Wow, a really interesting video @Solvetheriddle and what a big brain Arthur Mensch is! I am so glad I saw this video, because it frames the big question that is at the heart of a large part of my investment strategy. So much so, that we should open a new forum "LLMs vs Enterprise Software".
The General Question
Arthur articulates what has been a developing hypothsis of mine over the last year or two ... general AI models will continue to develop and converge, but the path to value lies in specialised application and development within industry verticals.
He is clearly onto this point in articulating that the path to value lies in specialised application within the enterprise context and access to the enterprise data.
My thesis is that the owners of end customer data will be best placed to create value, whether thats $RMD, $WTC, $SDR, $XRO, $PME, $CAT, $TNE, ... I hold all, as well as $REA, $CAR, etc. as well as banks and big retailers etc. which I don't hold.
One model for the competitive landscape's evolution is the agentic AI development that $WTC is following. Treating the AI model as a commoditised service that can readily switch between and maintain a firewall around its proprietary data.
Clearly, Mistral.ai has identified the need to engage with the verticals and access the enterprise data, so that perhaps frames a key frontier for competition and industry evolution from here. Will the AI developers be able to access the process and data architecture of enterprises, so as to be able to lead the next generation of re-architecting enterprise software? I don't know.
We've already seen several iterations of this over the last 3-4 decades.
First, software has codified processes, with local databases (local to a specific process or function) allowing process efficiency improvement through the increasing improvements in how process information and workflow is managed within the process.
Second, the information (data) embedded within processes and systems, has been extracted and consolidated iinto an enterprise data layer, with the application layer built on top. This is essentially what the big ERPs did at enterprise level during the 1990's - 2000's. It allows efficiencies and productivity gains across processes, at the enterprise level. Many companies and industries are still at the relatively early stages of exploiting this (witness my earlier post today about adoption of IT systems in managing medical imaging data, regarding $PME.)
Thirdly, over the last 20 years, with the advent and deployment of the cloud, we've seen the software providers move from providing the code into the enterprise, to holding code and sucking out the data via cloud. That has made the enterprise software providers very valuable and powerful. They are specialists in the industry, they have codified standarised processes, and they can aggregate and lock-uo data for an industry across mutiple customers. This allows them to conduct analytics and learning on that data to provide mode value-added serfvices, as well as to innovate to make the processes better.
So the "power" lies both in the "process knowledge" and the "enterprise data".
In each of the IT revolutions that have unfolded over the last 50 years or so, "experts" have predicted revolutions, and while these have occurred, they have taken much longer than predicted. Why? Because a number of things get in the way: people (including human adaption), governance, regulation, "accidents" etc.
I think two industry players to watch carefully are SAP and Oracle. Both have made transitiions to cloud-based SaaS offerings over the last decade. But each is large enough that they have specialist teams in multiple industry verticals. But more importantly, each is also developing their own LLMs! And as Arthur has made clear, the technology around general LLMs is converging, and their are both Europe and Chinese open-source offerings, that appear to be converging with OpenAI, Google, etc.
Key Takeaway: SAP and Oracle appear to be to have the ingredients to be long term winners.
$WTC and CTO
So where does that leave a business like $WTO?
While SAP and Oracle have scale, and can build and deploy their own LLMs within their moats, a company like $WTC has its focus on a single vertical.
You've asked about CTO.Transport optimisation is a mathematical challenge that goes back at least as far as the 1950s. The text I recently taught supply chain management to MBAs, for example, has a exercise for students to optimise transport costs by framing a linear optimisation problem for moving materials around factories and warehouses. Academics, manufacturers, and software company have been working on this problem for decades.
And so it is hardly surprising that there are mutliple CTO-like initiatives going on around the world at the moment. The problem for CTO is that for it to work, you have to get multiple parties to agree to collaborate: transport hubs, shippers, logistics firms, freight forwarders. And that's really challenging. With that in mind, if you watch the relevant segment of the $WTC Investor Day, I think there are multiple signs in the presentation that $WTC are coming up against this in their current NSW pilot.
The "mathematical optimisation" is "easy". The commercial and operational implementation is the hard bit, and of course, the human behavioural resistance to change makes it even harder. I suspect there is plenty of disppointment that lies ahead in terms of timelines and revenue growth for CTO.
However, there is value in the wider supply chain integration, and AI is going to play a role here. And that's why I think $WTC had to acquire Blume, Envase, Matchbox and get going on CTO, and finally building up to E2Open. If they tried to build the process knowledge, client relationships and data organically, it would have taken another 10-15 years. Any by that time it will be game over. The likes of SAP will have won.
In summary, on the specifics of $WTC and CTO, I think it is "game on". I don't think the LLMs themselves are the challenge.
My belief is:
"You won't lose your job to AI, but to someone who is better at using AI than you are."
I think this absolutely applies to enterprise software. And so, I am focused on which enterprise software firms are proving to be most adept at using AI.
Finally, are we in an AI Bubble?
I'm not sure. But for all the $trillions going in to AI infastructure, I think the penny is going to drop (maybe this year) that the "path to value" lies in accessing enterprise process and information. All my experience tells me that this will take place over a longer timeframe than current AI valuations expect. If that penny drops, then some of the heat will come out of the AI sector, and we could see some very significant value corrections. This is not unusual,... afterall, haven't we seen this in pretty much every historical technology innovation cycle?
(And for that reason, I wouldn't want to be a shareholder of OpenAI!)