Benedict's Newsletter: No. 632

In this issue: Anthropic versus the Dod, more bubble vision, Meta AMD, Block AI-washing, and deflationary bubbles

NO. 632   FREE EDITION   SUNDAY 1 MAR 2026 SPONSORED BY LINEAR Product teams at OpenAI, Coinbase, and Automattic switched to Linear Curious to try it but stuck in a contract with another issue tracker? Talk to Linear before Feb 15th and they'll cover the gap until your existing contract runs out. Learn about switching My work How will OpenAI compete? OpenAI has some big questions. It doesn't have unique tech. It has a big user base, but with limited engagement and stickiness and no network effect. The incumbents have matched the tech and are leveraging their product and distribution. And a lot of the value and leverage will come from new experiences that haven't been invented yet, and it can't invent all of those itself. What's the plan? LINK AI eats the world Twice a year, I produce a big presentation exploring macro and strategic trends in the tech industry. The latest edition: 'AI eats the world'. LINK News Who chooses?  Anthropic's contract with the US military says that it can't be used for 'domestic surveillance' (which is probably illegal anyway) or autonomous weapons systems, because Anthropic thinks that the tech isn't reliable enough for that yet. The Secretary of War, Pete Hegseth, has decided this is an unacceptable limitation and, in the manner of the Trump government, declared he will use all sorts of legal powers to force Anthropic to remove these limits, including banning the rest of the government from using it, and/or threatening a takeover. He may not be allowed to use those powers in this way, but plenty of other Trump officials are already demonstrating they don't recognise any legal or constitutional limits on presidential power, so that might be moot. OpenAI reacted by saying that it would be willing to provide what the DoD wants, which prompted a backlash inside and outside tech, and Anthropic is suddenly on the app store charts. This might be another 'delete Uber' moment (that one had limited impact, but Lyft was a weaker alternative).  However, the underlying issue is worth considering with nuance. Sure, Anthropic can write a contract, but what should it say? There are only a handful of companies that can build this - should their CEOs, unelected and unaccountable to anyone except their shareholders, decide what capabilities the US military has? Should they decide what an acceptable level of accuracy is, when they're not the ones in combat and don't know the context?  Meanwhile, the deeper context, of course, is that Anthropic is supposed to be the 'safe', 'ethical' lab, that's worried about AI killing us all, so how can they sell the military a system to kill people? Well, that's what the military does, and pacifism only works if everyone else is a pacifist too, which the last 5 years and last 87 years demonstrate is a foolish plan. So who decides? ANTHROPIC, OPENAI Bubble scenarios Another week, another viral blog post scaring the markets, this time from Citrini Research. As with previous ones, the ideas and arguments are not great (see eg this exasperated rebuttal from Citadel), but the fact that things like this create such strong reactions is more interesting. See this week's column. LINK, CITADEL Meta x AMD - more circular revenue Everyone wants a second source: Meta has done a deal with AMD for chips. Meta will buy up to 6 gigawatts of chips for up to $100bn over the next five years, and gets warrants to buy up to 10% of AMD at $0.01 per share (current price $196), which is worth $33bn now, but Meta would only get the full allocation if AMD's share price rises to $600, making 10% worth, well, $100bn. AMD did a similar deal with OpenAI last October. These kinds of deals always work well on the way up. META, AMD OpenAI's money tap The makers are getting more and more nervous, but OpenAI is still raising - or at least, locking in position before the music stops. This week it raised a $110bn fund (including $50bn from Amazon), which is apparently a record (and triple the largest ever IPO, Aramco's $29.4bn in 2019). LINK AI-Washing at Block?  Block (formerly Square), the PoS company run by Jack Dorsey, announced that it would cut 40% of headcount, attributing this to AI automation. Most people in tech attribute this rather to Dorsey just massively over-hiring, much as he did at Twitter - Block went from 5.5k heads in 2020 to over 11k in 2024. There will be a lot more of this kind of AI-washing. LINK The week in AI Anthropic expanded on its complaint that the leading Chinese models are mostly just scraping US models. Of course they are - they'll use any advantage they can get, especially if the US blocks cutting-edge chip exports (though the story there seems to change every week), but worth bearing in mind when you wonder why Chinese models are so good without equivalent capex. LINK More new/updates tools from Google: enhancements to Flow, its 'content studio', and a new version of Nano Banana, the flagship image generator. FLOW, BANANA Selling back doors The US is prosecuting a vendor who sold hacking tools used by the US government to anyone with enough money. This is why tech companies refuse to create backdoors in their products for law enforcement - if you create a skeleton key for the police, the police will lose it. LINK About What matters in tech? What's going on, what might it mean, and what will happen next? I've spent 25 years analysing mobile, media and technology, and worked in equity research, strategy, consulting and venture capital. I'm now an independent analyst, and I speak and consult on strategy and technology for companies around the world. FIND OUT MORE Ideas OpenAI made a lot of noise about 'Stargate' and its $100bn of capex, but I was very doubtful that much of it was real. The Information reports that a year later, nothing much has happened. LINK The messy middle of retailer deployment of generative AI. LINK A US intelligence agency report on the dangers of malware inserted into LLMs, or indeed into their training data. LINK Meanwhile, Google published its own cyber forecast. LINK Stripe's annual letter has some interesting ways to think about agentic commerce. Also note their interest in crypto as alternative payment rails. I have always taken blockchains much more seriously for their potential as arcane finance industry plumbing than as something that any consumer should see or use as software. LINK How is AI changing advertising? Part of the deployment phase of each new tech, and now AI, is specialisation - you need to know a lot about ads, or chips, or logistics to know what this will mean for each industry. LINK A UK SEO 'agency' bought a network of UK games sites, fired everyone, set up a bunch of fake journalist profiles, and published AI-generated spam promoting online casinos. People have been using automation to spam Google for 20 years, and Google has been banning them for just as long. LINK Ukrainian drones continue to hit Russian infrastructure over a thousand miles from the border. We're watching a slow-motion Pearl Harbour, and it doesn't even make headlines anymore. LINK Meanwhile, the FT reports on a debate in the military build-up: drones are consumables, like artillery shells, but where you can stockpile shells for decades, drones go obsolete a lot quicker. But how much quicker, and how much of that is just about updating the software? LINK Outside interests An online archive of Walkmans and all the other pocket tape players. LINK Animating a classic Chinese scroll painting. On an artistic level this is kind of crass (and apparently the boat is supposed to look like it's going to crash, symbolising the weakness in the empire), but it's hard to beat for child-like delight, and for the sense of possibility - imagine making this without AI. LINK Data The Fed on infrastructure spending in trade stats. LINK And NBER has global data on enterprise AI adoption so far. LINK Apollo points out that most of the declines in hiring and employment that people try to attribute to AI are also very highly correlated to more traditional macro changes. Opinion in tech is pretty divided about this, since many of the changes seem to start before one would expect generative AI to be widely deployed, or even before anyone outside tech really knew about it, and because some of the companies involved massively over-hired during Covid. LINK Bain's annual private equity report. LINK Updated Pew data on how US teenagers use generative AI.  LINK OpenAI launched 'Signals' - some (rather vague) usage trends data. LINK Edison Research thinks US podcast listening is now larger than spoken word radio. LINK Preview from the Premium edition Scare trades and deflationary bubbles  One of my memories of the dotcom crash and the parallel telecoms crash was that every month or so, a story would go around that was really obviously stupid, both at the time and now looking back, and the stocks would go down 10-15%. But also, the stocks were way too expensive. So really, what was happening was that silly stories catalysed well-founded concerns.  I thought of that looking at the blog post that spiked the markets this week, from a firm called 'Citrini'. It talked about an AI threat to marketplaces that shows they know nothing about marketplaces, a macroeconomic impact that shows they don't understand macroeconomics, and plenty more - it's a kind of compendium of the Dunning-Kruger syndrome. In response, a bunch of people put their heads in their hands, said 'I can't believe I have to waste time on this', and laid out one part of why this is probably wrong - Citadel, clearly exasperated, is a good example.  However, as we all try to digest every day's torrent of AI news, I think there's a more interesting point.  Most bubbles are about a transformative new thing (railways, radio, the internet) that people think will create a huge amount of new value. It turns out that the value takes longer, and comes to different companies, and maybe never matches the prices at the peak, but the hype was about creation. The internet destroyed some value too, most obviously in newspapers, but that wasn't on anyone's mind in the dotcom bubble - just look at the Time Warner merger.  However, in this bubble, if, as we did in 1999, you believe that all of this will work Right Now, this afternoon or maybe tomorrow, you can also believe that the first consequence will be value destruction, not value creation.   Upgrade to Premium You're getting the Free edition. Subscribers to the Premium edition got this two days ago on Sunday evening, together with an exclusive column, complete access to the archive of over 600 issues, and more. SUBSCRIBE OR UPGRADE THIS WAS SENT TO abdulkerim.bosnia.business@blogger.com ON 3 March, 2026, BECAUSE YOU SUBSCRIBED AT BEN-EVANS.COM. You can UNSUBSCRIBE, OR READ THE PRIVACY POLICY. OUR MAILING ADDRESS IS Benedict Evans, 2 Golden Square, London, W1, United Kingdom.   COPYRIGHT © 2013-2026 Benedict Evans, ALL RIGHTS RESERVED.