Canonical URL
Do not index
Do not index
Two big strategic moves in the AI space are extraordinarily important to how the world works (or not) and will continue to work (or not). Fair warning, they’re both connected to Stardog indirectly so if you’re only here for the commercial stuff, then maybe this won’t be yr favorite post? I hope you read it anyway since these two big AI moves are not only interesting but also, as I said, important.
So, two big AI moves. One is recent and ongoing. The other is older but still reverberating. The first is software; the second, hardware. The first is the biggest blunder in the Internet era. The second is the greatest strategic acquisition in tech history that no one talks about. While I’m not a Wall Street tech analyst, I do pay attention to both AI and the history of IT.
Google’s Epoch Blunder
Google’s epoch blunder is “epoch” instead of “epic” because, if they go through with it, the blunder will end one epoch, which we can all “the web as the greatest information system ever created”, and begin another, which I will call “Google uses AI to enshittify the Web”.
Look, man, Google does a lot of things. From moonshots to AI research to critical infrastructure like DNS, web caching, browsers, mobile. But the value of Google to the planet is simple: Web Search.
The cultural value of Google Web Search is simple: it turned the Web from an academic curiosity into the most important part of the Internet, the part that ate all the other parts of the Internet. The value of Web Search to Google shareholders is simple, too: the single greatest cash machine ever created in tech.
And now they literally want to throw it all away, but admit they have no idea how to make the AI Web as useful as Web Search since they have no idea how to prevent it from literally making shit up.
This one matters a lot because it’s ongoing, in fact, we’re at the very beginning and it’s not too late for Google to reverse course, though I’m pretty sure they won’t do that.
Google is in a great position from the perspective of tech, but the commercial position is much murkier. To monetize its AI advantage, it has to bring that tech to bear commercially. Where else can it do that with the same scale and pace except Web Search? But, if in doing that they destroy the quality of Web Search, then it will be the biggest blunder in tech history.
NVIDIA’s The Greatest Acquisition, Ever
NVIDIA’s $6.9 billion acquisition of a relatively obscure Israeli networking company called Mellanox in 2019 is the greatest acquisition, that no one really talks about, by any org in tech, ever.
But first let’s be even more clear; general consensus is that three acquisitions are the biggest ever. I claim NVIDIA’s acquisition of Mellanox is bigger than Google's acquisition of Android in 2005, Facebook's of Instagram in 2012, or Microsoft's of LinkedIn in 2016.
NVIDIA’s dominance in the AI space is absolute in mid-2024. Among AI’s other disruptions, it’s colliding the enterprise tech space—which is big—with the high-performance computing (HPC) space, which isn’t as big, but is critical in terms of pace and direction of innovation. The easiest way to see this is to compare Infiniband and Ethernet. We get the Ultra Ethernet Consortium, on the way to 1.6TB performance levels, not because of the Cloud or Enterprise IT, neither of which need anything like that level of performance. No, we get it because of Infiniband because of AI because NVIDIA acquired Mellanox, putting its unified infrastructure performance advantage out of reach of all but a handful of companies (and then only because a monster like Dell joins with a behemoth like Broadcom, together with AMD)—and it’s not at all clear that any of them can catch up.
In a straight knife fight between Google and NVIDIA and AWS, I don’t know who wins. Each is utterly formidable; but, ironically, Google and AWS are hampered by their hyperscaler position because neither is going to sell their special sauce outside their global computing environments. NVIDIA is going everywhere: robotics, AI PCs, edge devices on factory floors, the Cloud, etc. All of that is backed by NVIDIA as a one-stop shop for AI and compute infrastructure.
The acquisition of Mellanox means that—
- NVIDIA has a foothold in enterprise computing at the network layer, which is why it now finds itself in a fight with Broadcom.
- Suddenly with the release of Grace Hopper and, soon, Blackwell, the Supercomputer 500 sees NVIDIA-systems at the top and I expect that trend to continue.
- AI training and inference workloads, which are an unending growth machine, so far, are only available in one hyperconverged stack from a single vendor. Yes, Google TPUs are formidable; but how do you run them in Azure or OCI or…? You don’t.
Takeaways?
I depend hourly on Google Web Search. It’s been for almost 25 years a valuable and reliable information utility, the most valuable and useful ever created. I don’t want AI junk in my search results! I don’t know anyone who does.
NVIDIA has no real competition in terms of unified compute infrastructure for AI. That only matters to the extent that AI is a game-changer. I think that it is and I want it everywhere, except in Web Search, for as long as the machine tells me to put glue on my pizza!
