a16z Growth just published its September State of Markets deck. It runs 90 slides, and most of it covers macro: chips, memory, power, defense, robotaxis, and what they call the rotation “from bits to atoms.” All of that matters, but it isn’t where most of us work.
The B2B, venture, and startup sections are where the data gets useful for founders. Some of it confirms what you already feel in your pipeline. Some of it contradicts the headlines, and in a couple of places a slide’s own title contradicts its own data.
Here are the 10 learnings that matter for B2B + AI founders and investors.
#1. Horizontal B2B Apps Now Trade at a Median 2.7x TTM Revenue
This is the number to know. Per JPMAM data in the deck, median EV/TTM revenue by segment as of H1 2026 looks like this:
- Cloud, data, and AI infrastructure: 9.1x
- Security and identity: 6.8x
- Vertical software: 4.6x
- Consumer, commerce, and transactional platforms: 4.0x
- Horizontal software: 2.7x
Every category compressed. Horizontal apps compressed the most, and infrastructure held up best. a16z ranks the damage as cyber first, then vertical, then horizontal, with cyber and observability as the standout categories.
The market is pricing defensibility:
- If you sell a horizontal app that an AI-native competitor could plausibly rebuild, the market assumes someone will.
- If you own the data layer, the security layer, or a vertical workflow with real switching costs, you get paid more than 3x as much per dollar of revenue.
The founder decision: If you’re horizontal, your next fundraise or M&A conversation will start from 2.7x. Either get a credible infrastructure or vertical story, or plan around that multiple.
#2. Public B2B Companies Growing 20–40% Trade at Roughly 9–13x Forward Revenue. Those Growing 10–20% Trade at 4–5x
The second chart on the same slide splits public software by growth rate on TEV/forward revenue:
- High growth (20–40%): multiples have bounced around but mostly held in roughly the 9x to 13x range.
- Medium growth (10–20%): multiples slid to around 4x to 5x and stayed there.
a16z’s summary is that high-growth multiples have normalized and medium-growth multiples haven’t recovered. That’s accurate. The gap between 25% growth and 15% growth is worth 2x to 3x on your valuation.
In short, the market didn’t abandon B2B. It stopped paying for mediocre growth.
The founder decision: If you are at scale, going from 15% growth to 22% is worth more to your enterprise value than almost any margin improvement you could make this year. Allocate accordingly.
#3. About 75% of Public Software Is Profitable, and Only About 30% Grows 20%+
This slide describes the whole public B2B market. High-growth public B2B companies are now rare. Low-growth, high-margin ones are the norm.
a16z’s phrasing is that software has “traded growth for profitability.” Public B2B spent the last few years becoming a cash-flow business. Boards and investors got what they asked for: profitable companies growing in the low teens.
The market then priced them exactly like profitable companies growing in the low teens, which is the 4x to 5x from Learning #2.
The founder decision: Profitability is table stakes now and earns no premium. If you’re approaching an IPO or a sale, you need both profitability and growth, because 75% of the comp set already has the profitability.
#4. Top-Quartile Public B2B Growth Has Stabilized Around 20%, With the Median at About 12–13%
The deck’s growth-by-cohort chart shows reported quarterly revenue growth for public software and B2B. For the most recent quarters:
- 90th percentile: around 29–30%
- 75th percentile: around 20–22%
- Median: around 12–13%
- 25th percentile: high single digits
The important finding is that the lines stopped falling. Growth flattened over the past several quarters instead of continuing to decline. Operating leverage metrics (OpEx/revenue, margin change, revenue per employee) are stable or improving across the board. Revenue per employee is still climbing.
a16z notes that the fundamentals have been “remarkably stable” through the SaaSpocalypse selloff. The stock prices moved more than the businesses did.
#5. Stripe Data Shows New B2B Companies Growing 500–600%, but Mature B2B Bottomed at About 19%
This is the most interesting slide in the deck, and the one where the headline needs the most care.
a16z’s title claims B2B accelerated into the SaaSpocalypse, using Stripe payment data. The chart splits companies by age:
- Young B2B companies (under 1 year): year-over-year revenue growth ran in the low hundreds of percent through 2025, then went nearly vertical to about 500–600% by early 2026. It has since cooled slightly but remains above 500%.
- Mature B2B companies (1+ years): growth peaked around 32% in early 2025, decelerated to about 19% around January 2026, and has recovered to only about 24%.
The acceleration is almost entirely brand-new companies. Mature B2B slowed down through the SaaSpocalypse and has only partially recovered.
Both halves matter. The young-company number shows how fast AI-native B2B products are scaling from zero. That’s the new competitive set for every incumbent. The mature number shows that established B2B companies, the ones most founders are running, went through a real slowdown, and it wasn’t just sentiment.
The founder decision: If you’re an established B2B company, your competitive threat is a company that didn’t exist 12 months ago and is growing 5x. Assign someone to track new entrants in your category by name every quarter.
#6. 57% of U.S. Unicorns Grow 20% or Less, and 55% Have Under Two Years of Runway
a16z’s unicorn slide is titled “Mostly Profitable, Mostly Growing ≤ 20%.” Per the SVB data it cites, the second half is right. The first half isn’t supported by its own table.
