a16z State of Markets (September 2026): A Bilingual Digest

a16z argues that tech has outgrown its "cottage industry" origins and become the prime mover of the global economy, which it calls the "Everything Cycle": almost every company is now a tech company in some sense, and tech leads investment, earnings growth and margin expansion. It adds that this year's gains rest on earnings rather than price, multiples have compressed, and leadership has moved from the Mag 7 to the rest of tech (a16z p.5, p.10–12).

• a16z says AI infrastructure accounts for all or nearly all net new construction spending, arguably the economy's only pro-cyclical driver right now; its chart title says high-tech (equipment, software and R&D) is 55% of all U.S. capital spending, with the source dated May 5, 2026 (a16z p.5).
• According to a16z, among the 100 largest public companies ex-China, tech is about 35% by count and about 60% by market cap as of August 28, 2026; the market-cap share trends up since 2010. a16z notes its tech definition is broad and may not be exhaustive (a16z p.6).
• a16z holds that for at least a decade tech has taken the bulk of value creation and earnings growth; its chart shows tech behind about 76% of the 2026 increase in S&P 500 net income, with data as of August 28, 2026 (a16z p.7).
• According to a16z, the end point of S&P 500 quarterly margin (earnings over sales) is labeled 2Q26F at 17.5%; a16z calls margins historically wide, with nearly every sector expanding recently and tech the standout (a16z p.8).
• a16z says the overall market is hard to call a "bubble": new highs come from earnings while multiples compress. Its table (Duality Research, August 30, 2026) shows tech earnings per share up 55.7% year to date, price return +22.1% and P/E change -21.6% (a16z p.10–11).
• According to a16z, leveraged ETF assets were labeled $192 billion on August 28, 2026; a16z says that, unlike in the meme-stonk era, retail has piled into consensus themes, and leverage and momentum add volatility (a16z p.13).

a16z argues that AI is a generational platform shift: capex keeps outrunning forecasts, credit markets are stepping in to fund it, and for now returns remain well above the cost of borrowing.

• a16z cites Vanguard calculations from Bloomberg data (as of July 31, 2026): consensus 2026 capex for hyperscalers (Alphabet, Amazon, Meta, Microsoft, Oracle) is $777 billion, rising above $1 trillion a year from 2027. (a16z p.15)
• a16z's view is that forecasters under-shoot capex every year. Per the CapIQ consensus it cites (as of September 18, 2026), the 2028 forecast was close to $600 billion in December 2025 (chart reading) and above $1 trillion by September 2026. (a16z p.16)
• a16z expects the steep drop in hyperscaler free cash flow to last until about 2028, with credit markets stepping in. Per Bloomberg and JPMAM (as of August 31, 2026), new investment-grade debt and SPV financing by hyperscalers and Nvidia was about $200 billion in 2025 and above $300 billion in 2026 (chart readings). (a16z p.17–18)
• Per BofA Global Research (August 2026), cited by a16z, hyperscaler return on invested capital (ROIC) is about 20%, versus a weighted average cost of capital (WACC) near 10% and a cost of debt near 4% (chart readings). a16z notes these firms borrowed little until recently and now put the money into physical assets with steady, rising cash flows. (a16z p.19)
• a16z says the buildout has set off a scramble for skilled labor. Per data it cites, employment in data center-exposed construction is up more than 300,000 since 2022 (BLS, Goldman Sachs), and median posted base pay for data center facilities managers (Indeed, January to June 2026) is 64% above non-data center postings with the same title. (a16z p.20)
• a16z argues that data centers may be lowering power bills: a large, constant buyer of electricity absorbs much of the grid's fixed cost, leaving less for other customers. A study it cites (Watten, Bistline and Blanford, Aug. 24, 2026) estimates that each 10% rise in data center capacity went with an average drop of about 0.4% in residential retail electricity prices. (a16z p.21)

a16z argues that demand is validating this wave of AI capex: lab revenue, cloud backlogs, GPU rental prices and token usage all signal strong demand, while enterprise adoption remains shallow and early.

