The deck opens with a cover, a short introduction and a table of contents, then sets out its survey base of about 305 executives at software companies building AI products. An executive summary follows, presenting four shifts: convergence on the application layer, multi-model provider mixes, AI becoming a revenue and margin story, and AI reshaping org charts. The body is organised into four sections: AI Models and Infrastructure (product landscape, model types and providers, selection criteria, quality and security, data readiness), AI Go-to-Market and Economics (GTM motions, pricing models, AI revenue share, gross margins), Talent and Organization (org structure, headcount plans, a spotlight on forward deployed engineering), and AI for Internal Productivity (internal spend, adoption, agents, and functional case studies). Most content slides pair a headline sentence with a single chart or ranked table and a short sourced note. The deck closes with disclosures and a list of portfolio firms. Slides worth studying include the executive summary, the provider-share chart on slide 10 and the margin projection on slide 21.
Slides worth studying
- Slide 5: The executive summary reduces the whole report to four numbered shifts, each with a bold headline and a short paragraph, which makes the argument easy to scan.
- Slide 10: A paired bar chart compares 2025 and 2026 provider shares with the provider logos as labels, so the change in leadership reads at a glance.
- Slide 21: A large figure with a simple three-point trend line and a small product-type breakdown lets one projection carry the slide without extra text.
- Slide 24: A single stacked bar with five labelled segments shows the split of planning responses clearly, with the largest segment highlighted in the brand colour.
Details
- Company: ICONIQ Capital
- Year: 2026
- Type: Report-style deck
- Industry: Venture capital and AI software
- Pages: 47
The hosted copy of this deck is being prepared. Newsletter readers will find the download here.
Review score: 8 out of 10, Strong
Presentation quality, judged with the SlideDeck Co review heuristic. Ask, evidence and idea merit are not rated here.
The report is built to be read without a presenter: a contents page on 3 with page ranges, a methodology page on 4, an executive summary on 5 that numbers the four shifts, and section dividers with large numerals. Each chart page has a full-sentence finding as its title, a small section label above it and a source line below, and the green scale is used consistently so Q4 2025 and Q2 2026 can be told apart across slides 9, 10, 16 and 17. Slide 21 turns a three-point margin projection into three circles with the step changes labelled. The costs are density and type size: slides 26, 27 and 37 pack ten to fourteen rows with labels that are hard to read at presentation size, slide 23 fits three panels of small bars, and several titles run to two long lines.
Worth keeping: Keep the contents and executive summary pages, the sentence titles with section labels, the two-period colour coding and the highlight pages such as 21 and 33.
Most useful repair: Shorten titles to one line with the number in the subtitle, and split slides 26, 27 and 37 so each shows no more than eight rows at a readable size.
What would change the score: Readable type on the dense survey pages would make this a 9; the editorial structure already is.
Confidence high. 47 of 47 pages inspected on contact sheets, 4 at readable size. Reviewed 2026-11-11. Model-assisted review (claude-fable-5-1, in session).
Full text of the slides (47 slides)
Every slide's words and a description of its charts and images, so the deck can be searched and rebuilt. Also available as Markdown.
Slide 1
JULY 2026 / 2026 State of AI / The Builder’s Economy / ICONIQ / Private and Strictly Confidential / Copyright © 2026 ICONIQ Capital, LLC. All Rights Reserved / For Professional Clients Only. ICONIQ Partners (UK) LLP (973080) is an appointed representative of / Kroll Securities Ltd (466458) which is authorized and regulated by the Financial Conduct Authority
Visuals: Full-slide photograph-style illustration of a glowing 3D wireframe landscape: undulating mesh terrain of peaks and valleys receding toward the horizon, rendered as fine dotted grid lines with scattered luminous particles. Composition is a low oblique camera angle with the ridges rising from the lower half and fading into haze near the top; the image covers the full slide with no crop beyond the edges. Lighting is soft and diffuse, with faint warm yellow-orange light streaks running through the middle band. Dominant colours are teal #1E83A2, green #1FA35A, and blue-violet #5A6FA8. A teal tint at about 29% opacity (#1E83A2) overlays the image to soften it. A small ICONIQ wordmark in light grey serif capitals (#D9DCDF) sits top-right, about 297 by 70 pixels. A short light-blue accent bar (#5BC0DE approx.) sits under the date label.
Slide 2
Introduction / Building and operationalizing AI products has moved from competitive differentiation to table stakes. In January 2026, we / published “The State of AI: Bi-Annual Snapshot” to elevate the voices of architects, engineers, and product leaders driving the / disciplined execution of delivering and scaling AI products. / What began as a race to ship AI features has matured into a more consequential challenge: turning AI capabilities into durable, / economically sound businesses. Many of builders who are pulling ahead have made deliberate tradeoffs across the full stack – from / model selection and data infrastructure to org design, pricing architecture, and internal deployment – and are managing the cost, / trust, and go-to-market implications that come with each. / This report is ICONIQ's third bi-annual report of the AI builder landscape, grounded in our proprietary Q4 2025 and Q2 2026 / surveys of executives at software companies building AI products, alongside perspectives from our ICONIQ Community. Across / both waves, one shift stands out: the market has moved from proving AI works to proving AI pays. AI products are approaching half / of revenue at the companies we surveyed, margins are expanding, and the operators pulling ahead are the ones treating pricing, / cost, and org design as product decisions rather than afterthoughts. / Explore Our AI Perspectives / Private & Strictly Confidential / 2
Visuals: Two non-text elements. (1) A teal rounded-rectangle button, fill #438998, at x 6.1% y 83.1% w 20.1% h 5.6%, with a soft grey drop shadow falling down and to the right (rendered as two shadow images, 548 by 96 px and 487 by 94 px, stacked under the button); the button label is white. (2) The ICONIQ wordmark logo, a black sans-serif logotype image 131 by 31 px at x 3.6% y 95.2% w 4.9% h 2.1%, at bottom left. No photographs, charts or illustrations.
