Read the report in its own context
State of Markets is a 90-page September 2026 presentation by a16z Growth, the growth-investing team at Andreessen Horowitz. It combines market analysis, third-party datasets, company disclosures and forecasts. The material was supplied for USD2’s strategy development.
Its sections span public markets, the AI platform shift, the software and hardware supply chain, private markets and broader technology diffusion. USD2 draws operating lessons from that context. Public-company multiples, hyperscaler investment forecasts and venture-market observations are not forecasts of Fund 1 returns or WAB earnings.
Adoption is widespread; sustained measurement is rarer
Page 27 reproduces Apollo analysis of S&P 500 companies’ AI disclosures. For Q2 2026, 69% pointed to live deployment with usage or adoption statistics, while 2% disclosed a metric tracked over time. These are disclosure categories, not mutually exclusive groups or a measurement of all firms’ internal activity.
The useful inference for USD2 is to make ongoing measurement part of every workflow from inception: the baseline, verified outcome, operating cost and error rate. Reporting that AI is in use is a weaker test than showing whether the institution performs its work better.
INDEPENDENT RESEARCH · Q2 2026 · State of Markets, September 2026
AI disclosure: adoption and measurement
S&P 500 companies in the Apollo disclosure analysis reproduced by a16z. Categories overlap.
| Measure | % of S&P 500 companies |
|---|---|
| Stated AI plan or goal | 74 |
| Live deployment / usage | 69 |
| Quantified result | 29 |
| Metric tracked over time | 2 |
| Broken out and tracked | 0 |
S&P 500 company universe. Reported disclosure categories, not mutually exclusive shares or a test of actual AI capability. Transcribed from page 27; underlying data not independently re-estimated. Zero disclosed in a category does not imply zero internal measurement.
a16z Growth · State of Markets · p. 27 · citing Apollo Daily Spark, 11 September 2026Lean staffing is a design objective to validate
Page 31 associates high-intensity AI adoption with increased entry-level headcount share relative to not-yet adopters. It also shows demand for technical roles. The underlying observations do not establish that AI causes either higher or lower total staffing.
For USD2 and WAB, the objective is greater operating capacity per responsible person. The actual team must support management, customer relationships, model development, risk, compliance and independent challenge. A fixed headcount or percentage saving cannot be inferred from this report.
Develop a portfolio of proprietary capabilities
Pages 33–37 describe differences in usage intensity, caching and the mix of frontier and other models. The commercial implication is to route each task to a capability that meets its quality and cost requirements, while retaining a coherent institutional workflow.
USD2’s direction is an owned internal platform and proprietary domain models. Documentary extraction, entity matching, trade analysis and risk surveillance need their own training rights, evaluation sets and operational limits. High model usage by itself does not establish productive output.
Company examples point to tests, rather than promised results
Page 28 collects company statements on underwriting, claims operations and cost efficiency, among other applications. They describe different populations, products and metrics, often without the information needed to reproduce the result.
For WAB, those examples motivate a comparison against its own workflow baseline. Credit evaluation requires out-of-sample testing, subsequent repayment outcomes and customer-treatment review. Service automation requires measured corrections, escalation and complete costs. Another institution’s percentage improvement is not adopted as a WAB forecast.
Build institutional advantage through execution
The report’s software discussion distinguishes businesses through profitability and defensibility. USD2’s inference is that proprietary models become more useful when connected to lawful domain data, operating records, tested processes and durable commercial relationships.
WFA’s relationships and QAF’s trade programme provide the starting commercial context. The internal intelligence system aims to turn that context into accurate records, accountable decisions and repeatable servicing. It remains in development and is not represented as an existing source of earnings.
Measurement notes
The chart transcribes the Q2 2026 blue bars on page 27: stated AI plan or goal 74%; live deployment with usage or adoption statistics 69%; quantified result 29%; tracked metric 2%; broken out and tracked 0%. The report credits Apollo Daily Spark, 11 September 2026. This is a reproduction of the supplied report’s figures, not an independently re-estimated dataset.
The five categories overlap. A zero in the final disclosure category does not mean companies have no internal measurement. Other report charts use different samples and definitions; they should not be combined into an aggregate banking or employment forecast.
READ THE SOURCE
Publications and reference material
a16z Growth · State of Markets · September 2026 · supplied 90-page report
AMCM · Technology and Cyber Risk Management · 017/B/2023AMCM · Banking Outsourcing Guideline · 020/B/2023FSB · Responsible adoption of AI · June 2026 consultationNIST · Generative AI Risk Management Profile · July 2024