• Revenue: a16z's chart puts OpenAI and Anthropic revenue at $23 billion in 2025 and about $100 billion in 2026E, with a question mark on the second figure; lab data come from YipitData, public-software data from CapIQ. a16z calls this demand unlike anything seen before, and the chart title adds that the labs have added more revenue in 2026 than all public software (excluding clouds) combined. (a16z p.23)
• Backlog: a16z's chart of company filings stacks Microsoft RPO, GCS (Google Cloud) backlog and Amazon RPO to about $1.7 trillion in Q2 2026, estimated from the axis since no total is printed; a16z says backlogs have doubled year over year. a16z reads hyperscaler commentary as demand exceeding supply, and notes free cash flow is near negative for now but expected to rebound strongly in two years. (a16z p.24)
• GPU rental prices: Silicon Data (source dated September 7, 2026) shows the A100, the oldest generation on the chart, at $1.59 per GPU-hour at the chart's end point. a16z argues fears of fast GPU obsolescence are overstated: compute demand keeps accelerating, so even the oldest chips hold or gain rental rates and inferred residual value, and depreciation periods may even prove too short. (a16z p.26)
• Adoption: Apollo data cited by a16z (source dated September 11, 2026) show 69% of S&P 500 companies pointing to a live AI deployment in Q2 2026 (the chart's label adds usage or adoption stats), but only 2% disclosing a metric they track over time. a16z's view is that adoption reaches widely and matters, yet remains thin in depth; a second chart, from AlphaWise and Morgan Stanley Research, shows the share of companies reporting quantifiable AI impact trending up. (a16z p.27)
• Power users: in YipitData figures cited by a16z, the chart title says the top 1% of users by AI vendor spend outspend the top 10% by about 8x on a median basis. a16z argues growth is concentrated in the top 1% of users, whose usage sits orders of magnitude above the field, with a gap widening since early 2026, and expects the rest of the market to catch up. (a16z p.32–33)
• Agents: OpenRouter data cited by a16z carry a marker that agentic token usage passed human usage on February 6, 2026, and label the chart's latest point at 7.3 trillion tokens, on a 7-day-average basis. a16z argues demand keeps rising as intelligence gets cheaper, with agents driving parabolic token growth because routing, caching and other efficiency work makes them affordable, and calls this Jevons Paradox playing out live. (a16z p.32, p.34, p.35)

a16z argues that AI has set off a broad rotation from bits to atoms: earnings and capex are moving toward chips, power and manufacturing, then spreading to rockets, robots and defense. It adds that AI infrastructure is not the only driver, and that capital-heavy companies have recently reversed a decade of capital-light dominance in markets (a16z pp. 42-43, 52).

• a16z titles its chart “Hyperscaler FCF Has Become Semiconductor FCF.” Per BofA Investment Research data cited by a16z (as of Aug. 30, 2026), 12-month forward free cash flow for the semiconductor group (including Nvidia) reads above 400 billion on the chart (axis unit billions, currency unstated, an axis reading), while the hyperscaler group (including Microsoft and Oracle) has dropped below zero on the same measure (a16z p.43).
• a16z writes that infrastructure and industrial needs are massive, with about $90 trillion of global investment necessary by 2040, and uses this to argue the rotation goes beyond chips; the chart is sourced to JPMAM's Guide to Markets (Aug. 2026) (a16z p.45).
• a16z argues that electricity demand growth comes from more than data centers (left chart sourced to Bloomberg NEF). On the right, per S&P Global/RRA estimates cited by a16z (Apr. 23, 2026), annual utilities capex is projected to reach $277 billion in 2028, the highest year on the chart (a16z p.47).
• a16z's SensorTower chart of U.S. weekly active users shows Waymo with the longest line, running well above Tesla Robotaxi and Zoox in its later stretch. Another chart, per TechCrunch (source dated May 27, 2026), puts Waymo robotaxi rides at 500,000 per week, after paid service began in San Francisco in August 2023 and spread to more cities (a16z p.50).
• a16z says the AI supply chain has been the winning trade, with no bigger winner than memory. Per DRAMeXchange, TrendForce and Morgan Stanley Research data (source dated Aug. 20, 2026), the spot price of MLC 64Gb NAND stands at $36.59 at its latest reading after a steep rise. A Capital IQ chart on the same page (through Sep. 9, 2026) shows the Semis/Storage and AI Supply Chain baskets up year to date and ahead of the NASDAQ 100 and S&P 500; its footnote calls the universe and categorization illustrative (a16z p.53).
• a16z argues supply still lags demand: GPUs arrive on time, while other links in the chain face unusually long lead times, and powering the chips is an open question. A footnote citing electrical equipment trackers (Terrapin, DistroForge) puts average transformer lead times at about 128 weeks (a16z p.55).

a16z argues that the early-2026 “SaaSpocalypse” reads more like a “prove it” repricing: multiples contracted unevenly across software, while revenue growth and operating leverage show little visible damage so far (a16z p.57).