Slide 3
Table of Contents / Section / Topics Covered / Page / AI Models & Infrastructure / • AI Product Landscape / • Models Powering AI Products and Providers / • Performance, Quality, and Security for AI Models / • Data Readiness / 6-14 / AI Go-to-Market & / Economics / • GTM Motions & Pricing Models / • Revenue and Costs of AI Products / • Gross Margin Optimization / 15-21 / Talent & Organization / • Organizational Structure of AI Companies / • How AI is Impacting Org Structures / • Spotlight: Forward Deployed Engineering / 22-31 / AI for Internal Productivity / • Spend on AI Tooling / • Adoption and Impact of AI Tools and Agents / • Case Studies: How Functional Teams Are Implementing AI / 32-43 / ICONIQ / 3
Visuals: The slide has one small logo and no photographs, charts, or diagrams. Logo: the ICONIQ wordmark, set as a 131 by 31 pixel image at x 3.6%, y 95.2%, w 4.9%, h 2.1%, in black on white. Rules: one horizontal dark teal line (about #22454C, roughly 0.2% of slide height thick) spanning x 4.3% to 93.3% at about y 21.5%; three thin light grey lines (about #D9D9D9, roughly 0.1% thick) spanning the same width at about y 38.5%, 57.3% and 73.7%. Hyperlinked page ranges in the Page column are underlined in blue.
Slide 4
Data
Sources
& Methodology1
This study summarizes data from a Q2 2026 survey of ~305 executives2 at software companies building AI products, including CEOs, Heads of Engineering, Heads of AI, and Heads of Product.
Throughout this report, we compare insights to our prior State of AI report, published in Q4 20253, “2026 State of AI: Bi-Annual Snapshot”, where applicable. Where necessary, longitudinal data has been normalized to account for differences in firmographics to ensure trends are representative of the data.
We also weave in insights and what we believe to be best practices from AI leaders from the ICONIQ community.
All industry perspectives shared in this report have been anonymized to protect company-level information.
Respondent Firmographics
By ARR or Revenue Range
% of Respondents
9% 10% 10% 10% 10% 10% 7% 4% 9% 7% 7% 9%
$1M-$5M $5M-$9M $10M - $24M $25M - $49M $50M - $99M $100M-$149M $150M-$199M $200M-$249M $250M-$499M $500M-$749M $750M-$999M $1B+
By Headquarters
% of Respondents
87% 13%
North America Europe
In this report, select companies are referred to as “high growth companies” because they meet the following criteria:
➢ 100%+ YoY revenue growth if <$25M Revenue;
➢ 50%+ YoY revenue growth if $25M-250M Revenue;
➢ 30%+ YoY revenue growth if $250M+ Revenue
High Growth Companies
% of respondents
19%
Revenue Range
% of High-Growth Respondents
48% 24% 28%
<$100M $100-$500M $500M+
1 – This data was collected anonymously by an external survey. Survey responses include some but not all ICONIQ Venture and Growth portfolio companies as well as companies not part of ICONIQ Venture and Growth’s portfolio.
2 – Certain questions in the survey were optional or routed based on persona. Accordingly, some n-Size numbers in this presentation are less than 305.
3 – The Q4 2025 report summarizes data from a December 2025 survey of ~300 executives at software companies, including CEOs, Heads of Engineering, Heads of AI, Heads of Product, Chief Revenue Officers, and Chief Financial Officers.
ICONIQ
Private & Strictly Confidential
4
Visuals: 1) Vertical bar chart titled 'By ARR or Revenue Range', subtitle '% of Respondents'. Twelve bars, dark green #255E35, equal width, with value labels above each bar. Categories left to right with labels rotated about 45 degrees: $1M-$5M 9%; $5M-$9M 10%; $10M - $24M 10%; $25M - $49M 10%; $50M - $99M 10%; $100M-$149M 10%; $150M-$199M 7%; $200M-$249M 4%; $250M-$499M 9%; $500M-$749M 7%; $750M-$999M 7%; $1B+ 9%. No y-axis line or gridlines; a thin light grey (#D9D9D9) baseline under the bars. Range of the value axis about 0 to 10 percent is implied by bar heights; the bars are drawn to scale with the tallest (10%) at about 30 percent of the chart height above the baseline. No legend. 2) Vertical bar chart titled 'By Headquarters', subtitle '% of Respondents'. Two dark green (#255E35) bars: North America 87% (tall bar, about 87 percent of the plot height) and Europe 13% (short bar, about 13 percent). Category labels 'North America' and 'Europe' sit centred under the bars on a thin light grey baseline. No axis, no gridlines, no legend. 3) Donut chart 'High Growth Companies': ring with a dark green (#255E35) segment of 19 percent at the top-left (from 12 o'clock counter-clockwise to about 10 o'clock) and a light grey (#C9C9C9) remainder of 81 percent. Centre text '19%' in green #255E35, large serif. Label to the left: 'High Growth Companies' with '% of respondents' in italic beneath. The ring is solid, with an inner hole about 60 percent of the outer diameter. No legend. 4) Vertical bar chart 'Revenue Range', subtitle '% of High-Growth Respondents'. Three dark green (#255E35) bars: <$100M 48% (tallest), $100-$500M 24% (shortest), $500M+ 28% (middle height). Value labels above each bar; category labels beneath on a thin light grey baseline. No axis or gridlines. 5) ICONIQ wordmark logo (dark near-black #000000 serif wordmark 'ICONIQ') at bottom-left. No photographs or illustrations. Bullet glyphs are Wingdings arrowhead symbols (➢) in dark grey.
Slide 5
Executive Summary / Four shifts we see defining the AI market today / 01 / Builders have converged on the application / layer and are betting on agentic depth / Nearly two-thirds of builders are building horizontal or vertical apps (vertical alone at 43% of companies analyzed) and they’re moving into high-complexity domains, with financial services and healthcare use cases rising in popularity. Agentic capabilities and agent reliability/orchestration are the top 2 customer-facing priorities for the coming year. / 02 / Multi-model appears to be the default with the provider mix reshuffled / To power AI products, most builders now run ~3.3 providers on average (up from 3.1 in Q4’25). Among survey respondents, Anthropic moved to the top spot (51% Q4 ‘25 → 81% in Q2 ‘26) as the field has broadened. Selection criteria held steady (reliability, then cost), but security/privacy and SOC2/SLAs both climbed, signaling enterprise-readiness pressure as products mature. / 03 / AI is now more of a revenue and margin story / rather than an experiment / AI products grew from 32% of revenue (2025 actuals) to a projected 42% this year, with gross margins moving from 45% to 53% and on track for ~59% by 2027. Part of rising margins is driving by pricing model changes: consumption- / and outcome-based models are rising (42% and 23%, with companies now blending 1.7 models on average vs. 1.5 in Q4’25), as builders align price to the cost and value of usage. / 04 / AI is reshaping the org chart, not just the / product / AI-driven companies¹ run flatter and more cross-functional than their peers. 45% of companies are planning a different mix of roles and 33% are actively planning a smaller team. While companies continue to see gains on internal productivity, it is difficult to predict internal AI’s true cost, with overruns emerging from tokens, data infrastructure, and enablement / Source: Perspectives from the ICONIQ GenAI Surveys (December 2025 & May 2026) and perspectives from the ICONIQ team and network of AI leaders consisting of our community of CIO/CDOs overseeing AI initiatives in enterprises, CTOs, our Technical Advisory Board, and others in our network / 1 – Companies with 50%+ revenue driven by AI products / ICONIQ / Private & Strictly Confidential / 5
Visuals: none. The slide has no charts, tables, photographs or illustrations. The only graphic marks are four short green accent bars and thin light grey horizontal rules above the numbered sections, plus the ICONIQ wordmark in the footer.
