8 February 2026
News
SaaS is dead!
Software stocks fell heavily on fears that generative AI will automate away their entire business - particularly legal software and research companies, after Anthropic added a legal plugin to Cowork. Thomson Reuters was down 15%. See this week’s column. LINK, COWORK
The capex boom continues
The four big cloud companies have now announced their December quarter results and given outlooks for 2026. As expected, between them Meta, Alphabet, and AWS spent about $260bn on capex in 2025 and plan close to $500bn in 2026. Microsoft hasn’t given guidance for the year but also expects growth, meaning we will get to something over $600bn. This will probably require them to raise capital, since this is most or all of their cashflows. I don’t do stock prices here, but market reaction was very mixed, which is interesting: people are wondering if Satya Nadella’s OpenAI strategy is working out at all well.
Meanwhile, Google’s search revenue was up 17% year on year - remember when people said Google search was dead? Google and Meta both have a lot of headroom just for using AI to rework their existing businesses. LINK
The week in AI
Last autumn Nvidia announced a plan to invest ‘up to $100bn’ in OpenAI (which became part of the famous ‘circular revenue’ narrative), but this week the WSJ claims the deal was stalled and might not close. Jensen Huang and Sam Altman claimed all is good. The market is febrile. LINK
OpenAI launched ‘Frontiers’, a tool for enterprises to manage agentic systems. LINK
Anthropic launched an ad campaign boasting that it doesn’t run ads (but reserving the right to in the future). It’s easy to say you won’t run ads when you have no consumer users. This is silly. LINK
The FT reports that KPMG pushed its own auditor, Grant Thornton, to pass on AI cost savings, possibly not considering that all of its own clients will do the same. LINK
Grok CSAM
Both the French and British regulators are looking into how much non-consensual porn and CSAM Grok produced when Elon Musk decided to remove the guardrails. He claims this is a ‘free speech’ issue. Maybe Epstein should have tried that as a defence. LINK
Social bans for teenagers
The Spanish government says it will follow Australia in banning social media for under-16s. France, Denmark and the UK are considering similar moves. I am on the sceptical side of this (I see moral panic and cherry-picked data) but it’s certainly a trend. LINK
X mergers
More a financial engineering story than a tech story: Elon Musk merged his xAI AI lab (which also owns Twitter) into SpaceX, valuing xAI at $230bn (based on a funding round in January) and SpaceX at $1tr. The combined company plans an IPO in June, which hopes to raise $50bn (beating Saudi Aramco’s record of $29bn in 2019). Musk claims that the combined company will aim to build data centres in space, which he claims will happen in ‘two to three years’. Musk’s timelines obviously have no predictive value, but he’s not the only person thinking about this, and it may not be (quite) as impractical as it sounds. However, the more pressing motivation might be that SpaceX has positive cashflow ($16bn of cashflow last year, apparently) and is free to subsidise xAI’s model-building. Meanwhile, the IPO might just be to beat OpenAI and Anthropic to the market - there’s only so much capital at any one time. LINK
Ideas
The Information reports that OpenAI is hiring hundreds of consultants (AKA ‘sales engineers’ or ‘forward-deployed engineers’) to help enterprise clients implement ChatGPT use-cases. SaaS is dead and consulting is dead because the AI can do it all…. except you need the consultants to implement the AI as SaaS. LINK
“Hate brings views” - rage-posting lies on TikTok for the revenue share. LINK
The NY Times has a fascinating story about the impact of generative AI on romance fiction. Both the genre and the technology are predictable and formulaic, and the authenticity of the author’s experience is not typically central, so this is ground zero for adoption. Some people are very upset that they bought something ‘made’ by machine, but the more interesting question might be what happens when the whole book can be generated for you from scratch as you read it (which will probably also come to porn). LINK
Remember OpenClaw? I wrote last week that there were obvious security questions (which is one reason Google and Apple haven’t launched something similar yet): this week it turned out that the most popular third-party plug-ins have malware. LINK
The WSJ has a long profile of Polymarket, the leading ‘prediction engine’ / gambling machine. LINK
Microsoft has spent the last two years adding AI features everywhere and calling them all ‘Copilot’, and now users are confused and irritated. All of the incumbents have added AI everywhere (incumbents always try to make the new thing a fetter) and of course not all of those features work out, but giving it all the same name means people expect it to work together, and it doesn’t. In a sense ‘Copilot’ is the new ‘Watson’ - a marketing umbrella, not a specific, coherent product. LINK
The Washington Post had a bad week, but some other US news companies are growing strongly. LINK
The tricky local politics of AI data centres. They use land, and (maybe) push up power bills, but they’re also jobs and the future. LINK
A McKinsey think piece on what AI might mean next for film and TV production. LINK
Outside interests
The US federal government has now arrested so many people illegally as part of its attempted immigration crackdown that DoJ lawyers can’t keep up with all the court orders requiring releases: close to 100 people are now being held in violation of court orders, many more are waiting for those orders, the DoJ lawyers say they can’t get ICE to respond, and judges (some appointed by Trump) threaten contempt proceedings. But of course, this is in Minnesota, not San Francisco, so it doesn’t matter, right? POLITICO, LAW
The Cooper Hewitt museum in New York is planning ‘the art of noise’ LINK
Anti-depressant or Tolkien character? LINK
Data
OpenAI’s lead in consumer use is shrinking. LINK
Column
SaaS is dead?