• Cyber and observability was the strongest software category this year. Per a16z (CapIQ, Aug. 31, 2026), its cap-weighted year-to-date return among $IGV constituents is about +85% (chart reading), while the other categories sit around or below zero (a16z p.58).
• Segments saw different multiple declines. Per a16z (JPMAM, Aug. 20, 2026), median EV/TTM revenue for horizontal software slid from 12.3x in 2020 to 2.7x in H1 2026, and vertical software from 14.2x to 4.6x. Per a16z (CapIQ, Sep. 18, 2026), median TEV/forward revenue for software companies growing 20%–40% came down from a peak near 28x around late 2020 to about 9.5x; both are chart readings (a16z p.59).
• Revenue growth cooled and then leveled off, and a16z reads operating leverage as mostly steady or better. Per a16z (CapIQ, ex-China, as of Sep. 11, 2026), median quarterly year-over-year revenue growth across all software companies fell from about 25% in Q1 2022 to about 12%–13%, and a16z says it has since stabilized (a16z p.60).
• a16z argues that software, SaaS included, has given up growth to earn profits. Per a16z (CapIQ, Sep. 8, 2026, ex-China), about 75% of public software companies are now profitable, while only about 30% grow TTM revenue by 20% or more (a16z p.61).
• Newspapers offer a comparison. Per a16z (FactSet, Goldman Sachs), U.S. newspaper shares started sliding 5 years before earnings gave way (2002–2011 chart), and a16z asks whether software could follow print media (a16z p.62).
• a16z then asks whether AI could be a “rising tide” that lifts ROE across companies. Per a16z (FactSet, Datastream, Goldman Sachs, Jul. 7, 2026; trimmed mean), the forecast line for capital-light companies ends near 25 around 2027, against about 16 for capital-intensive ones; the axis has no unit (a16z p.63).

a16z's argument in this chapter is that much of today's value creation sits with companies that are still private: the largest are bigger than ever (p.66), they stay private longer (p.67), and exit value and fund results are concentrating in a few hands (pp.68–71).

• Per a16z's PitchBook data, US VC-backed exit value reached $2,188 billion in 2026 year to date (through June 30); the highest earlier bar on the chart is $865 billion, in 2021. (a16z p.65)
• Per a16z's PitchBook and CapIQ data (as of Sept. 1, 2026), the five largest private tech companies outside China (Anthropic, OpenAI, Databricks, Stripe, Waymo) have last-reported valuations labeled $2.2 trillion in total, above the $1.73 trillion cumulative first-day-close value of the past decade's tech IPOs. (a16z p.66) Per PitchBook-NVCA data (as of June 30, 2026), active US unicorns total $5.34 trillion in post-money valuation, against $3.5 trillion for the Russell 2000 (2026 market caps as of Apr. 30). (a16z p.67)
• Per a16z's PitchBook data (as of July 20, 2026), the top 1% of US exits made up 84% of exit value in H1 2026 and the top 10% made up 94%, against 17%–43% and 54%–79% in each year of 2018–2025. a16z concludes that concentration has intensified. (a16z p.68)
• Per a16z's Carta data, in Q2 2026 the median tender-offer subscription rate was 93.1% and median seller participation was 57.9%. The slide headline calls subscription rates all-time highs; a16z also argues that discounts are uncommon while top performers command premiums. (a16z p.69)
• Per a16z's PitchBook and CapIQ data, indexed to 100 in 2017, the Top 30 VC portfolios reached 2,333 by 2026, against 542 for the top 10 public companies. a16z argues that managers show the same concentration. (a16z p.70)
• a16z argues that VC fund performance is more dispersed than ever, with top-decile funds far ahead of the rest (PitchBook IRRs by vintage, as of Dec. 31, 2025). In Carta data as of Q1 2026, the 2017 vintage's 90th-percentile net TVPI is 3.46x, against a median of 1.64x. (a16z p.71)

a16z argues that startups have entered a new cycle too: after the zero-rate era (ZIRP), growth gave way to profitability, funding concentrated in AI, unicorns got younger, and the AI generation sits on a different growth curve.

• According to a16z, the median recently funded US VC-backed tech company grew revenue roughly 60% or more year over year on the chart; a16z says that matches or beats peak ZIRP, and fresh rounds still tend to demand fast growth. (a16z p.73)
• Per JPMAM data cited by a16z, the AI-related share of US VC deal value rose every year, from 15% in 2016 to 86% in 2026; a16z adds that beneath the concentration is a mix shift, with dollars also reaching legal and accounting services and capital-intensive areas like semis, power and defense. (a16z p.74)
• Per SVB data cited by a16z, the chart's 2023–2026 trend lines are labeled with a 37% decline in the median age of new unicorns and a 7% rise in the median age of all active unicorns. (a16z p.75)
• In the same SVB data, 42% of 2026 US VC-backed tech unicorns grew revenue 0% to 20% year over year, the largest bucket; a16z's headline says most unicorns grow 20% or less. (a16z p.75)
• a16z marks the 2022 vintage's year-4 revenue growth curve as about 3x steeper; the page does not state the baseline, and the chart excludes large AI hyperscalers. (a16z p.76)
• Per proprietary data the page labels approximate, top AI apps with base-year ARR of $1-50 million had median YoY growth of 423%; those at $50-200 million had 150%. (a16z p.76)

a16z argues that data from private tech companies increasingly gives the earliest read on major economic questions (a16z p.77). It closes with seven untapped opportunities: heavy-use consumer, robotics, autonomy, AI x Bio, personal health, enterprise diffusion beyond coding, and a new era of American Dynamism (a16z p.89).

• According to a16z's OpenRouter data, weekly tokens rose from 4.7T in September 2025 to 126.2T in September 2026 (as of Sept. 16), about 27x, and have doubled twice since June (a16z p.78). a16z calls compute price a key KPI and says Kalshi has been making a market in it (a16z p.81).
• a16z argues AI adoption is “wide and shallow” for now. In OAI Signals data (Aug. 12, 2026), the share of weekly active enterprise users using Plugins was 95% in the OpenAI group, 21% at frontier firms (top 10%) and 9% at typical firms (middle 10%) (a16z p.79). A Databricks blog (Aug. 7, 2026) shows its Smart Router solving 92.3% of coding tasks at $2.13 each, labeled 35% lower cost than Claude Opus 5; a16z says falling costs make more use cases viable (a16z p.80).
• By a16z's estimate, four private inference companies added about 3/4 of the 2026 ARR that Snowflake and Datadog added, at about 1/5 of their equity value (private firms at latest funding valuations; public figures from CapIQ, Sept. 16, 2026) (a16z p.82). Stripe Economics data (Sept. 10, 2026) show seasonally adjusted year-over-year SaaS revenue growth for the young (under one year) group peaking near 620% in the first half of 2026, which a16z calls a “narrative violation” (a16z p.83).
• In Protege data, health systems with high ambient AI adoption (share of notes referencing AI tools) show about 1,000 in-system deaths per 100k patients in 2026, versus about 1,600 at low adoption. The groups track closely in 2023 to 2025, and the error bars are wide and overlap. a16z says it is early but reads this as AI already saving lives (a16z p.84). EliseAI data (Jan. 16, 2026) show tour conversions up 5.28% after voice localization; a16z calls such gains often incremental, yet cumulative (a16z p.85).
• a16z says defense priorities have changed, with newer munitions that are smaller, cheaper and suited to mass production. One LRASM at $4.49 million per round buys at least 8 Castelion Blackbeard ER rounds, each under $500,000 (a16z p.86).
• In a CrimRxiv working paper cited by a16z (Aug. 16, 2026), motor vehicle theft fell about 11% in the 12 months after Flock deployment versus the prior 12 months, with 95% confidence limits shown (a16z p.87).