Slide 6
01
AI Models &
Infrastructure
Private & Strictly Confidential
ICONIQ
Visuals: One photograph, an illustration rendered as a glowing teal wireframe sphere. Subject: a large translucent dome made of fine luminous particles and dotted lines, with flowing wave-like mesh ribbons running horizontally across its lower half. Action: none, a static abstract render. Composition: portrait framing, sphere cropped at the top and sides of the box so its upper arc is near the top edge, with the lower ribbons running off the bottom edge. Lighting: soft internal glow, brightest cyan highlights along the ribbons, darker edges. Dominant colours: teal #1A8C9C at the top, deeper teal #137A8A in the lower areas, bright cyan #35D2E0 highlights, white-cyan particle dots. The photo is boxed with rounded corners of about 2% of slide width, not full-bleed, with no overlay or tint. Also one number '01' in light grey, a large serif display numeral, decorative. Also the ICONIQ wordmark logo in black, bottom-left, rendered as spaced uppercase letters.
Slide 7
AI Models & Infrastructure / Builders have converged on the application layer, with nearly two-thirds of the market focusing on horizontal and vertical / AI applications. Within this, data analytics and productivity tools are the most targeted use cases / What is the primary AI product you are building? / % of Respondents, Single-Select, N=304 / Vertical AI applications / Highly specialized applications designed for / a specific industry or function / 43% / Horizontal AI applications / Enterprise applications that can be used / across multiple industries or functions / 20% / Consumer AI applications / AI products primarily targeted at individual / consumers / 11% / Core AI models / technologies / Foundational AI models or specialized ML / algorithms / 10% / AI platforms / infrastructure / Tools that facilitate deployment, hosting, / monitoring, and scaling of AI in production / 10% / AI development tooling / Tools that support the building, evaluation, / and iteration of AI products / 6% / Which use case does your primary AI product target? / % of Respondents, Horizontal & Vertical AI Builders Only, Single-Select, N=192 / Q2'26 Rank / Use Case / Q4'25 Rank / Q2'26 Rank / Use Case / Q4'25 Rank / #1 / Data Analytics And Business Intelligence / 1 / #7 / IT & Security / 4 / #2 / Productivity & Collaboration / 2 / #8 / HR & Recruiting Tools / 10 / #3 / Financial Services / 8 / #9 / Legal & Compliance / 9 / #4 / Marketing & Sales / 5 / #10 / Product & Design / 7 / #5 / Healthcare & Life Sciences / 11 / #11 / Developer Tools / 6 / #6 / Customer Support / 3 / #12 / Other / 12 / Major Shifts / Companies are increasingly focusing on use cases with highly complex and specialized workflows, e.g., financial / services and healthcare, compared to 6 months ago. The workflow-embedded nature of vertical products appears to / give the product a natural moat that is harder for general-purpose AI models to displace overnight. / Source: Perspectives from the ICONIQ GenAI Surveys (December 2025 & May 2026) and perspectives from the ICONIQ team and network of AI leaders consisting of our community of CIO/CDOs overseeing AI initiatives in enterprises, CTOs, our Technical Advisory Board, and / others in our network / Private & Strictly Confidential / 7
Visuals: Chart: horizontal bar chart, untitled apart from its panel heading 'What is the primary AI product you are building?' with subtitle '% of Respondents, Single-Select, N=304'. No axes, no gridlines, no axis ranges; bars are left-aligned from a common baseline at x 20.7 and value labels sit at each bar end. Six series values printed: Vertical AI applications 43%, Horizontal AI applications 20%, Consumer AI applications 11%, Core AI models / technologies 10%, AI platforms / infrastructure 10%, AI development tooling 6%. Bars are all one colour, light green #94D5A6, with bar lengths proportional to those values. Highlighted: the first two categories (Vertical and Horizontal AI applications) carry dark green star icons. No legend. Table: 'Which use case does your primary AI product target?' split into two side-by-side blocks, each with columns Q2'26 Rank, Use Case, Q4'25 Rank. Rows in the left block: #1 Data Analytics And Business Intelligence, Q4'25 rank 1 with a grey dash (no change); #2 Productivity & Collaboration, 2, grey dash; #3 Financial Services, 8, green up triangle; #4 Marketing & Sales, 5, grey dash; #5 Healthcare & Life Sciences, 11, green up triangle; #6 Customer Support, 3, red down triangle. Rows in the right block: #7 IT & Security, 4, red down triangle; #8 HR & Recruiting Tools, 10, grey dash; #9 Legal & Compliance, 9, grey dash; #10 Product & Design, 7, red down triangle; #11 Developer Tools, 6, red down triangle; #12 Other, 12, grey dash. Use-case text is coloured: green #255E35 for Financial Services and Healthcare & Life Sciences; red #C00000 for Customer Support, IT & Security, Product & Design and Developer Tools; black for the rest. Logo: ICONIQ wordmark, dark charcoal, bottom-left. Other visuals: a thin dark-green vertical divider line and a light grey callout box behind the Major Shifts text. No photographs or icons apart from the star glyphs and triangle markers.
Slide 8
AI Models & Infrastructure / Deep workflow integration is the priority. Many builders are designing their products for habitual use and prioritizing investing in agentic capabilities over the next 12 months / How would you characterize the typical usage pattern of your primary AI product among active customers? / % of Respondents, Single-Select, N=305 / 84% / Of products target usage of >1 time per week for active customers / Full usage distribution / Daily use / Several times per week / Ad hoc / Weekly / 60% / 24% / 10% / 6% / Where are you prioritizing customer-facing AI product investment over the next 12 months? / Ranked by share of respondents placing each category in their top 3, N=305 / 1 Agentic AI capabilities 66% / 2 Agent reliability & orchestration infrastructure 35% / 3 Deeper vertical or domain-specific functionality 35% / 4 Application layer differentiation 34% / 5 Enterprise readiness 27% / 6 Context engineering 24% / 7 Multimodal capabilities 20% / 8 Integrations and ecosystem 18% / 9 Cost and inference optimization 14% / 10 Evaluation and reliability infrastructure 14% / 11 Model layer differentiation 13% / Source: Perspectives from the ICONIQ GenAI Surveys (May 2026) and perspectives from the ICONIQ team and network of AI leaders consisting of our community of CIO/CDOs overseeing AI initiatives in enterprises, CTOs, our Technical Advisory Board, and others in our network / ICONIQ / Private & Strictly Confidential / 8
Visuals: Two visuals. (1) Horizontal bar chart inside the dark green left panel, titled 'Full usage distribution' under the headline figure 84%. It has four categories from top to bottom with no printed axes: Daily use 60%, Several times per week 24%, Ad hoc 10%, Weekly 6%. Bars are solid rounded-corner rectangles in mid green #4DB96B with a thin darker green outline, left-aligned at about 24.4% of the slide width; the bar lengths are approximately 7.5%, 5.3%, 3.3% and 1.9% of slide width, so the bars are not strictly proportional to the printed values. Value labels sit right-aligned at about 36% of slide width: 60% in green #4DB96B, 24%, 10% and 6% in pale mint #E8F2EC. There is no legend and no highlighted point. (2) Ranked list of eleven categories, each with a numbered circular badge at about 44.6% of slide width, 2.2% wide and 4% tall. Badges 1 to 3 are dark green #255E35 with white numerals and their labels and values are dark green #255E35. Badges 4 to 11 are mid grey #6B6B6B with white numerals and their labels are near-black #151515. Thin light grey #E5E5E5 hairline rules separate rows. Values are printed at the right edge of each row. The ICONIQ wordmark is a small black logo image at bottom-left.
Slide 9
AI Models & Infrastructure
Licensed third-party APIs are the most popular model type to power customer-facing AI products, but adoption of open-source models is increasing. We expect this shift to continue as cost and control advantages compound
Q4 2025¹
Q2 2026²
Which type of model(s) power your customer-facing AI product?
% of Respondents, Select All That Apply
85%
80%
37%
40%
33%
37%
Licensed 3rd-party APIs
e.g., OpenAI, Anthropic, Google
Open-source models
e.g., Llama, Mistral
Proprietary foundation models
Pre-trained/trained from scratch, company-owned
# of Model Types Selected in Q2’26
% of Respondents, Select All That Apply
# of Model Types
1
55%
2
34%
3
11%
58%
Of respondents fine-tune and customize on top of models in Q2’26
Source: Perspectives from the ICONIQ GenAI Surveys (May 2026) and perspectives from the ICONIQ team and network of AI leaders consisting of our community of CIO/CDOs overseeing AI initiatives in enterprises, CTOs, our Technical Advisory Board, and others in our network; 1 – N = 288; 2 – N=292
Private & Strictly Confidential
9
Visuals: Chart 1, grouped vertical column chart. Title: 'Which type of model(s) power your customer-facing AI product?' with 'customer-facing' underlined; subtitle '% of Respondents, Select All That Apply'. Categories on x-axis: Licensed 3rd-party APIs (e.g., OpenAI, Anthropic, Google); Open-source models (e.g., Llama, Mistral); Proprietary foundation models (Pre-trained/trained from scratch, company-owned). Y-axis has no printed scale, labels or gridlines; a single grey baseline runs under the bars. Series Q4 2025 (light green bars) values: Licensed 3rd-party APIs 85%; Open-source models 37%; Proprietary foundation models 33%. Series Q2 2026 (dark green bars) values: Licensed 3rd-party APIs 80%; Open-source models 40%; Proprietary foundation models 37%. Legend: Q4 2025 superscript 1 in light green #4DB96B text above Q2 2026 superscript 2 in dark green #255E35 text at top-left of the chart area; footnote 1 is N = 288 and footnote 2 is N=292. Bar heights are proportional to the printed values, with the 85% bar tallest. No bars are highlighted. Chart 2, horizontal bar chart in the top-right panel. Title '# of Model Types Selected in Q2’26', subtitle '% of Respondents, Select All That Apply'. Vertical axis label '# of Model Types' rotated 90 degrees, categories 1, 2, 3 from top to bottom. Bars in light green #4DB96B: 1 = 55%, 2 = 34%, 3 = 11%, with value labels at the bar ends. No x-axis scale is printed. Visual 3, large-number callout in the bottom-right panel: '58%' set in very large light-green numerals. Logo: ICONIQ black wordmark at bottom-left.
Slide 10
AI Models & Infrastructure
Similarly, builders are choosing a wider variety of model providers (~3.3 models on average). Anthropic has grown to the
top provider among survey respondents over the last 6 months
Top Model Providers
% of Respondents, Select All That Apply
Q4 2025¹
Q2 2026²
81%
77%
71%
56%
51%
50%
30%
26%
27%
22%
21% 21%
8%
9%
8%
8%
7%
7%
5%
6%
Data not available
4%
4%
3%
4%
3%
ANTHROP\C
OpenAI
Azure
aws
Meta
databricks
Mistral AI
deepseek
Alibaba.com
Moonshot AI
Other
On average, companies selected ~3.3
model providers in Q2 2026 vs ~3.1 in
Q4 2025.