For U.S. VC-backed tech unicorns in 2026:
- Growth: 42% grow 0–20%, and another 15% are shrinking. That’s 57% at 20% or below.
- Margin: 4% have margins above 25% and 21% have margins of 0–25%. That’s about 25% profitable, not “mostly.”
- Runway: 26% have 0–1 year and 29% have 1–2 years. That’s 55% with under two years of runway.
The herd is also aging. The median age of all active unicorns is up 7% to roughly 15 years. Meanwhile, new unicorns are 37% younger than they were a few years ago.
So there are two populations: a large, aging cohort of low-growth, mostly unprofitable companies burning toward a deadline, and a small new cohort of young AI companies reaching $1B+ faster than ever.
#7. Startups That Recently Raised Are Growing 60–70%, Versus 15–30% for Everyone Else
a16z plots median revenue growth against median profit margin for VC-backed startups by sector. After ZIRP ended, the median startup did exactly what public companies did: it cut burn, moved toward breakeven, and gave up growth to get there. As of Q2 2026, median startups at scale across fintech, enterprise, and consumer internet sit around 15–30% growth with modestly negative margins.
The companies that actually raised money recently look completely different. They’re growing at roughly 60–70%, in line with or faster than peak ZIRP, with deeply negative margins.
The bar to raise hasn’t come down. It’s where it was in 2021, and fewer companies clear it.
The founder decision: If you’re growing 25% at breakeven at scale, you have a good business that probably can’t raise a priced equity round at a good valuation. Plan to fund growth from cash flow, venture debt, or a strategic partner, and don’t spend six months on a fundraise that the data says won’t close.
#8. For the 2024 Vintage, Top-Decile VC Net IRR Is 40.5% and the Median Is -3.3%
This is from Carta data covering 2,773 funds and about $119B in committed capital, as of Q1 2026.
For the 2024 vintage:
- 90th percentile net IRR: 40.5%
- Median: -3.3%
- 25th percentile: -14.6%
That’s a 44-point spread between top decile and median. a16z’s companion chart shows dispersion of VC IRRs at its widest in the dataset.
DPI is the harder number. For every vintage from 2021 on, median DPI is 0.00x. Top decile for 2021 through 2023 ranges from about 0.06x to 0.16x. LPs in those vintages have received essentially no cash back.
Secondaries tell the same story. a16z notes that discounts are few and far between, and the best companies trade at a premium. The power law is steeper than it’s ever been, for both companies and managers.
The founder decision: When you pick investors, their fund’s position on this curve affects you. A fund with 0.00x DPI and a struggling vintage will be slower on follow-ons and more nervous at the board table. Ask where they sit before you sign.
#9. Only 20% of Organizations Cite Cost as a Constraint on AI, and AI Spend Is Still Below Cloud’s 4% of IT Budgets in Year Two
McKinsey data in the deck shows only 20% of organizations say AI-related costs have limited their adoption. A majority of every sector except one expects to increase AI spending over the next year.
For context, cloud reached about 4% of IT budgets in its second year. Most estimates put AI below that today, and a16z argues AI’s TAM is considerably larger than IT budgets, because AI competes for labor spend, not just software spend.
a16z also describes adoption as wide but shallow. A tiny fraction of power users consume orders of magnitude more AI than everyone else, and that gap has widened since early 2026. Growth in token demand is being driven by agents and by that top sliver of users.
The founder decision: If you sell AI products, your buyer mostly has budget. What’s scarce is depth of deployment. Price and package for the power users inside each account, because they drive expansion revenue, and most of your customers’ employees haven’t gotten there yet.
#10. Companies With High AI Adoption Increased Entry-Level Hiring Share by 1.15 Points. Low Adopters Cut It by 0.52 Points
Revelio Labs and Ramp data in the deck tracks entry-level headcount share at companies by AI adoption intensity. Starting 24 months after adoption:
- High-intensity AI adopters: entry-level headcount share up an average of 1.15 percentage points
- Low-intensity AI adopters: down an average of 0.52 points
Separately, BLS data via The Economist shows above-trend U.S. hiring for technical roles in 2026, including hundreds of thousands of additional AI-related white-collar jobs, with software developers, math and data science, and info-security making up most of the gain.
This contradicts the “AI eliminates entry-level jobs” story, at least so far. The companies using AI most are hiring more junior people, likely because each junior person with AI can produce much more.
The founder decision: Don’t freeze junior hiring on the theory that AI replaces it. The data says the companies furthest ahead on AI are doing the opposite.
A Few More Worth Knowing
- Older GPUs aren’t losing value. Rental prices for GPUs four and five years old are roughly where they were a year ago. Depreciation bears have been wrong so far, because demand for compute keeps outrunning supply.
- Hyperscaler cloud backlogs doubled year-over-year. Demand still exceeds supply, and free cash flow at hyperscalers is expected to stay compressed until around 2028 as capex absorbs it.
- Cheaper models aren’t killing frontier models. Open-weight models are gaining share of tokens, but total token spend is growing, and frontier models still command a premium for hard problems. Model competition isn’t zero-sum.
- The public market isn’t a Mag 7 story anymore. The biggest tech companies lagged the indexes this year because they’re pouring profits into the AI buildout, and the returns went to the semiconductor supply chain instead.
