KPMG's Q3 pulse of 2,100 senior leaders splits the enterprise world into two camps — companies still experimenting with AI and companies already banking ROI — and the gap between them reads like a roadmap: harness layers, model routers, sovereignty strategies, and a management layer that treats AI economics as seriously as AI capability. Plus Meta's best month since 2022, Anthropic's pre-Thanksgiving IPO sprint, and TPUs in orbit.
The best business AI users aren't buying better models — they're building better management layers.
KPMG's Q3 pulse survey of 2,100 senior leaders across 20 countries breaks organizations out by where they sit on the AI journey, and the distance between the experimenters and the established-ROI companies maps where everyone else is headed: 86% of mature organizations run a formal AI harness layer, more than half have enterprise-wide sovereignty strategies and C-suite accountability for AI-informed decisions, and nearly a quarter have built model routers. Use cases are shifting from pure efficiency toward revenue, cost control is maturing into economic management — and despite all the new discipline, planned spend is still going up.
Meta stock rose 27% in September — its best single month since the November 2022 start of the year of efficiency — even with a 5% slide this week on news of an OpenAI competitor to Muse. The agent's early success has added roughly half a trillion dollars in market cap and, more importantly, convinced the market that Meta has a viable AI strategy.
Needham's Laura Martin credits Meta with pivoting from the metaverse to personal agentic AI but holds her rating, calling Meta 'notoriously slow at monetizing new products' — Muse is free for all but power users, and user data is segregated from the ad business. The telling shift: six months ago investors questioned Meta's basic AI competence; now they're questioning its ability to cash in on a successful bet.
Bloomberg reports Anthropic aims to start marketing its IPO the week of November 9, with a roughly two-week roadshow pointing to a first trading day around Thanksgiving week, plus an institutional investor day on October 14. It's expected to be the largest IPO in history, topping SpaceX's $75 billion raise, with potential investors reportedly buying the $1.8–2 trillion target valuation.
The leaked S1 shows an $8 billion operating loss on $4.6 billion of 2025 revenue, and the doomsaying followed. NLW's strongly held position: with more than 10x growth in 2026, investors won't price this deal on last year's numbers — if the IPO underperforms, the 2025 operating loss won't be the reason.
— President Trump, to Time Magazine. After a two-hour dinner with Dario Amodei ahead of this week's gathering of tech leaders, Trump told Time the Anthropic CEO's views are 'much different than what is portrayed in the media' — suggesting much of the White House–Anthropic acrimony has been driven by staffers rather than the president himself.
Per Time, Trump spent hours talking to Grok after a December meeting with Elon Musk — including asking how Venezuelans would react if the US captured Maduro. The chatbot said many would likely celebrate; after the mission, celebrations broke out, and officials say Trump came away thinking Grok was ingenious. Commenters immediately wondered what else LLM advice might be shaping.
A Falcon 9 placed a satellite carrying four Google TPUs into orbit — the first launch under Project Suncatcher, Google's orbital data center collaboration with SpaceX and Planet Labs. The chips will run in 15-minute bursts as a stress test; next comes a pair of larger satellites, then plans for an 80-TPU-satellite network and a mega satellite the length of a soccer field, all hinging on Starship.
OpenAI says the three violated policies on handling sensitive company information; sources told The Information it involved sharing data with an outside AI evaluations organization. Reactions split along predictable lines — some see retribution against whistleblowers losing key chain-of-thought monitoring researchers, others say safety work isn't a moral exemption from confidentiality rules.
From the outside it's impossible to judge whether OpenAI was justified, because we don't know what was leaked — and interpretations will mostly track what people already believe about AI safety. The point everyone should agree on: this is exactly why formal channels for reporting and third-party evaluation need to be up and running as soon as possible.
Anthropic users report queries being routed to a model with an updated knowledge cutoff and slick new design taste, with early chatter pointing to serious gains in design, 3D work, and spatial intelligence. The funny part: Opus 5.5 is already such a monster that the jump to Fable may feel less dramatic than it really is.
Even solopreneurs and freelancers should care where big companies are on the adoption curve: many interact with enterprises as service providers, and enterprise behavior shapes how AI itself evolves. If Wall Street decides volume-up-spend-down is the permanent trend, it changes how much investors are willing to backstop the infrastructure build-out.
RAMP's latest AI Index shows companies growing AI usage while paying less for it. Notably, it's not an open source story — open models remain under 5% of business spend. Per RAMP lead economist Eric Karazian, the decline is driven almost exclusively by price competition between OpenAI and Anthropic.
OpenAI's new enterprise marketplace lets companies apply spend commitments to open models sold through OpenAI, not just OpenAI tokens. Enterprises get the multi-model flexibility and cheaper open weights they increasingly want; OpenAI keeps the lock-in and the spend inside its ecosystem even when it flows to other token merchants.
The Q3 AI Pulse survey of 2,100+ senior leaders across 20 countries now breaks results out by where organizations sit on the AI journey. The gap between companies still experimenting and those with established ROI reads like a map of where business AI users in general will go over the next several months.
Among organizations with established AI ROI, 48% report significant employee adoption of AI agents, versus just 15% of organizations still in the experimentation phase. 2026 was the year agents became real — but the enterprise is experiencing that reality very unevenly.
Mature organizations maintain a formal interface through which their people interact with AI — a 55-point gap over experimenters, 31% of whom have one. The cyber gap is even starker: 58% of established-ROI organizations are deploying AI-assisted cyber defense versus just 8% of experimenters.
53% of established-ROI organizations report having an enterprise-wide sovereignty strategy — versus just 8% of experimenters. The questions around handing data to the model labs, and the more complex architectures that avoid doing so, are moving from podcast discussion to formal enterprise policy.
KPMG's summary of the shift: what matters now is the layer that makes AI work inside the confines and context of the organization. 53% of mature orgs place accountability for AI-informed decisions at the C-suite, 23% have built model routing capability, and across every dimension — from scaling to driving adoption — maturity correlates with formal, enterprise-wide approaches.
Productivity as a key AI goal fell from 42% of organizations in Q1 to 37% in Q3. Gaining ground instead: human-AI collaboration (+4 points), responsible AI, governance, trust and security (+4), and adaptability and resilience (+3). Operating priorities are becoming what AI is for.
A quarter of organizations are developing multi-agent systems and a fifth are orchestrating multiple agents. Per KPMG, revenue-focused agent strategies have risen since Q1 while efficiency-focused strategies declined — and most organizations now pursue the two together rather than trading one off against the other.
Cost visibility is now widespread — 77% of established-ROI orgs have AI cost monitoring dashboards (vs 43% of experimenters), 73% run cost reviews at the point of AI approval (vs 50%), and 46% have usage or token budgets (vs 31%). The next step, per KPMG: linking cost to value, which 48% of mature organizations now do consistently.
One of NLW's standing soapboxes: being overly restrictive with usage or token budgets during the experimentation phase can be fairly limiting — though it's hard to judge individual programs without seeing what organizations actually count as a budget.
For anyone worried that the new focus on cost efficiency means decreased investment: KPMG found average planned AI investment over the next 12 months jumped from $186 million in Q1 to $210 million in Q3. NLW's view — we're still barely scratching the surface of what we will ultimately spend on intelligence.