Source: Perspectives from the ICONIQ GenAI Surveys (December 2025 & May 2026) and perspectives from the ICONIQ team and network of AI leaders consisting of our community of CIO/CDOs overseeing AI initiatives in enterprises, CTOs, our Technical Advisory Board, and others in our network; 1 – N = 194; 2–N=301
ICONIQ
Private & Strictly Confidential
10
Visuals: Vertical grouped bar chart titled Top Model Providers, subtitled % of Respondents, Select All That Apply. Categories left to right: ANTHROP\C, OpenAI, Google, Azure, aws, Meta, databricks, Mistral AI, deepseek, Alibaba.com, Moonshot AI, xAI logo, Other. Each category has a light green bar (Q4 2025, N=194) and a dark green bar (Q2 2026, N=301); no axis lines or gridlines and no numeric axis are printed, values are printed above each bar. Values Q4 2025 / Q2 2026: Anthropic 51% / 81%; OpenAI 77% / 71%; Google 56% / 50%; Azure 30% / 26%; AWS 27% / 22%; Meta 21% / 21%; Databricks 8% / 9%; Mistral AI 8% / 8%; DeepSeek 7% / 7%; Alibaba 5% / 6%; Moonshot AI shows no Q4 2025 bar, a vertical italic label Data not available, and a Q2 2026 bar at 4%; xAI 4% / 3%; Other 4% / 3%. Legend at upper left: Q4 2025 in light green text above Q2 2026 in dark green text, with superscripts 1 and 2 referring to the source line. Callout box text: On average, companies selected ~3.3 model providers in Q2 2026 vs ~3.1 in Q4 2025, with ~3.3 and ~3.1 in bold; it is a plain grey rectangle, not a data series. Logos sit in a row under the bars: Anthropic wordmark in black caps, OpenAI logo with wordmark in black, Google wordmark in its multicolour, Microsoft Azure logo in blue with wordmark, AWS smile logo in black with orange arrow, Meta infinity logo and wordmark in blue, Databricks logo in red with wordmark in black, Mistral AI in black with orange-red pixel mark, DeepSeek whale icon with wordmark in blue, Alibaba.com in orange with script mark, Moonshot AI black circular icon with wordmark, an xAI X mark in black, and the text Other. No photographs or illustrations.
Slide 11
AI Models & Infrastructure
Top model selection criteria have largely remained consistent over the last 6 months, with model reliability & accuracy
remaining in the top ranked priority
Top Considerations When Choosing a Foundational Model for Customer-Facing Use Cases
Ranked by share of respondents placing each in their top 3, N=305
Q2’26 Rank
Consideration
Q4’25 Rank
#1 Model Reliability & Accuracy 1
#2 Cost 2
#3 Security / Privacy 4
#4 Ability to fine-tune / Customize 3
#5 Latency 5
#6 Speed of Model Advancement 7
#7 Model Transparency / Explainability 8
#8 SOC2 / Enterprise SLAs 10
#9 Competitiveness with Peers 6
#10 Open Source 9
Perspectives from the ICONIQ Network
Security/privacy, explainability, and SOC2/SLAs have climbed in
importance over the last 6 months, signaling enterprise-readiness
pressure as products mature.
Security is becoming a switching cost.
An enterprise software CISO noted that the first provider to lead on
enterprise controls (account blocking, execution governance, governance
hooks, etc.) becomes the default and drives consolidation.
Observability is the new blocker.
A CISO at a major data platform flagged a lack of telemetry and visibility into
what agentic tools are actually doing as an acute and growing concern.
Regulatory explainability is a hard gate.
A CISO at a global insurer can't yet describe to regulators what deployed
agents are doing or how they're monitored which instead of capability, is the
core blocker.
The gaps are live, not hypothetical.
A global bank's security team frames five production domains (identity,
shadow AI, data provenance, supply chain, insider risk) as the lens through
which model providers are now evaluated.
Source: Perspectives from the ICONIQ GenAI Surveys (December 2025 & May 2026) and perspectives from the ICONIQ team and network of AI leaders consisting of our community of CIO/CDOs overseeing AI initiatives in enterprises, CTOs, our Technical Advisory Board, and
others in our network
ICONIQ
Private & Strictly Confidential
11
Visuals: Table: a ranked list of 10 considerations with two rank columns. Columns are Q2’26 Rank (left, #1 to #10), Consideration (text), and Q4’25 Rank (right, prior rank numbers). Rows top to bottom: #1 Model Reliability & Accuracy, prior rank 1, indicator neutral dash; #2 Cost, prior 2, dash; #3 Security / Privacy, prior 4, up; #4 Ability to fine-tune / Customize, prior 3, down; #5 Latency, prior 5, dash; #6 Speed of Model Advancement, prior 7, up; #7 Model Transparency / Explainability, prior 8, up; #8 SOC2 / Enterprise SLAs, prior 10, up; #9 Competitiveness with Peers, prior 6, down; #10 Open Source, prior 9, down. Rows #3, #6, #7 and #8 are set in dark green text. Indicators are small solid triangles (green up, red down) and short grey dashes. No chart. Table has no row rules; the header row is a single teal band with thin white vertical dividers between its cells. Right-side panel holds text only, no photograph or illustration. ICONIQ wordmark logo bottom-left.
Slide 12
AI Models & Infrastructure / Builders are widely adopting a range of approaches to help ensure model output quality and safety, including user / feedback-driven evals and audit logging / QUALITY & RELIABILITY / Quality assurance is mostly reactive today. Builders catch issues through / feedback, monitoring, and human review rather than pre-deployment testing. / Proactive adversarial testing remains a minority practice at 21%. / SAFETY & SECURITY / Data protection guardrails, including audit logging, scope restrictions, and PII / controls are widely adopted. Adoption is lower for safeguards aimed at AI-specific / risks (e.g., prompt injection, data exfiltration, autonomous model actions). / How builders ensure output quality / % of Respondents, Select All That Apply, N=302 / 1 User feedback-driven evals 67% / 2 Model monitoring & observability 63% / 3 Human oversight & approval workflows 59% / 4 Fine-tuning & prompt optimization 55% / 5 Automated validation / eval frameworks 48% / 6 Grounding in verified or proprietary data 41% / 7 Adversarial testing / red-teaming 21% / 8 No formal quality practices yet 3% / Guardrails in place / % of Respondents, Select All That Apply, N=302 / 1 Audit logging of actions & data access 64% / 2 Topic or scope restrictions 59% / 3 PII detection & redaction 55% / 4 Input filtering 54% / 5 Least-privilege access controls 45% / 6 Prompt injection detection & prevention 44% / 7 Data exfiltration prevention 38% / 8 Limited model autonomy in high-risk cases 34% / 9 No formal guardrails 7% / Source: Perspectives from the ICONIQ GenAI Surveys (May 2026) and perspectives from the ICONIQ team and network of AI leaders consisting of our community of CIO/CDOs overseeing AI initiatives in enterprises, CTOs, our Technical Advisory Board, and others in our / network / ICONIQ / Private & Strictly Confidential / 12
Visuals: Two horizontal bar charts with no axes or gridlines. Left chart: title 'How builders ensure output quality'; subtitle '% of Respondents, Select All That Apply, N=302'; 8 bars with printed values: User feedback-driven evals 67%; Model monitoring & observability 63%; Human oversight & approval workflows 59%; Fine-tuning & prompt optimization 55%; Automated validation / eval frameworks 48%; Grounding in verified or proprietary data 41%; Adversarial testing / red-teaming 21%; No formal quality practices yet 3%. The top three bars are dark green #033A15, the next four mid-green #4DB96B, and the last bar is pale green. Highlighted points: none beyond colour grouping. Right chart: title 'Guardrails in place'; same subtitle; 9 bars with printed values: Audit logging of actions & data access 64%; Topic or scope restrictions 59%; PII detection & redaction 55%; Input filtering 54%; Least-privilege access controls 45%; Prompt injection detection & prevention 44%; Data exfiltration prevention 38%; Limited model autonomy in high-risk cases 34%; No formal guardrails 7%. The top three bars are dark green, the next five mid-green, and the last pale green. Bars are drawn on light grey #F2F2F2 tracks and bar lengths follow the printed values. Legend: none. No photographs, icons or other images besides the ICONIQ wordmark.
