ICONIQ j** ust released the third edition of its State of AI report,** subtitled “The Builder’s Economy,” built on a Q2 2026 survey of roughly 305 executives at software companies building AI products. The thesis moved in six months. The January edition was about proving AI works. This one is about proving AI pays.
AI products are now approaching half of revenue at the companies surveyed, margins are expanding, and the operators pulling ahead treat pricing, cost, and org design as product decisions instead of afterthoughts. A few numbers also reversed hard from the last report, which makes this a good moment to reset your priors. Here are the 10 metrics every B2B founder and operator should know cold.
1. Frontier Leaders vs Open Source: Open Source Is Now 40% of the App Layer #
The model story in 2026 isn’t lab vs lab, it’s the frontier leaders against open source. The frontier order did reshuffle fast: Anthropic overtook OpenAI as the most-used provider (81% vs 71%, up from 51% in Q4 2025), with Google at 50%. But that’s a subplot. The average builder runs 3.3 providers, so no single lab is the story.
The story is the category fight. By model type, licensed frontier APIs still lead at 80%, but that’s down from 85%, while open source climbed to 40% (from 37%) and now edges past proprietary in-house models at 37%. ICONIQ expects open source to keep gaining as cost and control advantages compound. The fork for builders isn’t which frontier lab to pick, it’s whether to rent frontier intelligence or run open models you control.
Unexpected learning: The leaders’ grip is looser than the provider ranking looks. Frontier API usage is sliding (85% to 80%) while open source is rising (37% to 40%), and 58% of builders now fine-tune or customize on top of whatever they run, so almost nobody uses a model off the shelf. The moat is shifting from whose model you use to who controls the stack.
2. AI Revenue For “Non AI Native” Startups: From 32% to 42% to 53% in Two Years #
AI products went from 32% of revenue in 2025 to a projected 42% this year, and ICONIQ has 2027 crossing 53%, the level the report flags as the majority threshold. In two years, AI goes from a third of the business to more than half of it.
That crossover is the whole premise of “The Builder’s Economy.” Once AI is the majority of revenue, it stops being an experiment line or a roadmap bet and becomes what your margins, pricing, and valuation ride on. You resource it, staff it, and report it like the core business, because it is the core business.
Unexpected learning: Read the arc as momentum, not market share. The number is a self-reported projection, it excludes AI-natives (95%+ AI revenue), and it’s an unweighted average of ratios, so a $10M shop that’s 90% AI counts the same as a $2B company that’s 10% AI.
3. Vertical Apps Are 43% of What Builders Make #
Vertical AI applications are now the single most common product being built at 43%, well ahead of horizontal apps at 20% and consumer at 11%. Builders converged on the application layer, and inside it they’re going deep into specific industries.
Financial services jumped to the #3 targeted use case from #8, and healthcare rose to #5 from #11, while customer support fell from #3 to #6. Teams are walking away from the easy, generic use cases and into hard, regulated domains where the workflow itself becomes the moat. Model-layer differentiation, by contrast, ranks dead last as a priority at 13%.
Unexpected learning: The use cases gaining fastest are the hardest and most regulated. Financial services and healthcare are climbing precisely because their workflow depth is what a general-purpose model can’t copy overnight.
4. 66% Rank Agentic Capabilities as Their #1 Priority #
Asked where they’re investing in customer-facing product over the next 12 months, 66% put agentic AI capabilities in their top three, the clear leader. Agent reliability and orchestration infrastructure came next at 35%. The roadmap is pointed at agentic depth.
That’s the ambition. The reality inside these same companies is it’s still a work in progress.
Unexpected learning: Agents are the top investment priority and the weakest performer at the same time. Internal agent productivity impact is stuck under 30% across every revenue band, and only 5% of teams say their agents rarely need human intervention. Nearly half report agents need a human occasionally, and another 39% need one frequently. The bet is real, but the payoff hasn’t arrived.
5. AI Product Gross Margins Hit 53%, Heading to 59% #
Average gross margin on AI products climbed from 45% in 2025 to a projected 53% this year, on track for 59% by 2027. That’s 14 points of expansion in two years, driven by falling inference costs, routing strategies, and revenue scale rather than pricing power. The margin panic of 2024 is over.
High-growth companies are pulling ahead on this too, projected at 64% gross margin in 2027 versus 58% for everyone else.
Unexpected learning: The highest-margin layer is the picks and shovels. AI infrastructure and platform products are projected to hit 67% gross margin in 2027, above application-layer products at 60%. That flips the usual assumption that infrastructure is capital-heavy and low-margin.
