ICONIQ Capital just released their January 2026 “State of AI: Bi-Annual Snapshot” report, surveying approximately 300 executives at software companies building AI products. The respondents include CEOs, Heads of Engineering, Heads of AI, Heads of Product, CROs, and CFOs. The companies span a wide range: from sub-$5M to $1B+ in ARR, with growth rates from 0% to 150%+, and 85% headquartered in North America with 15% in Europe.
This is ICONIQ’s second bi-annual survey (following their Q2 2025 “AI Builder’s Playbook”), so we’re starting to see meaningful longitudinal trends. The core thesis: we’ve shifted from the race to experiment with large models to the challenge of scaling AI into durable, economically sound products. Building AI features is no longer a competitive advantage—it’s table stakes.
Here are the five things that matter most for founders right now.
1. Vertical AI Applications Are Where the Value Is Being Created
Nearly 70% of companies are now building vertical AI applications, up from 59% six months ago. And 49% of teams say application-layer innovation—unique UX, workflows, integrations—is their primary source of differentiation.
The model layer is commoditizing fast. Companies are using an average of 3.1 model providers now, up from 2.8 six months ago. OpenAI remains dominant at 77%, but Google/Gemini jumped to 55% (from 43%), and Anthropic/Claude is at 51%.
The takeaway is clear: don’t try to compete on models. Compete on what you build on top of them. The winners will be the companies that deeply understand specific workflows in specific industries and build AI that solves those problems end-to-end.
2. AI Gross Margins Are Improving—But Only If You’re Disciplined
Average AI product gross margins are projected to hit 52% in 2026, up from 41% in 2024. That’s meaningful improvement, but there’s a catch.
Companies with “balanced differentiation” (combining model and product innovation) report the highest margins at 53%. Pure application-layer companies are at 45%. The cost structure is also shifting dramatically as products scale: talent costs drop from 32% to 26% of total spend, while model inference climbs from 20% to 23%.
What this tells us is that long-term margin leadership depends on model selection, routing strategies, and infrastructure efficiency. The best teams are routing the majority of tasks to smaller, cheaper models and only escalating complex cases to frontier models. If you’re burning through GPT-4 tokens on tasks that GPT-3.5 can handle, you’re leaving money on the table.
3. Pricing Is Still a Mess—And 37% of Companies Plan to Change It
Here’s where things get interesting. 58% of companies still use subscription/platform pricing, but usage-based (35%) and outcome-based (18%) models have grown significantly. The outcome-based number jumped from just 2% in Q2 2025.
But the real story is that 37% of companies plan to change their AI pricing model in the next year. The top drivers are customer demand for consumption-based or outcome-based pricing (46%), customer demand for more predictable pricing (40%), and competitive pressure (39%).
The emerging best practice from the report: start with a light subscription for platform access plus usage for volume while outcomes are uncertain. Once outcomes stabilize, shift toward heavier subscription for predictability. One company in the survey averaged 1.6M monthly calls, cut average call time from 15 minutes to 4-5 minutes with 3x customer satisfaction improvement. At that scale, outcome-based pricing would have been more expensive for the customer, so they renegotiated to subscription-heavy.
Hybrid models with pricing safeguards (49% use annual commitments, 29% use overages at tiered rates) are becoming the pragmatic answer.
4. R&D Is Eating the AI Budget—And High-Growth Companies Are Spending Even More
Companies are allocating larger portions of their R&D budgets to AI development. The numbers by revenue range tell the story: sub-$100M companies went from 25% to 45% of R&D on AI, $100-250M went from 15% to 36%, and $250-500M went from 15% to 33%.
But here’s the number that should get your attention: high-growth companies (100%+ YoY ARR growth) are spending 57% of R&D on AI, compared to 38% on average.
This is a clear signal. The companies growing fastest are investing most aggressively in AI development. If you’re not allocating a significant portion of your engineering resources to AI, you’re falling behind.
5. AI Is a Force Multiplier, Not a Headcount Killer (Yet)
Despite all the hype about AI replacing workers, the data tells a different story. 42% of companies report no significant impact to headcount plans from AI adoption. 35% report a slight decrease, and 15% are actually increasing headcount (hiring for AI-related roles).
The productivity gains are real though. High-growth companies report 36% of their code is now written with AI assistance, up from 29% six months ago. Content generation and documentation show 35-42% productivity improvements. Coding assistance, testing, and code review show 31% gains.
What’s actually happening is workforce composition is shifting. Companies are prioritizing AI-fluent talent while de-emphasizing administrative and repetitive roles. AI is making top performers 10x more productive and helping newer employees upskill faster.
We Are Well Into The Execution Era
The experimentation phase is over. The companies winning in AI right now are executing on four dimensions: product (vertical applications with deep workflow integration), cost (efficient model routing and infrastructure), trust (better evaluation frameworks and hallucination mitigation), and go-to-market (pricing models aligned with value delivery).
If you’re building an AI-powered B2B product, stop worrying about which model to use. Start worrying about whether your application layer is differentiated enough, whether your cost structure is sustainable, and whether your pricing captures the value you’re creating.
The execution era has begun.