What does AI do to software? We can make a pretty obvious list of building blocks to think about.
First, clearly, AI means that it will be an order of magnitude cheaper to get any given piece of software written. That means more software, quicker, and hence more things being automated with software that might not have been economic to automate before. I have suggested for a while that we could compare this to AWS, which was also an order of magnitude change in how quickly and easily you could build and deploy software.
Second, just as clearly, AI will unlock a broad new set of tasks that could not be automated with software before, at any price.
These two combined mean there will be a lot of churn in the software industry, as people make new things and new companies, and compete on new grounds. Some incumbents will make this a feature and others will be demolished by a new way of solving or even defining the customer’s problem. We had this with SaaS too. Also like SaaS, this will change the margin structure (although right now we don’t really know where LLM costs will end up).
After that, things get a lot more frothy. There is a view that we won't have far more software but far less software, because now you won't need that piece of enterprise tooling to solve your problem. Now, a lawyer or head of accounts payable or a supply chain manager can just ask ChatGPT or Claude to do the thing. You won't need a subscription to Westlaw because you can just get CoPilot or Cowork to find the citations you want.
This thesis has pretty much everybody who knows anything about software (including Jensen Huang) rolling their eyes and saying “I can't believe I have to waste my time explaining why this is nonsense”. (It's also the kind of irrational volatility that you get in bubbles, and crashes, once people start getting nervous.)
There are really two things to understand here. Firstly, narrowly, companies need agreed, consistent sets of processes that are done in the same way by everybody. Everyone has to be using the same inventory management system that works in the same way, and not just the same database. It doesn't work if one person decides to make one for themselves.
Second, and more important, when you actually go and look at successful software, the users generally didn't see the problem, didn't see how you would solve it, and could not have sat down and thought about what should happen on every screen, how it should get built, and how you get everybody to use it. There is an enormous difference between knowing something about how your company and how your job works and being able to identify a set of problems and a set of workflows amd think about how those could be automated.
In other words, the fact that you're writing the code in natural language doesn't mean that you don't have to work out what the computer should do.
Setting the nonsense aside, however, I think there are perhaps a couple of more interesting places that this could go.
First, the typical big company often has hundreds or thousands, or tens of thousands of separate legacy systems, databases, and applications that have accumulated over time. Dell, for example, has 4,700 apps and 10k databases. Part of the promise of agents is to connect, synchronise and rationalise all of those, and to change what the software estate looks like and how people can interact with it. So, both more software and less software.
And second, rather than imagining that people create their own tools, I think it's more interesting to think about how much is done without dedicated tools, with CSVs and things like Tableau or Excel. Today, to put it very simply, you might do that task inside Oracle or you might do it in a vertical SaaS app with specialised tools, but if it’s a one-off, ‘improvised’ problem, where you haven’t built a dedicated workflow, you might export the CSV and analyse it in Tableau or Excel. A chatbot might be a new, different, and expanded way to handle those kinds of improved problems - it won’t replace software, but expand the space around it. In other words, there is software that is formalised, institutionalised process, and then there is software that is improvisation. You won't replace process with improvisation - you don’t replace Salesforce with ChatGPT any more than you replace it with Excel. But there's a lot more that you could automate if you could improvise more.