The AI safety letter that three of the largest model builders endorsed over the weekend is, in Matthew Bloxham's reading, the fastest route to any rules at all — because the alternative runs through Congress.
Investors spent two sessions selling on the idea that self-imposed guardrails would slow AI demand. Bloxham said the demand does not depend on the frontier, because the models enterprises already have are sophisticated enough for what they are building.
"I don't see any near-term slowdown in demand."
Bloxham covers technology, media and telecoms for Bloomberg Intelligence out of London, which puts him on both sides of the regulatory split the segment is about.
The full segment is covered here so you can skip it.
Here are the 6 calls that matter.
👤 Guest: Matthew Bloxham, Senior Tech Media Telecom Analyst at Bloomberg Intelligence, based in London
🎙️ Hosts: Paul Sweeney and Scarlet Fu, who anchor Bloomberg Intelligence's weekday program
📰 Published: 15 September 2026 on Bloomberg Intelligence (Bloomberg Podcasts)
🔴 YouTube | 🟢 Spotify | 🟣 Apple Podcasts | ⏱️ 8 min
Key Takeaways
Voluntary industry guidelines are the realistic near-term outcome, not legislation
He reads the weekend letter, endorsed by the three big players, as the start of exactly that
Anthropic has been profitable two quarters running and he thinks it still wants to list this year
Whether it does turns on whether the selling of the past two sessions carries on
A frontier slowdown would not cut AI demand, because the models already shipped are good enough
Money diverted from frontier research does not leave the sector — it goes to efficiency, power use and speed
Whatever US legislation eventually arrives will be light touch, on the evidence of US history
Europe is less enthusiastic about AI but also less inflamed, because its checks on data centers already exist
1. Guidelines Before Laws
Paul Sweeney set the segment up on the split in Washington: the people running the largest AI companies saying they should slow down, against political pressure to keep moving. He asked Bloxham how it plays out.
His answer for the short term is voluntary industry guidelines, and he thinks the industry is already heading there. The evidence he gave is the letter that came out over the weekend, which he said was endorsed by all three of the big main players — the quickest route to a standard they are all prepared to abide by.
Legislation is the slower and less reliable path. He said the US government might eventually agree something that sets tighter constraints on what is not permissible, but that legislation takes a long time to get through the process and does not always end up in a good place for the industry.
Scarlet Fu called that a very polite way of describing Congress getting anything done.
2. Anthropic's IPO Window
Fu noted that OpenAI has made clear its listing is a 2027 event rather than a 2026 one, and asked what that means for an Anthropic filing.
Bloxham said Anthropic has been quiet on timing, and gave his own read rather than a company line: he imagines they are still keen to go this year, flagging it as his view.
The supporting evidence he cited is Anthropic's own messaging through press channels that it was profitable for a second quarter in a row, which he takes as a sign the momentum is holding.
The gating factor is the tape, not the business. Listing schedules, he said, are incredibly sensitive to stock market sentiment, so it turns on whether the selling pressure of the past couple of days persists or the market bounces back.
The debate driving that selling is specifically about guardrails. He described an argument running that day and the day before over how far self-imposed restrictions would slow AI demand — which would feed straight through to Anthropic's revenue outlook.
3. No Slowdown in Demand
On that argument Bloxham took the other side, and the reason he gave is about installed capability rather than sentiment.
His own view is that there is no near-term slowdown in demand coming.
The mechanism is that enterprises are not waiting on the frontier. "The models that enterprises have at their disposal today are plenty sophisticated enough for them to deploy very sophisticated overlays to their businesses."
So the demand outlook, he said, is still going to look pretty robust — a claim that survives a pause at the leading edge because the buying is happening a generation behind it.
4. Where the Money Goes
Sweeney relayed what Dan Ives had said earlier that morning on the Bloomberg Surveillance program — that the money and the investment keep flowing with no slowdown — and asked whether that matches what the companies themselves are saying.
Bloxham agreed, with a hedge that it is still early days.
His larger point is that capital does not leave when the frontier slows, it moves sideways. If there is an agreed set of principles that slows leading-edge frontier model development, he said a lot of that investment finds its way into other parts of the models the companies already have, to make them more competitive and create differentiation.
The specific alternatives he named are efficiency measures rather than capability ones: power consumption, speed and other parameters, instead of pushing the boundaries of artificial intelligence forward.
There are, on his read, still lots of areas that the investment can go to.
5. Light Touch, Eventually
Fu asked what he will be listening and watching for next as the public outcry over AI safety builds.
What he is watching is the companies, not the politicians. His primary focus is on what the companies themselves say.
His near-term expectation stays the same: a voluntary industry framework the companies agree to abide by.
He expects an enormous amount of political rhetoric from every direction about what should or should not be done, and said nobody is expecting legislation to be imposed anytime soon.
When it does come, he expects it to be mild. His sense, from everything seen through the history of the US, is that whatever legislation arrives is going to be fairly light touch.
6. Europe's Calmer Mood
Sweeney's last question was about geography: AI is getting pushback in the US from the marketplace, from consumer groups and now from politics, and he asked how the mood compares in the UK and Europe.
Bloxham said the European mood is holding up pretty well, and that adoption is pretty good if less enthusiastic than in the US.
The difference he identified is that the constraints are already in place rather than newly demanded. Most European markets, he said, already have more checks and balances on things like where new data centers get built and the environmental impact of that.
The result is a slower pace of development along the environmentally and politically sensitive dimensions, and a debate that is not as magnified in Europe as it is in the US.
Bonus Insights
Sweeney opened the segment on the politics: President Trump telling AI companies to continue to spend and continue to invest, with Fu summarizing the message as "There's nothing to worry about. Move on, move on, move on."
Sweeney's own view is that the president has a point, because this is a competitive race between China and the West, so pulling back for any reason carries the risk of falling behind.
Fu's counter was the calendar: lawmakers are gearing up for the midterm elections and have to be seen listening to constituents, many of whom have concerns about what AI does to the economy, to their jobs and to their livelihoods.
Bloxham's bottom line is that the AI trade does not hinge on what happens at the frontier: the rules will arrive as voluntary industry guidelines long before they arrive as law, the models enterprises are already buying are good enough to keep revenue growing through a frontier pause, and money pulled back from leading-edge research gets spent on making the existing models cheaper and faster instead.
Products, Companies & Tools Mentioned
OpenAI (Has made clear its listing is a 2027 event rather than a 2026 one, which is the benchmark the hosts measured Anthropic against)
Anthropic (Quiet on IPO timing, but flagged through press channels that it was profitable for a second quarter in a row; he thinks it is still keen to list this year)
Books & Resources Mentioned
The AI safety letter published over the weekend (Endorsed by all three of the big main model builders; he reads it as the start of the voluntary industry framework he expects)
Watch the full episode:
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: