Source: six episodes of The AI Daily Brief (raw/What_the_Best_Business_AI_Users_Are_Doing_Different.md, raw/Consumers_Will_Never_Pay_for_AI.md, raw/AI_s_First_Full_Field-Level_Disruption.md, raw/The_Best_New_AI_Products_and_the_Key_Trends_They_Represent.md, raw/The_5_Debates_Shaping_AI.md, raw/Where_Gemini_4_Argon_Fits.md) and The Epoch Brief of 2026-10-08 (raw/newsletter-epoch-ai-2c141f4d88.md).
Creator: Nathaniel Whittemore (NLW), The AI Daily Brief URL: https://www.youtube.com/watch?v=Rx7G-8myrd8 (KPMG), https://www.youtube.com/watch?v=J3n4IUwmuOM (a16z), https://www.youtube.com/watch?v=GoTIt8aPG0Y (The Information), https://www.youtube.com/watch?v=thO4Y5RYfHo (Ramp), https://www.youtube.com/watch?v=w7SmNJxUHLc (usage concentration), https://www.youtube.com/watch?v=SX9LQD3yINg (Character AI) Platform: YouTube (podcast) Newsletter: Epoch AI, “The Epoch Brief - October 8, 2026”, https://epochai.substack.com/p/the-epoch-brief-october-8-2026
Four adoption datasets arrived in October 2026: KPMG’s Q3 AI Pulse on enterprises, a16z’s seventh Top-100 consumer apps list, Ramp’s spend data, and The Information’s subscriber survey. Epoch and Ipsos added US usage-frequency polling. Read together, they show a gap between organisations that have built a management layer around AI and those still experimenting, and a power-law split in who actually pays. Every figure except Epoch’s comes through NLW reading the report on air, so it is secondary. Use it as benchmark data for client conversations, not as audited statistics.
Key Takeaways
- The ROI gap is about management, not models. In KPMG’s Q3 Pulse, organisations with established ROI are far ahead on agent adoption (48% vs 15%), a formal “AI harness layer” (86% vs 31%) and cost dashboards (77% vs 43%).
- Spend falls while volume rises. Ramp: token spend fell 5.2% week over week while token volume rose. Ramp’s economist attributes the drop to OpenAI–Anthropic competition, not open-source models, which “remain less than 5% of business spend”.
- Revenue is extremely concentrated. Ramp: 80% of OpenAI’s and Anthropic’s enterprise revenue comes from 1% of their customers. a16z: the top 1% of consumer AI payers spend 25 median.
- Few consumers pay. Only 2.2% of US households had a paid AI subscription (PNC, April 2026), while more than two-thirds of Americans use AI weekly.
- Daily use more than doubled. Epoch/Ipsos: US adults using AI 6–7 days in the prior week rose from 8% (March) to 19% (August 2026).
- Claude is strong with paying, technical users. a16z: Claude caught Gemini on paid US subscribers and converts users to higher tiers better. The Information’s readers: Claude Code 45%, Codex 33%, Cursor 23%.
- Open weights are used widely, Chinese ones much less. The Information: 72% use open-weight models regularly and 20% run most workloads on them, but only 21% use Chinese open weights.
KPMG Q3 2026 AI Pulse
More than 2,100 senior leaders in 20 countries. KPMG now splits results into organisations with established ROI and those experimenting.
| Measure | Established ROI | Experimenting |
|---|---|---|
| Significant employee adoption of agents | 48% | 15% |
| AI-assisted cyber defence | 58% | 8% |
| Formal “AI harness layer” | 86% | 31% |
| Enterprise-wide sovereignty strategy | 53% | 8% |
| AI cost-monitoring dashboards | 77% | 43% |
| Cost review at approval | 73% | 50% |
| Usage or token budgets | 46% | 31% |
Other findings:
- 53% place accountability for AI-informed decisions at C-suite level, and 23% have built model routing.
- Among established-ROI organisations, 48% consistently assess value against cost. KPMG calls this a move from “cost control” to “AI economic management”.
- Productivity as a key goal fell from 42% (Q1) to 37% (Q3). Human–AI collaboration and responsible AI/governance each rose four points.
- “A quarter of organizations report developing multi-agent systems and a fifth are orchestrating multiple agents”. Revenue-focused agent strategies rose since Q1 while efficiency-focused ones declined.
- Average planned AI investment for the next 12 months rose from 210M (Q3).
- NLW’s caveat (opinion): tight token budgets during experimentation can be limiting, and it is unclear what respondents count as a budget.
a16z Top-100 Consumer AI Apps, 7th edition
- First revenue ranking (top 50 by monthly revenue, using Yipit data), alongside the usual web and mobile traffic rankings. Only 11 debuts, the fewest across seven editions.
- Consumer AI’s first real market is people paying for tools to build, create and get work done.
- Claude caught Gemini on paid US subscribers during 2026 and does better at converting users to higher-cost tiers. ChatGPT beat Claude on net new subscriptions in July and August, reversing Claude’s lead earlier in the year.
- Where startups still win: own a differentiated model (Suno, ElevenLabs), own a multi-model experience (Cursor, OpenRouter), or serve a narrow audience (OpenEvidence for physicians, Venice for privacy-focused users).
- The top 1% of AI spenders outspend the bottom 50% combined: 25 median. They buy build, create and work tools at much higher rates, for example Higgsfield at 15× and Runway at 10× the average purchase rate.
- Categories with no AI app in the top 100: streaming and media, retail marketplaces, social and messaging, travel, personal finance, jobs, real estate, gaming and dating.
Ramp
- Spend vs volume: token spend fell 5.2% week over week while volume rose. Ramp economist Ara Kharazian (captioned “Eric Kerazian”; spelling normalised, unverified): open-source models “remain less than 5% of business spend”, and the decline is “driven almost exclusively by competition between OpenAI and Anthropic”.
- October fastest-growing vendors: Jev (TypeSafe) ranked first on growth relative to size and gained “a full point of market share within a month”, categorised as a foundation LLM. Anthropic, SpaceX AI and OpenAI were all in the top five for market-share gain. DeepSeek, OpenRouter and Featherless also appeared. Runway and Higgsfield made the market-share growth list, which Kharazian says “will push AI spend up”. Ramp’s own caveat: its customers are early adopters.
- Concentration: “80% of OpenAI and Anthropic’s enterprise revenue comes from just 1% of their customers.”
Ramp update, late October (added 2026-11-01)
- As relayed on The AI Daily Brief, Ramp’s newer data shows the top spenders roughly tripled AI spend per employee since the summer. Businesses in the top 10% (not the top 1%) went from about 800 per employee, growing faster than the median.
- The top of the chart is startups spending around $9,000 per employee per month.
- Source:
raw/The_Most_Concerning_AI_Bubble_Signals.md. How this concentration feeds bubble worries: AI Bubble Signals, Late October 2026.
The Information subscriber survey
Readers are tech-forward: two-thirds work at organisations that build AI applications.
- Returns: 35% say AI returns are “multiples” of spend, 17% positive but below hopes, 30% too early to tell, 8% neutral, 8% net negative.
- Vendors: OpenAI is back on top after Claude led in the summer survey. ChatGPT, Claude and Gemini are used by 65–78% of readers. Grok, Microsoft Copilot and Perplexity are at 25–29%.
- Agents: OpenClaw use is 12%, its high for the year. Claude Code 45%, Codex 33%, and Cursor doubled to 23%. A year earlier, 59% had never used an agent.
- Open weights: 72% use them regularly; 20% say they drive most workloads, 10% about even, 42% a minority. Only 21% use Chinese open-weight models.
- Hiring: 22% hire less, 10% more, 21% too early to tell; almost half say AI has not changed hiring.
US usage and consumer economics
- Frequency (Epoch/Ipsos): near-daily use (6–7 days in the prior week) rose from 8% in March to 19% in August 2026. One-day-a-week use fell from 17% to 10%. NLW cites the same 8% → 19% figure and says more than two-thirds of Americans now use AI weekly.
- Messages (Epoch ChatGPT Usage explorer): 8.3M messages from 5,000 opt-in YouGov panelists. Median monthly messages per active user rose from 14 (January 2023) to 36 (December 2025). The most active 10% sent 63% of messages. The data is unweighted.
- Paying households: 2.2% of US households (PNC, April). Menlo Ventures estimates global consumer AI spend of $40B in 2026, more than 3× 2025.
- Power users in coding: Cursor reported that the top 10% of users accounted for nearly two-thirds of tokens in the prior month (early September).
- A consumer-economics warning: Character AI filed for bankruptcy. The CEO’s affidavit says it spent 2.8M a month.
Try It
- Benchmark a client against the KPMG table. Ask which of the seven measures they have. A harness layer, cost dashboard and cost review at approval are the cheapest gaps to close.
- Track spend and volume separately. Ramp shows they can move in opposite directions. Report cost per task or per outcome, not total token spend. ^[inferred] See Cost and Intelligence Levers.
- Design for the paying 1%. If you sell an AI product to consumers, the a16z and Cursor data say a small group of power users carries the revenue. Price and support for them first.
- Don’t assume buyers use Chinese open weights. Only 21% of a tech-forward audience does, against 72% for open weights generally. Offer a US or EU-hosted open-weight option. ^[inferred] See Chinese Models Decision Framework.
Open Questions
- All KPMG, a16z, Ramp and Information figures are NLW’s on-air readings. None of the primary reports was fetched. KPMG’s definition of “established ROI” and its sample split are not given.
- Ramp’s economist is named from auto-captions; the spelling is normalised and unverified.
- Ramp’s 80%/1% figure is described as published “recently”. The date and method were not given on air.
- Sample bias. Ramp’s customers and The Information’s readers are both early adopters, so their figures overstate the general market.
- The 2.2% household figure dates from April 2026, before the summer’s agent launches.
Related
- WalkMe State of Digital Adoption 2026 — earlier enterprise adoption benchmark
- How Americans View AI (Pew) — public attitudes, with later polling waves
- OpenAI Enterprise Signals — OpenAI’s own frontier-firm usage data
- How Work Is Changing Inside Anthropic — the same month’s first-party data from inside one lab
- Anthropic’s Position — customer concentration and the Microsoft/Meta spend cuts
- Admin Spend Visibility and Control — how to build the cost dashboards and budgets KPMG measures
- Cost and Intelligence Levers — token budgets and routing in practice
- Chinese Models Decision Framework — token share vs spend share