Slide 13
AI Models & Infrastructure
Context engineering is becoming increasingly important to AI product development as builders combine RAG, memory,
agent-to-agent context, among other techniques, to manage and provide context to models at runtime
How does your AI product manage and provide context to models at runtime?
% of Respondents, Select All That Apply, N=294
1
Retrieve relevant data (RAG)
RAG, knowledge bases
63%
2
Structure prompts or inputs
Prompt templates, few-shot examples
56%
3
Dynamically assemble context
Routing, multi-step pipelines
51%
4
Maintain memory across interactions
Cross-session context, user history
47%
5
Share context across workflows / agents
Multi-agent systems, tool coordination
47%
6
Minimal context management (static prompts)
16%
7
No formal approach to context management
3%
Source: Perspectives from the ICONIQ GenAI Surveys (May 2026) and perspectives from the ICONIQ team and network of AI leaders consisting of our community of CIO/CDOs overseeing AI initiatives in enterprises, CTOs, our Technical Advisory Board, and others in our
network
ICONIQ
Private & Strictly Confidential
13
Visuals: Horizontal bar chart titled by the question 'How does your AI product manage and provide context to models at runtime?' with subtitle '% of Respondents, Select All That Apply, N=294'. Value axis is percent of respondents, 0% to 63% scale implied by the track length, with no printed axis ticks; bars are ranked 1 to 7 top to bottom. Series values: 1 Retrieve relevant data (RAG) 63%; 2 Structure prompts or inputs 56%; 3 Dynamically assemble context 51%; 4 Maintain memory across interactions 47%; 5 Share context across workflows / agents 47%; 6 Minimal context management (static prompts) 16%; 7 No formal approach to context management 3%. Each bar is drawn to scale against a full track of 63%, so bar lengths are proportional to the printed values. Bars 1 to 5 are in graded greens darkening from top; bars 6 and 7 are pale green and visibly de-emphasised, with greyed labels. Sub-labels: RAG, knowledge bases; Prompt templates, few-shot examples; Routing, multi-step pipelines; Cross-session context, user history; Multi-agent systems, tool coordination. No legend. No highlighted points beyond the colour gradient. Also a small dark ICONIQ wordmark logo at bottom left, black on white.
Slide 14
AI Models & Infrastructure / Increased investment in AI infrastructure appears to be paying off, with more companies stating their data foundation is / fully ready and optimized to support accurate AI workflows / How would you rate your data foundation’s readiness to support accurate AI workflows?1 / % of Respondents, Single-Select / <$100M / $100M - $500M / $500M+ / Fully Ready / Mostly Ready / Somewhat Ready / Not Ready / 10% / 59% / 28% / 3% / 19% / 55% / 24% / 2% / 10% / 53% / 32% / 5% / 18% / 48% / 29% / 5% / 4% / 56% / 36% / 4% / 22% / 54% / 24% / Q4 2025 / Q2 2026 / Q4 2025 / Q2 2026 / Q4 2025 / Q2 2026 / Source: Perspectives from the ICONIQ GenAI Surveys (December 2025 & May 2026) and perspectives from the ICONIQ team and network of AI leaders consisting of our community of CIO/CDOs overseeing AI initiatives in enterprises, CTOs, our Technical Advisory Board, and / others in our network. N-Sizes: Q4 2025 – N=198; Q2 2026 – N=304 / 1 – Fully ready = Governed, high-quality, and labeled data optimized for AI workflows; Mostly Ready = Solid data pipelines, minor gaps; Somewhat ready = Basic data systems in place, limited for AI; Not ready = Significant data gaps or quality issues / ICONIQ / Private & Strictly Confidential / 14
Visuals: One stacked column chart, vertical, 100% stacked by respondent share. Title: How would you rate your data foundation’s readiness to support accurate AI workflows? (superscript 1 refers to footnote). Unit: % of Respondents, Single-Select. Vertical axis: no numeric axis or gridlines; instead four row labels at left (top to bottom): Fully Ready, Mostly Ready, Somewhat Ready, Not Ready, each aligned to its segment band. Horizontal axis: no line labels other than category labels. Three groups with headers <$100M, $100M - $500M and $500M+; each group has two columns labelled Q4 2025 and Q2 2026. Segments from top to bottom: Fully Ready (dark teal #326772), Mostly Ready (light teal #86BECA), Somewhat Ready (light grey #C9C9C9), Not Ready (mid grey #A5A5A5). Values: <$100M Q4 2025: Fully 10%, Mostly 59%, Somewhat 28%, Not 3%. <$100M Q2 2026: Fully 19%, Mostly 55%, Somewhat 24%, Not 2%. $100M - $500M Q4 2025: Fully 10%, Mostly 53%, Somewhat 32%, Not 5%. $100M - $500M Q2 2026: Fully 18%, Mostly 48%, Somewhat 29%, Not 5%. $500M+ Q4 2025: Fully 4%, Mostly 56%, Somewhat 36%, Not 4%. $500M+ Q2 2026: Fully 22%, Mostly 54%, Somewhat 24%; no Not Ready segment or label is printed. Each column is about 5.6% of slide width and spans from about 38.9% to 80.7% of slide height, drawn flat with no outlines. Legend: none separate; row labels act as legend. Highlighted points: none. Footnote defines the readiness levels. No legend box.
