6 April 2025
News
Trade wars
You don’t need this newsletter to know that this week Donald Trump started a global trade war, aiming to dismantle the economic system that has made the USA by far the richest country in the world. However, there are a few direct implications for tech companies, beyond the macro-economic effects that apply to everyone. See this week’s column below. LINK
The end of de minimis
Part of Trump’s tariff push is ending the ‘de minimis’ rule that exempted inbound packages with a declared value of less than $800 from customs inspection and duty. This has been subject to pretty bi-partisan pressure for a while, as Shein and Temu shipped direct from manufacturers in their networks to US consumers, bypassing normal tariffs and safety rules as well as things like sanctions on cotton from Xinjiang. They had already begun trying to build up more conventional shipping and local distribution, but of course the broader tariffs make that moot. LINK, RULE
Meta returns to the LLM leaderboards
Meta led the open source trend of LLM development, but hasn’t launched a new version of its flagship Llama foundation model since last summer, and the models were good but not the best (though DeepSeek used them as part of its own open source models). But this Saturday it released Llama 4, which gets it to the top of the leaderboards - it scores 1417 on LMArena versus 1439 for Google’s Gemini 2.5 (1st place) and 1410 for the latest ChatGPT 4o (3rd). There is some debate quite what these benchmarks really tell us, but Meta is SOTA, whatever that means, and since it’s open source (mostly) you can download that model yourself (unless you live in the EU, in which case this is over the threshold for ‘systemic risk’ under the EU Act and you’d better get a lawyer first). NB - there is some suspicion that the release was on Saturday to get ahead of someone else’s big release this week. LINK
OpenAI returns to Open?
Sam Altman announced that OpenAI will release an open LLM for the first time since GPT2 in 2019. No word on how good that model will be compared to the closed SOTA, though. People complained for years that OpenAI isn’t open, and Altman has spent almost as long loudly declaring that open source LLMs are massively dangerous and a threat to whatever the listener thinks is most important. Altman, meanwhile, spends a lot of his time manoeuvring and positioning so, as Metternich said when Talleyrand died, “What did he mean by that?” LINK
The week in AI
As reported last week, OpenAI closed a funding round raising $40bn at a $300bn post-money valuation, the largest private round in history. Fun fact: the top 20 tech venture rounds in 1999 added up to $2.9bn ($5.6bn adjusted for inflation) and all US tech venture funding in 1999 was only $30.5bn. LINK
And some context - The Information reports that OpenAI has 20m paid subscribers, implying a revenue run-rate of close to $5bn even before counting API revenue. LINK
There are growing reports that Microsoft is pulling back for at least some of its data centre construction. It’s not clear how much this is reconfiguration based on developments in LLM technology, chips or geopolitics, or concerns about tariffs, or OpenAI shifting its infrastructure elsewhere. The company has given guidance of $80bn of spending for the 12m to June. LINK
Amazon is experimenting with showing product and inventory from other companies’ websites within search results if it doesn’t stock the product itself, and using an AI agent to help the customer buy from that website. This seems very clever, but also potentially very brittle. LINK
TikTok tick tock
Amazon and AppLovin (in-game ads and analytics) have submitted bids to buy TikTok, and my old employer A16Z is apparently involved with Oracle's bid. However, for a second time, Trump declared that he was extending the deadline for a sale by 75 days (and again without any apparently legal authority to do this, so this is really just a declaration that he won’t enforce the law). The real story appears to be that China won’t let Bytedance make the sale, while Trump has declared that he wants to use this as bargaining for his import taxes, although the tax on Chinese imports seems to add up to $200bn-300bn a year (54% on 2024 US imports from China of $440bn), against which a one-off sale of TikTok at anything from $100-200bn is only one chip. LINK, A16Z
Apple’s fintech shuffle
Goldman Sachs has been trying for a while to get out of its credit card and savings account deal with Apple as it U-turns out of consumer, and now, apparently, Visa is offering $100m for Apple to switch network away from Mastercard. Amex is apparently bidding to take both the network and issue role. The card and savings products are good, but US-only and not very large - a lot of the interest is about whatever Apple does with this next, I suspect. LINK
The James Bond of enterprise software
Rippling, which makes enterprise HR software, accused Deel, a competitor, of paying an employee to pass it confidential internal information. There are some wild accusations, a lawsuit, and a little bit of history. LINK
Ideas
A write-up of MCP, the hot new API standard for LLMs (of which I expressed some skepticism last week). LINK
China has lots of empty AI data centres. This has always been the model, though - China funnels lots of investment into new spaces and sees who sinks and swims, which can be wasteful but ultimately produces winners (Oppo, BYD…). LINK
Wikipedia is being scraped continuously for AI training data and it’s complaining about the cost (on the other hand, Wikipedia is massivelycash-generative, so it can probably afford this). LINK
Doug Shapiro outlining the narrative that the last disruption in media was distribution becoming free and the next is creation becoming free. LINK
China is demonstrating a deep-sea fibre cutter. LINK
China’s Starlink alternative. Everyone is now thinking about their own Starlink alternative for military uses, as Ukraine has shown how essential it is to modern war and Trump and Musk have demonstrated that you can’t know if you can rely on the USA. LINK
The Chinese tech industry is pushing ‘Sparklink’, a possible replacement for Bluetooth. Standards wars are eternal. LINK
Using AI to make artillery targeting more efficient. LINK
Unilever wants an army of influencers to promote its products. LINK
Outside interests
“Let’s not reinvent the ocean” - the Ford exec who collected malapropisms. We need to talk about the elephant in the closet. LINK
A 10-metre scroll with 108 seals. LINK
Data
EY report on Indian media and online use. LINK
Bain report on Indian e-commerce. LINK
A Pew study on US perceptions of AI - as in many similar studies, about 10% of people are using LLMs regularly and two to three times more have tried them and not found them useful. I don’t think this is a time problem - I think it’s a product, interface and use-case problem. LINK
Column
Trade wars in tech
“There's a confusion about China. The popular conception is that companies come to China because of low labour costs. I'm not sure what part of China they go to, but the truth is, China stopped being a low-labour-cost country many years ago” - Tim Cook, 2018
Donald Trump is trying to end the economic system that made the USA the richest economy on Earth, with the declared aim of moving much manufacturing that is currently spread around the world to the USA. You probably already have an opinion about this, and you can read plenty of others from economists. The stock market is crashing, including tech stocks, as investors react to the prospect of a recession, inflation, and a major new tax burden. But as I said above, you don’t need me to tell you this - what does it mean as a tech analyst?
The company with the most obvious hit is Apple. Most of its products are assembled in China, which now has combined tariffs of 54%, and the ones that aren’t are mostly made in other countries with equally massive new tariffs - for example, Vietnam, at 46%. Hardware is 75% of Apple revenue, and it probably has a 35-40% gross margin on the hardware, so a $1,000 iPhone will probably face a $300-350 import tax in the USA. To state the obvious, neither Apple nor its suppliers have the margins to absorb that even if they wanted to, so prices will go up. And then, while the USA is about 40% of revenue, ‘Greater China’ itself is another 17%, and how will the Chinese government (or indeed Chinese consumers) react to this over time?
Can’t you recreate manufacturing in the USA, though? Isn’t this short term pain for long-term gain? Well, here there are three answers that apply to a lot of the sectors where that is suddenly being asked. First, yes, but it would be expensive and take a long time. It’s taken TSMC years and huge amounts of money to get to its first plant in the USA, and that’s just one company, not an attempt to replicate an entire ecosystem. Foxconn has struggled for years to make iPhones in India, and again, it takes time. Meanwhile, much of the capital equipment you’d need is itself subject to the same import tariffs (though chips themselves are not). Second, how do you plan to change your entire supply chain and invest tens of billions of dollars over 5-10 years when you have no idea what these tariffs will be next week, let alone in five years? This is a contradiction in many defences of these tariffs: if Trump is just negotiating to try to get a deal, and all this will look different in a week, what plans can you make? And third, go back to your high school economics class about Riccardo and comparative advantage: you’ll invest all of this time and money to get more expensive, less efficient production.
Meanwhile, imagine you’re a hardware startup that grew up in a world where the Shenzhen electronics cluster provided on-demand contract manufacturing to turn out your product on demand? Before, you were worried about the Chinese government - now your own government has added 54% to your BOM.
Then there’s Meta. Shein and Temu (see above) got a lot of their customers with aggressive advertising on Meta’s properties, as indeed did TikTok. About 10% of Meta’s ad revenue comes from Chinese companies - however, it’s not clear how much is e-commerce and how much is games and video (which are not taxed, yes). And then, a third to a half of Amazon Marketplace (which is 60% or more of total volume) is Chinese sellers. How price-elastic is that?
A third thing to wonder is data centres, where Google, Microsoft, Amazon and Meta have said they plan to spend something in the region of $300bn this year. At least three-quarters of that is in equipment, and most of that is imported. Of course, traditionally most data centres are distributed to be close to users, so they’re not all in the USA, but a lot are, especially as they’re focused on training (where having the compute all in one place is important, at least for now). Trump’s tariff announcement exempted ‘semiconductors’, but does that include the highly complex custom computing systems that Nvidia sells? This week? Next week? What about the transformers and gas turbines, and the HVAC, and the new liquid cooling? How much of that is made in the USA, or could be quickly? It would be ironic if Trump’s tariffs pushed big tech companies to move their AI development out of the USA.
These are just a few initial and obvious places to think about, but there will be a lot more: this is a very blunt change to an enormous number of complex systems. As I suggested above, every company that sells to consumers or enterprises is now thinking about a recession and inflation, and that will change how they think about budgets. Any company that imports, well, anything, now at a minimum has to spend a lot of time thinking about ways to avoid these tariffs, or just waiting them out, or wondering how much it can put up prices, and that inherently means doing things that are more expensive and less efficient than you’d be doing without the tariffs - or you’d already be doing them. Mario Draghi, author of the EU’s call to arms for more growth and less bureaucracy, called the EU’s regulation and especially its tech regulation as the equivalent of a giant tariff that the EU had changed itself, but all of this actually is a giant tariff that the US has charged on itself, and on the tech industry as much as anything else.