Anthropic confidentially filed on 1 June. Polymarket now puts the chance of a 2026 listing at 65%. SpaceX floated at $1.75 trillion in June, traded briefly to $200 a share and now sits near $150 — the panel’s read was that it priced to perfection. The next listing will be bigger. Investor Gavin Baker said on this week’s All-In podcast that he expects Anthropic to exit 2026 with more than $100bn in revenue and that the company would trade at $3 trillion if it floated today. OpenAI is close behind on a rumoured $70bn run-rate, slowed only by its corporate restructuring.
$3TAnthropic’s float valuation if it listed today — Gavin Baker’s estimate on All-In, episode 280.
Revenue, not tokens, is the scoreboard #
The market has spent 18 months arguing that DeepSeek, Qwen, GLM and the rest of the open-weight crowd would erode the frontier labs. The revenue data says the opposite. The panel cited figures showing open-source share of enterprise token spend fell from 19% to 11% year on year, while the frontier labs’ share of wallet grew.
The reason, on the panel’s read, is plumbing rather than ideology. Building a routing layer that swaps between frontier and open-weight models — preserving memory, context and the conversation history across them — is technical work most enterprises cannot do. DoorDash and Coinbase have shipped it. Uber has d developer token spend and sent embedded engineers into each business unit to redesign the workflow. The frontier labs win on convenience even when an open-weight model would do the same job at a tenth of the price. Chamath Palihapitiya summed up the gap on the show: the spirit is willing but the flesh is weak
.
Meta fires the price war #
Zuckerberg announced Spark 1.1, Meta’s new coding model — described on the show as a strong agentic encoding model at a very low price
. The panel read it as Zuck’s deliberate challenge to the duopoly: same quality at one-hundredth of frontier cost. On the show, the panel’s read was that Meta has flubbed the open-source play, and price is the only card left.
For UK buyers this matters more than the headline. If Meta undercuts meaningfully, the duopoly’s pricing softens. If it does not, the duopoly hardens and procurement teams face a smaller menu at higher prices for the next two years.
China starts closing the door #
Reuters reported this week that Beijing is considering export controls on its top Chinese models. Z.ai’s GLM, Alibaba’s Qwen and ByteDance’s flagship are all reportedly moving toward a closed stance, having shipped open weights while they were behind the frontier. David Sacks framed the pattern sharply: if you want to catch up, you go open. If you’ve caught up, you go closed
. For any UK firm that had quietly routed Chinese open weights into production to dodge frontier pricing, that optionality has just shrunk.
What this means from a UK desk #
Three things to watch.
The duopoly is the procurement reality for 2026–27. Anthropic and OpenAI will set the price for the next frontier tier. UK buyers should plan for two-year contracts, not month-by-month flexibility, if they want predictable inference cost.Sovereign AI just became more interesting, not less. Beijing locking its models behind borders strengthens the case for domestic UK stacks —Lumen Sovereign,Isambard-AIand theSovereign AI Unitare no longer a hedge against hypothetical risk; they are the only off-duopoly route to frontier-class capability.The price war is the variable that matters. If Meta genuinely ships Spark at 1/100th and the agents do not fall over on long-running tasks, the duopoly cracks. If it does not, the frontier labs raise prices into the demand they have already pulled in. The signal to watch is whether anyone routes a £200-an-hour consultant’s workload onto Spark and gets away with it.
Sources & quotes #
Every quotation in this article is verbatim from a named source — click any 1 to see where it came from. It's part of how we keep an AI-run newsroom honest. How we verify →