What Caught My Attention #9
17 April 2026
Back again.
Friday in London, although I have spent the last two weekends bouncing between here and Edinburgh. Work-wise, Anthropic shipped Claude Opus 4.7 this week alongside a fresh stack of tools, and I have been living inside Claude Cowork putting the new model through its paces on client work. The step change in coding and agentic task completion is great. Anyway, plenty happened this week. Here is what stuck with me.
The French have worked out what the Australians could not.
The big one first. Roland-Garros has confirmed wearables are in, the first Grand Slam to allow them, and WHOOP took an immediate bow. Three months ago the Australian Open was pulling bands off wrists like it was doing the sport a favour. The French have looked at the same question and reached the grown-up answer, which is that telling elite athletes they are forbidden from seeing their own biometrics is absurd. The timing could not have been scripted better, because Rory McIlroy won his second green jacket a couple of days earlier with the band on, and his numbers were across every timeline within hours. Heart rate of 135 on the 18th tee, 150 when the putt dropped, and a recovery score of 7% on Monday morning which tells you everything you need to know about Sunday night. I wear one and I like it, but set that aside. The thing I genuinely admire here is what modern sports marketing is supposed to look like. No agency deck, no brand safety memo, no campaign brief. A sponsored athlete wore the product, won the tournament, and let the data do the rest. Most brands would sell an organ for that weekend. The rest of the Slams fall in line by next year, and the question left on the table is which federation wants to be the one explaining why a wristband is a threat to the integrity of sport.
Britain is poorer than the poorest state in the USA.
Delusion sets in after the IEA survey this week. They asked Brits where the UK ranks in income per person against the fifty states. The average guess was 7th, wealthier than 43 of them. The real answer is 51st. Last. Below Mississippi, below Arkansas, below every single state in the union. Two-thirds of the country thinks we are heading in the wrong direction. I agree. This is the number that ought to end every other argument in British politics, and instead we are busy increasing dividend tax again and cutting writing-down allowances, with the OBR now forecasting 1.1% growth for the year. The OECD has told Reeves to overhaul the tax code. She will not. We have spent the best part of fifteen years litigating how to carve a pie that has stopped growing, and the result is that a median American in the poorest state in the country has more money in their pocket than the median person on this island. The country needs a government that treats growth as the point, rather than as something to apologise for. Until it gets one, the chart only gets worse.
The SaaS stock-comp reckoning.
Bill Gurley posted a long piece this week on stock-based compensation in SaaS, and it is the most sensible thing I have read about software businesses in a long time. His short version is that SBC was always a real cost, adjusted EBITDA was always a fiction, and the industry accepted the fiction because rates were low and growth was easy. Gurley says 95% of RSUs get sold on vest, which is the number that matters. Nobody holds them. Everyone treats them as cash, because they are cash, and the only people pretending otherwise were the CFOs presenting the deck. I wrote about a version of this a while ago in my thesis on business fundamentals, and the diagnosis has only hardened since then. If your operating model only works when share prices go up, you do not have an operating model, you have a favourable cycle. That arrangement is getting repriced in public now, and if I were advising a mid-tier SaaS board this week I would be pushing comp towards cash, trimming headcount to match, and eating six ugly months before the market notices you are the only one doing the arithmetic honestly. The tourists get exposed when the buyback tap slows, which it will.
Venture had its biggest quarter ever. AI ate it.
Relatedly, and not a coincidence. a16z’s charts of the week confirm Q1 2026 was the largest venture quarter on record, $300 billion into about 6,000 companies, up 150% year on year. Around 80% of that went to AI, and $188 billion of it went to four names: OpenAI, Anthropic, xAI and Waymo. Sixty-five per cent of the entire quarter concentrated in four companies. You can read this two ways and both are correct. One, venture is no longer a single asset class, it is a sovereign-scale bet on a handful of model labs plus a more normal market underneath. Two, there is no middle outcome from concentration this tight. Either one or two of those bets compound into the biggest companies ever built, or a crater opens and half the LP community decides venture was a mistake. I still think the infrastructure spend is broadly rational. I would not extend that opinion to every Series B that mentioned an agent in the pitch deck.
RFK unwinds the peptide ban.
