TL;DR
One AI lab, OpenAI, is turning into a real business with ~$70B in recurring revenue while Anthropic is asking for a $2T IPO and committing to half-a-trillion in compute on spec. GPU and infra power is consolidating around Nvidia, AMD, and even SpaceX, just as governments start buying frontier models for citizen services and consumers push back on AI-fueled dynamic pricing and behavioral targeting.
The big bet now is how much you lock into this concentrated stack versus keeping the freedom to walk when the economics or politics turn.
Key Events
Report
Frontier AI has split into two games: one actually throws off cash, the other burns it in hope of owning the future. The tension between those models is now the main risk variable for anyone wiring real money into this stack.
Anthropic is asking public markets for a $2T valuation while telling them to expect about $4.6B of 2025 revenue and a $42B net loss.
It is on the hook for roughly $518B of cloud and compute commitments, including an $84.5B deal with SpaceX. Around a quarter of its revenue comes from just two customers, and commentators openly question whether this is late-cycle bubble behavior.
In contrast, OpenAI is nearing $70B in ARR, with B2B sales more than doubling since July and consumer AI spend hitting $40B across 2B users, even as skeptics ask where actual profits are.
At DevDay, OpenAI rolled out GPT‑6.1 Sol, marketed as near‑Astra intelligence at about a fifth of Astra’s price and roughly half the cost of Sonnet 5.5.
Sol also comes with a 95% cache read discount and is reported to be about 30x cheaper than Sonnet 5.5 on some workloads. Alongside this, OpenAI re-opened its $200 Pro plan with new multipliers (Plus 1X, Pro 100 5X, Pro 200 10X) and a roughly 50% cut in included API spend, which some users experience as an effective jump toward $500 for similar usage.
Always-on Dots agents and the Decision API put OpenAI in the business of running persistent workflows, while mid-cycle term changes and the cancellation of Astra 6.1 over safety issues are driving some users to explore Anthropic, GLM, and open-source stacks.
compute suppliers: Nvidia’s chokehold, AMD’s breakout, SpaceX as dark horse Nvidia is reinforcing its central position with the Open Agent Safety Platform and OpenShell runtime, backed by more than a hundred partners, while its GPUs power OpenAI’s latest GPT‑6.1 models.
It is also leaning into financial engineering via a $150B stock buyback and is using insurers to offload some AI build-out risk, even as commentators worry about valuation and market concentration.
AMD answered with an $8.2B acquisition of World Labs, a post‑OpenAI‑deal stock pop of about 25%, and a run to a $1T+ market cap, plus kernel-level work that boosts Radeon iGPU AI performance by roughly 18–23%.
Meanwhile, SpaceX’s $84.5B compute contract with Anthropic effectively turns Starlink/Starship capacity into an AI cloud business, at the same time as the broader sector is told it needs $6T a year in revenue by 2031 just to justify the data center build-out.
The U.S. stood up America.gov, an AI front door to federal services powered by Grok and Gemini, and launched a separate Copilot-style chatbot, signaling willingness to rely on commercial frontier models for citizen-facing work.
At a White House lunch, Trump rebranded AI as super intelligence, claimed a U.S. lead, and signed an accord with lab leaders that leans on voluntary guardrails and self-regulation.
At the same time, Cruz blocked an AI safety bill, Khanna is drafting a ban on recursive tech until safeguards exist, and House Speaker Johnson is talking up voluntary rules while Congress mostly stalls.
Outside the U.S., the Netherlands is dumping Microsoft for NixOS over sanction worries, the UK is under pressure to deregulate for AI, and Australia just saw an OpenAI agent breach four government departments, including Medicare, prompting a public apology.
McDonald’s is trialing AI-driven dynamic pricing on Big Macs, in parallel with a push to more digital-only ordering, and online sentiment is full of accusations of greed and vows to switch to other chains.
DraftKings is using AI to flag chronic gamblers who are likelier to respond to promos and place losing bets, while carefully branding this as entertainment rather than gambling, which is drawing heavy criticism for exploiting vulnerable users.
Amazon’s own dynamic pricing and its noisy delivery drones are generating localized backlash, with some users explicitly looking for ways to route around the platform.
Commenters are increasingly skeptical that AI rollouts aimed at juicing ARPU are sustainable, tying them to layoffs, opaque algorithms, and a broad sense that companies are out over their skis on AI economics.
What This Means
The live trade is between doubling down on a concentrated, high-priced AI stack (labs, GPUs, data centers) and spreading exposure across multiple platforms while public and political tolerance for exploitative use cases visibly thins. The decision is whether you treat frontier AI as core infrastructure you must pre-pay for, or as a volatile input where you demand flexibility, transparency, and the option to walk.
On Watch
Interesting
We processed 10,000+ comments and posts to generate this report.
AI-generated content. Verify critical information independently.
Sources
Key Events
On Watch
Interesting