Microsoft's AI play(s) make it clear it's early in a long AI race.
Microsoft's multiple AI plays make clear how early we are in AI. Microsoft has put 3 entries into market on AI in different ways - launch of Copilot, ownership interest (or at least revenue sharing interest) in OpenAI, and now an acqu-hire of maybe the key guy in Inflection. Personally, I think these 3 bets all make sense as a portfolio play in an effort not to get left behind any of the key players in AI (and with Microsoft and Google/Alphabet's wallet and free cash flow, it seems there is unlikely to be one big winner in the space). Here are my thoughts:
1) The big players (including MSFT, GOOG, OpenAI) will prefer not to have a dominant player, so you will see a set of portfolio plays like this and these will evolve over time as the market evolves. Nobody wants to cede a major portion of the space to anyone else (witness Yahoo!'s historic cave and agreement to give Google the search results while they focused on display ads and other opportunities - thus giving away any chance of being a strong contender in the search to Google in exchange for a rev share - whoops). At the same time, nobody knows where the big winners will come from, so expect a lot more investments and a lot of hedging. Everyone watched Facebook really pull away as the dominant social media company through product and acquisition strategies, and nobody wants to let the other(s) become a Facebook in AI.
2) The big winner with multiple players trying to form some type of informal oligarchy is clearly going to be Nvidia. Already crushing it for the GPU market and with capital and a great set of products, they stand poised to become the Intel or Cisco of overall AI with all of the key players (OpenAI, MSFT, GOOG) likely to buy their GPUs and ever-faster and ever-costlier ones for a while. It's like selling Sun Netra servers to eBay in the late 90s - just put the rep on speed dial and start ordering 6-packs every couple of weeks and you're about right. Nvidia is in a great spot and everything coming out of their big conference (which was a direct conflict with World Agri-Tech so many of us could not be in 2 places at one time ...) suggests they intend to stay there. It's unclear how many competitors can actually develop a product set to be a serious competitor to them in the short term.
3) We are still building some pretty early-stage LLMs but the pace of play is as fast in the space as I've seen in tech (and I've been watching since the 80s, so it's been a while). We went from talking about "wow, look at that" for OpenAI and first-gen tools that worked at all about 18 months ago to a whole lot of infrastructure investment (more on Nvidia's 2023 in a future post - in a word - wow), training data strategies evolving (hello Reddit's $200M data deal with Google) to the possibility of next-gen AI desktop and mobile clients becoming real fast (hello Apple talking to Google about Gemini becoming some type of client component on iPhone).
4) All of this reminds me of the internet in the mid-90s. We had to build a lot of network plumbing, put a lot of browsers on desktops, enhance desktop and laptop connectivity, and get hundreds of millions of people using the internet (and email - one of the initial killer apps - hello Yahoo! mail) so that the application layer could really shine. That's why Cisco and a bunch of data center growth was a key enabler for the internet to really grow up fast. Once the plumbing got bought and built out, then we could really get to work with applications to start converting so much of the off line activity to on line. But if you had bet on applications initially, you had to have very patient investors and/or very patient capital while the infrastructure got built. Remember, Amazon didn't go public until 1997, eBay in 1998, Google in 2004, and Facebook was all the way in 2012 (and 3 of those 4 are still major players in their segment). Compare that with Sun (1986 IPO - remember "we're the dot in dot.com"?) and Cisco (1990 IPO) - equipment had to go first.
领英推荐
5) I am more confident than ever in predicting that AI will have the largest impact of any tech segment in the last 20+ years. Once e-commerce, search, mobile, and social networking were launched and evolved, each of them found their swim lane (some bigger than others) and become major tech segments. The possibility for AI is to help level up all of the tech segments, from enterprise to customer service to development teams. Plus the pure pace at which it is happening is tremendous to watch, with product launches, M&A, investments, and partnerships happening almost weekly it seems like.
I have said (early and often) that living in interesting times with quick change is the best way to live. Everything I'm seeing in AI suggests that (1) we live in interesting times; and (2) there's no end in sight to the interesting and the change.
Game on!
Managing Director at Voyager Capital
8 个月The other major innovation for pre-farmgate ag is Starlink, now available in 72 countries around the world. Nvidia GPUs on the farm get network effects with Starlink.
Founder @ AutoFarmWorks | We reduce production cost/acre for cash crop farmers by 12%
8 个月Agree 100%: 3) AI for it's own sake is irrelevant to growers. Sell value where it makes their operations better.
Innovating farming with actionable data and AI insights | Founder at innov8.ag
8 个月Well said. While the Google-Reddit deal made headlines, Microsoft also sitting on a content goldmine with LinkedIn.
Ag-tech | Innovation | 1st Gen Parking Lot Farmer
8 个月Completely agree & to specifically call out the importance of (3). Get in and prove a tangible value proposition. Then build the relationship to establish that other things can come to layer on top (that will also have a tangible value proposition). Start with here’s what I can help you with today & what can I help you with tomorrow?
Harnessing the forces of Creative Destruction
8 个月An interesting externality More AI -> More energy -> more energy offsets -> more investment by AI companies in regenerative ag. AI early success is expected in education, healthcare, and programming. Should AI companies think about AI for Ag to help access Regenerative offsets to their energy usage?