Robert F. Kennedy announced HHS is reversing the Biden FDA’s 2023 decision that shoved a list of peptides into Category 2 and killed legitimate compounding access overnight. Five are already off the list and fourteen more are being reviewed by an FDA panel in July, including BPC-157, CJC-1295 and Ipamorelin. I have been getting a proper education on this in the last few months, because my work with SDBotox has taken me deep into their peptide subdivision and the commercial reality of a market that has been running on Telegram and dodgy websites since 2023. Pushing compounding back into regulated pharmacies with doctors in the loop is the less reckless policy, whatever you think of the underlying science. It is also a commercial event. Longevity clinics, wellness brands and the compounding pharmacy stack are about to get formally legitimised, and every operator in the space now has until July to decide whether they are professionalising up or getting bought. The smart money is already moving, and the brands that were hedging on the grey-market side are going to find themselves outrun by regulated incumbents who were quietly waiting for the green light.
Allbirds wants you to think it is an AI company now.
Josh Kale laid out the full arc on X. Allbirds has absolutely pivoted. It has just pivoted in the least serious way I have seen since 1999. The business flogged its IP and operations for $39 million to American Exchange Group, rebranded as NewBird AI, announced a $50 million raise, and declared its new purpose is to buy GPUs and lease them out. The stock ran 600% in a day. There is no GPU team. No data centre experience. No customers. Four hundred million in accumulated losses and a Nasdaq delisting warning sitting in the top drawer. The 1999 version of this was tacking .com onto the end of a brand and watching the market lose its mind. The 2026 version is tacking AI onto the front and doing the same thing. Some of these punts will look clever in five years. Most of them are a polite exit door for existing shareholders dressed up as a growth story. If you bought NewBird at $17, good luck, and get a stop-loss.
Bieber cut out the agent and cleared ten million.
Coachella paid Justin Bieber a reported $10 million to headline last weekend, making him the highest-paid headliner in the festival’s history and beating Beyoncé’s $8 million from 2018. He sealed the deal directly with Goldenvoice without an agent, which means he keeps a much bigger chunk of the cheque than any previous headliner. His music got streamed 24.6 million times in the US the day after his set, a 54% jump. And Hailey used the moment to launch the first Rhode x The Biebers collaboration, pimple patches, because of course. There is a business lesson sitting in all of that. If an artist has enough leverage, the agent, the platform and the retailer are all optional, and the margin flows back to them. Bieber rebuilt that leverage through eighteen months of a public mental health reset and a stripped-back live product that gave the press nothing to write about except the music. The result is a higher margin on a bigger cheque than any headliner before him, plus a fully-owned consumer brand picking up free distribution off the back of his set. Most marketing teams would spend a year’s budget trying to engineer that. Bieber engineered it on his own.
Also caught my attention.
Amazon Leo unveiled its aviation antenna this week. One gigabit down, 400 megabits up, installs in a day. Starlink Aviation currently advertises up to 310 Mbps down and 44 Mbps up per terminal, with installations taking 10 to 14 days. Starlink has over 1,400 commercial aircraft connected and real customers flying today. Amazon has a spec sheet, a Delta commitment for 500 aircraft from 2028, a JetBlue install from 2027, and roughly 240 of a planned 3,232 satellites in orbit. On paper the Leo antenna is the better product. In practice Starlink has a three-year head start, more satellites, and actual airlines. Amazon has logistics, AWS relationships and patience that SpaceX cannot match. The seatback experience for the next decade is going to get decided on this.
TSMC posted Q1 earnings with profit up 58% year on year and revenue of $35 billion. It raised 2026 guidance to more than 30% growth in dollar terms, said advanced chips were 75% of wafer revenue, and flagged that capacity is going to be tight for at least three more years because fabs take that long to stand up. If anyone was quietly wondering whether AI demand was cooling, it is not.
Sam Altman’s San Francisco house was hit with a Molotov cocktail last weekend. A 20-year-old drove from Texas with a list of AI executives, an anti-AI manifesto and a stated intent to kill him. He is being charged with attempted murder and federal terrorism offences. This is genuinely dark and it is not an isolated act. The backlash against AI is moving from op-eds to physical risk, and Silicon Valley is about to discover what finance has known for thirty years about executive protection. Expect a lot of private security budgets to get signed off in Q2.
Opposition to AI data centres has now blocked $18 billion of projects and delayed another $46 billion across 24 US states, mostly on power and water grounds. The ceiling on the AI build-out in the United States is no longer chips. It is planning permission, grid capacity and local politics, and that is a much harder problem to throw money at.
That is all for this one. Have a great weekend. Follow me on X, Instagram and LinkedIn.
Ben Walker