Slide 15
02
AI Go-to-Market
Strategy & Economics
ICONIQ
Private & Strictly Confidential
Visuals: One photograph, a tall rounded-rectangle image at x 62.1%, y 11.9%, w 31.2%, h 76.2% (1048 by 1440 px, portrait). Subject: a large glowing translucent sphere or dome made of fine dots and fine luminous wavy mesh lines, with a faint bright rim along the upper edge. Action: static abstract 3D render; the wavy lines flow horizontally across the lower half of the sphere. Composition: sphere centred horizontally, its top arc near the top third, the lower wave bands running off the bottom and sides; the image is cropped at all four edges by its rounded box, not full-bleed. Lighting: soft, self-lit glow, brightest in the cyan lines and the sphere rim. Dominant colours: teal #1E8FA0 at top, deeper teal #137A8C in the middle, bright cyan #3FD3E6 highlights on the wave lines, with small white-cyan dots. Overlay: none. Corner radius about 1.5% of slide width. One logo: the ICONIQ wordmark, small black uppercase sans-serif letters at bottom left.
Slide 16
AI GTM & Economics
GTM strategies for AI builders are diversified across different motions, but hybrid motions (combination of PLG and SLG)
has gained the most traction over the last 6 months
Q4 2025 1
Q2 2026 2
Primary GTM Motion for AI Products
% of Respondents, Single-Select
39%
37%
31%
36%
30%
27%
Sales-led growth
Hybrid (evenly split between product- and sales-led
growth)
Product-led growth / self-serve
Source: Perspectives from the ICONIQ GenAI Surveys (December 2025 & May 2026) and perspectives from the ICONIQ team and network of AI leaders consisting of our community of CIO/CDOs overseeing AI initiatives in enterprises, CTOs, our Technical Advisory Board, and
others in our network; 1– N=288; 2–N=289
ICONIQ
Private & Strictly Confidential
16
Visuals: One clustered vertical bar chart titled 'Primary GTM Motion for AI Products', subtitle '% of Respondents, Single-Select'. Series: Q4 2025 (light teal #86BECA) and Q2 2026 (dark teal #326772). Categories left to right: Sales-led growth (Q4 2025 39%, Q2 2026 37%); Hybrid (evenly split between product- and sales-led growth) (Q4 2025 31%, Q2 2026 36%); Product-led growth / self-serve (Q4 2025 30%, Q2 2026 27%). Value labels sit above each bar in black. No y-axis, gridlines or tick labels are shown; bar heights are proportional to the printed values, with the baseline at about 77.7% of slide height. Legend is top left: 'Q4 2025' in light teal #86BECA with superscript 1, and 'Q2 2026' in dark teal #326772 with superscript 2. Footnote markers 1 and 2 refer to the source line (N=288 and N=289). Highlighted points: none. Bars are flat, unoutlined rectangles of equal width with about 2.5% gap between paired bars and wider gaps between groups; a thin light grey (#D9D9D9) horizontal baseline runs across the chart width.
Slide 17
AI GTM & Economics
Subscription/platform components are still the most common pricing model; however, consumption and outcome-based
pricing has grown in prevalence over the last 6 months to better align pricing to product value
Q4 2025¹ / Q2 2026²
AI Pricing Strategies
% of Respondents, Select All That Apply
Companies selected 1.7 pricing models on average in / Q2 2026 vs 1.5 pricing models in Q4 2025
58% 57% 35% 42% 23% 28% 18% 23% 17% 15%
Subscription/platform
Consumption-based
Seat-based
Outcome-based
AI product is offered at no extra / cost
Source: Perspectives from the ICONIQ GenAI Surveys (December 2025 & May 2026) and perspectives from the ICONIQ team and network of AI leaders consisting of our community of CIO/CDOs overseeing AI initiatives in enterprises, CTOs, our Technical Advisory Board, and / others in our network; ; 1– N=297; 2–N=305
Private & Strictly Confidential
17
Visuals: Clustered vertical column chart titled AI Pricing Strategies with subtitle % of Respondents, Select All That Apply. Categories left to right: Subscription/platform, Consumption-based, Seat-based, Outcome-based, AI product is offered at no extra cost. Two series per category: Q4 2025 in light teal #86BECA (left bar) and Q2 2026 in dark teal #326772 (right bar). Labelled values Q4 2025 / Q2 2026: Subscription/platform 58% / 57%; Consumption-based 35% / 42%; Seat-based 23% / 28%; Outcome-based 18% / 23%; AI product is offered at no extra cost 17% / 15%. No value axis or gridlines are printed; bar heights are proportional to the labelled values with the baseline at the bottom, so unlabelled scale is about 1 percentage point per fixed unit and is not printed. Legend at top left: Q4 2025 in light teal with superscript 1 above Q2 2026 in dark teal with superscript 2. No bars are highlighted. Callout box at top right: light grey #EDEDED rectangle with a soft drop shadow containing the text Companies selected 1.7 pricing models on average in Q2 2026 vs 1.5 pricing models in Q4 2025. Thin light grey baseline under the bars. ICONIQ wordmark logo bottom left in near-black.
