;AWS was offering,<p>it then became worthwhile for large companies (who can sustain such hardware development efforts) to design and build their own chips, either for cost reduction or functionality-enhancing reasons. Maybe they just saw the margins being achieved by sit-on-your-hands incumbents and decided, &quot;we could do this too, and better&quot;.

&gt; Everyone loves to talk about tech monopolies.<p>No monopoly loves to talk about how it is a monopoly though, most claim to have fierce competitors. Monopolies usually pretend to be part of much larger markets; where they are just &quot;a small fish in a big pond&quot;.

As a counter-point, all these companies might eventually lose the goodwill of the programmers by exercising too tight a control -- and having very specific tooling where knowledge can&#x27;t be transferred to other jobs -- and they could find themselves starved for quality hires.<p>Imagine if the average programmer needs 5 highly specialized laptops to improve their job prospects -- not many will do that even when they have the money.<p>I mean, until it gets to that point the big companies might not care anymore, sure.<p>But general-purpose computing is very far from dead or even dying. What we seem to be heading towards is more like a huge divide and a segmentation that might last decades.

I think AWS Inferentia [1], Amazon’s AI silicon, is much further along than the article suggests. The Alexa backend is mostly run on these chips now [2]. Other than this nitpick, the article is a great summary of how Google, Amazon, and Apple are vertically integrating silicon into their platforms.<p>[1] <a href="https:&#x2F;&#x2F;aws.amazon.com&#x2F;machine-learning&#x2F;inferentia&#x2F;" rel="nofollow">https:&#x2F;&#x2F;aws.amazon.com&#x2F;machine-learning&#x2F;inferentia&#x2F;</a><p>[2] <a href="https:&#x2F;&#x2F;aws.amazon.com&#x2F;blogs&#x2F;aws&#x2F;majority-of-alexa-now-running-on-faster-more-cost-effective-amazon-ec2-inf1-instances&#x2F;" rel="nofollow">https:&#x2F;&#x2F;aws.amazon.com&#x2F;blogs&#x2F;aws&#x2F;majority-of-alexa-now-runni...</a>

on the other end of the spectrum, i can definitely see a trend in DIY : 3d printing, knowledge sharing through social media, mooc and e-learning, risc-v , arduino, rasperry-pi, etc.<p>One could also see a future where computers become home-built, both software <i>and</i> hardware.

Neither of these points are relevant to the purpose of this article, however:<p>The table showing Google&#x27;s suppliers left out IBM and their POWER9-based datacenter systems. These include but are not limited to GCP offerings: <a href="https:&#x2F;&#x2F;www.datacenterknowledge.com&#x2F;google-alphabet&#x2F;google-launches-ibm-power-systems-cloud-service" rel="nofollow">https:&#x2F;&#x2F;www.datacenterknowledge.com&#x2F;google-alphabet&#x2F;google-l...</a><p>In the same vein, I must admit that I was impressed by the work these monopolies poured into the Open Compute Project. AFAIK, OCP datacenters are rare outside of these companies, but they are one of the net goods that have come out of tech monopoly vertical integration.

Hah, I really like that I was right, having written this (<a href="https:&#x2F;&#x2F;medium.com&#x2F;@turbinenreiter&#x2F;on-google-apple-and-the-computer-industrys-struggle-for-innovation-a9b4ad6e4212" rel="nofollow">https:&#x2F;&#x2F;medium.com&#x2F;@turbinenreiter&#x2F;on-google-apple-and-the-c...</a>) a while ago, making a similar point, although written way worse :D<p>I think that we (non-FANG, open-source, university) have to make sure we don&#x27;t lose the ability to innovate, due to not being able to buy the hardware we need. I.e. if we can&#x27;t buy an M1 class processor, we can&#x27;t invent new products that would be enabled by it.

My prediction on how this will play out is that it will become impossible to exceed the FAANG based on compute price, as the article also points out, but I predict that it won&#x27;t matter as much. If Amazon has a 50% cost advantage on AI by producing their own chips, then an algorithm working 2x as effective but inaccessible to Amazon would be enough to catch up.<p>So my prediction is that as chip designs get more exclusive, algorithmic research will become more important again.<p>Everyone is using interpreted languages because CPU time is still considered cheap. So if you want to prepare for this future as a coder, learn C++ or Rust.

&gt; We are going back to the old patterns of integration of both Software and Hardware.<p>Can we have some regulation against this, please?