6. Consumption Pricing Jumped to 42% #
Subscription still anchors most models at 57%, but consumption-based pricing climbed from 35% to 42% and outcome-based from 18% to 23%. Companies now blend 1.7 pricing models on average, up from 1.5, as they align price to the cost and value of usage.
The move is pragmatic. Builders are stacking a subscription floor with usage components so the bill tracks the value delivered without torching the margin when consumption spikes.
Unexpected learning: Among companies on consumption pricing, only 15% eat 100% of the inference cost. The other 84% pass at least some of the token bill through to the customer. The cost of tokens is increasingly the buyer’s problem, which protects vendor margins as usage grows.
7. Data Readiness Flipped: $500M+ Fully-Ready Jumped From 4% to 22% #
This is the reversal from the last report. Six months ago, only 4% of companies over $500M in revenue called their data foundation fully ready, and readiness got worse with scale. This edition shows the opposite: that number jumped to 22%, the biggest gain of any size band. Smaller companies improved too, from 10% to 19%.
The infrastructure investment is finally landing. What was the top execution bottleneck in January is now the metric moving fastest in the right direction.
Unexpected learning: Scale flipped from a liability to an advantage here. The largest companies posted the single biggest jump in data readiness, reversing the pattern from two quarters ago where getting bigger meant feeling less ready.
8. 78% Are Planning a Different Role Mix or a Smaller Team #
The org chart is changing as much as the product. 78% of companies are planning either a different mix of roles or a smaller team than they’d otherwise have. AI-driven companies already run flatter, with 72% operating on just one to four layers between the CEO and the most junior IC, versus 56% of their peers.
The efficiency shows up in output per head. ARR per FTE at high-growth companies is projected to reach $496K in 2027, up from $270K in 2025, and to beat non-high-growth peers by about 11%.
Unexpected learning: This is not a mass-layoff story. Only 33% plan a smaller team. The larger group, 45%, is changing the mix (fewer operational roles, more AI-fluent talent) with no net headcount reduction. Under the hood, R&D, sales, and product are growing while customer support and G&A shrink.
9. 50% Are Scaling Up Forward-Deployed Engineers #
Half of companies now plan to scale up their forward-deployed engineer model as a permanent part of their GTM motion. The complexity of AI deployments is pulling human engineers back into the sale, and enterprise FDE coverage is projected to reach 34% of customers by 2027.
FDEs are landing on the revenue side of the house. The most common description of their primary role is revenue driver at 38%, ahead of delivery and product-gap coverage.
Unexpected learning: FDEs are comped like reps. 76% of their variable pay is tied to customer retention and renewal, and 74% to upsell and expansion. They are quota-carrying product engineers, a new hybrid role showing up on the org chart.
10. Coding Assistance Delivers 48% Productivity Gains at High-Growth Companies #
Across internal use cases, coding assistance produced the biggest productivity gain and the widest gap between high-growth companies and everyone else: 48% versus 32%, a 16-point spread. High-growth teams also ramp new AI tools faster (2.5 months versus 3.5) and write more of their code with AI (59% versus 47%). R&D leads internal adoption with the highest maturity score at 3.7.
Unexpected learning: The gap between functions comes down to conversion. G&A actually has more AI tool access than GTM (59% versus 54%), but the lowest share of power users (28%) and the lowest maturity (3.1). Handing out seats isn’t the bottleneck. Turning access into daily, expert usage is.
Five More Stats Worth Flagging #
Quality assurance is still mostly reactive: only 21% of builders run proactive adversarial testing or red-teaming.92% of companies find internal AI’s true cost hard to predict, with token spend the top source of budget overruns (one workflow projected at $0.10 per run hit $1.50+).** Security/privacy and SOC2/SLAs both climbed the model-selection ranking**, a clear signal of enterprise-readiness pressure as products mature.** RAG is now table stakes at 63% adoption**, and context engineering (memory, dynamic assembly, agent-to-agent) is emerging as its own discipline.** Hybrid GTM gained the most ground of any motion**, now at 36% and nearly tied with sales-led at 37%.
What the Builder’s Economy Rewards in 2026 #
Line the ten metrics up and the pattern is the same. Launching AI was 2024. Proving it works was early 2026. Now the job is proving it pays, and the operators pulling ahead are the ones who treat pricing, cost, org design, and model choice as product decisions.
The winners this year aren’t the teams with the best model access, because everyone runs three or more now. They’re the teams turning AI into revenue, expanding margin through routing and inference discipline, redesigning the org around AI-fluent talent, and converting tool access into real usage. That’s the builder’s economy, and it rewards execution over everything else.
Source: ICONIQ Analytics + Insights, 2026 State of AI: The Builder’s Economy, July 2026, based on a Q2 2026 survey of ~305 executives at software companies building AI products.