&gt; This is a barrier to entry that few companies can really climb over anymore, with 500 million in R&amp;D only possible by a few companies<p>I wonder what could be done to reduce that barrier. It sounds as though there is a lot of value that could be unlocked by making it easier (cheaper) to get over it.<p>I imagine a company that could drop the Verification or Software costs down by an order of magnitude would suddenly find itself with a long lineup of customers. After all, if only 270 companies can currently get over that barrier, reducing the barrier by even $50m or $100m would open up the market to a large number of companies, all with fairly large pockets and a strong incentive to be customers.<p>And of course, I say this having never worked in the hardware design industry, so all are welcome to explain in excruciating detail how naïve I am.

Good read and I didn&#x27;t know about the rumors of Facebook&#x27;s ARM chips (for what though?), Microsoft&#x27;s Project Catapult, nor the imminent custom chips for Waymo.<p>I think the next frontier will likely be on network bandwidth, in particular user-facing production ones beyond the internal backbones. Currently it&#x27;s a finite resource that all network providers have to negotiate for and involves more than just the big tech companies. But as workloads continue shifting to the cloud (leveraging the cost efficiencies such as AWS Graviton chips), I&#x27;m expecting companies to vying for greater ownership of this external network bandwidth.<p>I&#x27;d be curious of any behind the scenes negotiations being made with Starlink, to leverage its growing satellite constellations to address this.

Sometimes I hear that &quot;hardware providers are holding dotcoms in a chokehold.&quot;<p>That&#x27;s utterly wrong. Silicon makers themselves are under intense pressure of dotcoms massive negotiating weights. Chasing the Facebook, Amazon, Google trio is not a pleasant experience as many OEMs found out.<p>Unlike hardware makers themselves, and OEMs, there is no way any component manufacturer can simply &quot;turn the switch off&quot; on a dotcom client. Such clients can simply wait out few hardware generations, and buy older products just fine.<p>Now, imagine yourself an OEM, and that a top chipmaker declined to ship him his latest, and greatest chip&#x2F;part. This becomes instantly less funny. Imagine a graphic card vendor having to ship cards with 3 years old GPUs, or a laptop vendor having to do the same.

With nary an antitrust enforcement in sight, who can be surprised?

&quot;No company better exhibits vertical integration than Apple.&quot;<p>Tesla.

I&#x27;m not buying the argument. Firstly I don&#x27;t think that designing competitive AI accelerators requires the resources that would restrict the market to a few high-end competitors. From what I understand, they are a sea of low precision DSPs specialized for convolution, connected by a high-speed interconnect. While I would hesitate to claim that they are simple to design, they don&#x27;t necessarily have the staggering amount of design complexity of a modern CPU or GPU.<p>Second, for CPUs I don&#x27;t think that a complete clean-sheet design makes sense for a lot of companies. Yes, I get it, Apple has taken the world by storm - but let&#x27;s not forget, that Apple gets a lot of mileage from very few designs, probably the same CPU core ends up in the iPhone, iPad, Apple TV and now the MacBook - with slight modifications. This is the same, for Samsung, ARM,h Qualcomm, Intel, etc. - most of their CPU designs can be aimed at the market for servers, game consoles, smart devices, PCs, Notebooks, etc. If AWS designs a server core, their entire market is AWS instances probably - not a small one by any means, but smaller than a dedicated CPU manufacturer&#x27;s, thus the sensible amount of investment they can make into CPU design is lower than that of a dedicated CPU manufacturer. I&#x27;m not saying it doesn&#x27;t make sense for them - it probably does - but it&#x27;s far too early to predict the demise of CPU companies.

&gt;Google - In-House Opportunity - TPU chip, Custom ARM chip for phone<p>Google will put their custom chip on their Chromebooks for sure, there were couple of Chromebooks already with some weird SoC earlier which was supposedly purpose built for Chromebooks, couldn&#x27;t remember the name I think it was from Spreadtrum.

That makes me wonder which widely adopted software will first have optimization for new generation non-standard hardware, for example Optane storage? Postgres? MongoDB? Redis? Any of competing time-series DBs?

The difference between 28, 16 and 5nm is much less than the sales people would like you to think.<p>The costs are exploding, not only are revenues going to drop because there is no reason to buy anything new, but combined with more &quot;subscription&quot; plans where you don&#x27;t own your wares and the shorter lifespan of planned obsolecence, you end up in the eye of the perfect storm.<p>Combine that with global peak energy + debt and covid is just another margin call.<p>Get out now or suffer the consequenses for the rest of eternity!<p>Linux on Raspberry 4 and Jetson Nano is the only exit and it will close sooner than you think.<p>Edit: PCIe NVMe and linux driver&#x2F;OS bugs on these are the last bottleneck, hopefully solved soon!