Slide 18
AI GTM & Economics
As AI becomes a more significant driver of revenue and customer base for companies building AI products, the economics
of building and selling AI products is becoming increasingly important
% of Revenue from AI Products¹
Average, Excludes AI-Native Companies, N=265
2 0 2 5
32%
Actual
Baseline
2 0 2 6
42%
Projected
+10pp vs. 2025
2 0 2 7
53%
Projected
Majority Threshold
Perspectives from the ICONIQ Network
1. AI adoption with customer bases frequently requires a Trojan horse
• Top-down "end-to-end automation" pitches reliably trigger
pushback from non-AI-native enterprise users
• What works: match the AI workflow to what users already do,
just faster, and let them discover the compounding effects
themselves. Adoption follows organically once two accelerated
steps visibly connect
• Customer literacy programs pay off even with no production
output. Getting users hands-on creates the internal advocates
needed to scale later
"We tried the full end-to-end automated pitch and everyone had that
knee-jerk reaction. Trojan-horsing it in, the same thing just faster,
got them to organically form their own view of what could be
automated.“ – CTO @ Series A, AI-Native Company
2. Feature distribution and customer education remain the bottleneck
• Shipping fast creates its own problem: the humans who talk to
customers need continuous education on new features to
translate them into adoption
• The more layers between product and customer, the more it
compounds across product marketing, sales enablement, and
account management
"I haven't found an AI-powered way to drive customer adoption
automatically. The humans who talk to customers are still the main
channel, and they need constant education.“ – Field CTO @ <$50M
AI-Enabled Company
Source: Perspectives from the ICONIQ GenAI Surveys (May 2026) and perspectives from the ICONIQ team and network of AI leaders consisting of our community of CIO/CDOs overseeing AI initiatives in enterprises, CTOs, our Technical Advisory Board, and others in our
network
1 – Includes companies where AI products drive >0% of revenue and excludes companies that have 95%+ of revenue driven by AI products, as these respondents are likely AI-native
Private & Strictly Confidential
18
Visuals: Not a conventional chart: three vertical stat cards, left to right, each showing a year, a large percentage and a status label. Card 1: year 2025, value 32%, label Actual, sub-label Baseline; light grey card #F2F2F2 with dark green number. Card 2: year 2026, value 42%, label Projected, pill reading +10pp vs. 2025 in mid-teal #438998 with white text; light blue card #D7E9ED with dark green number. Card 3: year 2027, value 53%, label Projected, pill reading Majority Threshold in dark teal #22454C with white text; mid-teal card #438998 with white number. Values are printed on the slide, so no estimates. Heading: % of Revenue from AI Products, superscript 1, subtitle Average, Excludes AI-Native Companies, N=265. No axes or legend. Panel on the right contains two numbered quoted perspectives with bullets, not a chart. Soft drop-shadow images sit beneath each card. ICONIQ wordmark logo bottom-left in black.
Slide 19
AI GTM & Economics
Talent costs tend to scale down, and inference costs tend to increase as a product scales. For companies using
consumption-based pricing, token/inference costs are generally a shared expense between providers and customers
Current breakdown of customer-facing AI product costs
Average, N=305
How does your organization handle the cost of
token/inference usage for your customer-facing AI products?
% of Respondents, Consumption-Based Pricing Users Only, N=72
Talent
Model Inference
Infra & Cloud
Model Training
Data Storage
Compliance
Other
34%
18%
16%
12%
12%
7%
30%
22%
17%
10%
13%
8%
28%
20%
17%
13%
13%
8%
Beta
GA
Scaling
We partially absorb costs and
partially pass through to customers
(e.g., via consumption-based pricing)
58%
We pass 100% of inference costs to
customers (e.g., via consumption-
based pricing)
26%
We absorb 100% of inference costs
(not passed down to customers)
15%
Source: Perspectives from the ICONIQ GenAI Surveys (May 2026) and perspectives from the ICONIQ team and network of AI leaders consisting of our community of CIO/CDOs overseeing AI initiatives in enterprises, CTOs, our Technical Advisory Board, and others in our
network
Private & Strictly Confidential
19
Visuals: Chart 1: stacked column chart titled 'Current breakdown of customer-facing AI product costs' (underlined 'customer-facing'), subtitle 'Average, N=305'. Three columns labelled Beta, GA and Scaling, each stacked to 100% with segments bottom to top: Other (thin strip, no value printed, olive #6C8C13 line), Compliance (Beta 7%, GA 8%, Scaling 8%; fill #DAF199), Data Storage (Beta 12%, GA 13%, Scaling 13%; fill #DBF1E1), Model Training (Beta 12%, GA 10%, Scaling 13%; fill #B8E3C4), Infra & Cloud (Beta 16%, GA 17%, Scaling 17%; fill #94D5A6), Model Inference (Beta 18%, GA 22%, Scaling 20%; fill #378D4F), Talent (Beta 34%, GA 30%, Scaling 28%; fill #255E35). Segment category names are printed as labels to the left of the Beta column, right-aligned, in their segment colour. No axis scale or gridlines; no legend other than the row labels. No highlighted points. Chart 2: horizontal bar chart inside the dark panel titled 'How does your organization handle the cost of token/inference usage for your customer-facing AI products?' (underlined 'customer-facing'), subtitle '% of Respondents, Consumption-Based Pricing Users Only, N=72'. Three bars, all in #4DB96B, left-aligned from x 80.2%, each with a value label at its right end: 'We partially absorb costs and partially pass through to customers (e.g., via consumption-based pricing)' 58% (longest bar); 'We pass 100% of inference costs to customers (e.g., via consumption-based pricing)' 26%; 'We absorb 100% of inference costs (not passed down to customers)' 15% (shortest bar). Bar lengths are proportional to the values (about 58, 26 and 15 units). No axis. Logo: ICONIQ wordmark in black at bottom-left, a flat image about 131 by 31 px, no icon.
Slide 20
AI GTM & Economics
Overall, companies are seeing unit economics improve moderately on a per-query or per-task basis by focusing primarily
on reducing inference costs, implementing routing strategies, and growing revenue for cost leverage
How have your unit economics on a per-query or per-task basis changed over
the last 12 months?
% of Respondents, Single-Select, N=303
47%
19%
22%
8%
3%
2%
Improved
significantly (cost
per query / task
down more than
30%)
Improved
moderately (cost
per query / task
down 10-30%)
Roughly flat
(change within +/-
10%)
Worsened (cost
per query / task
increased)
We do not track
cost at this level of
granularity
Too early to tell
Source: Perspectives from the ICONIQ GenAI Surveys (May 2026) and perspectives from the ICONIQ team and network of AI leaders consisting of our community of CIO/CDOs overseeing AI initiatives in enterprises, CTOs, our Technical Advisory Board, and others in our network
ICONIQ
Private & Strictly Confidential
20
Primary Levers For Improving AI Margins
Ranked by share of respondents placing each in their top 3
#1
Reducing inference costs
#2
Implementing routing strategies
#3
Growing revenue for cost leverage
#4
Switching to lower-cost / OSS models
#5
Raising prices for AI models
#6
Negotiating better provider pricing
#7
Reducing talent costs vs revenue
#8
Other
Visuals: Bar chart, vertical columns, titled 'How have your unit economics on a per-query or per-task basis changed over the last 12 months?' with subtitle '% of Respondents, Single-Select, N=303'. Six bars in #86BECA with printed values: Improved significantly (cost per query / task down more than 30%) 19%; Improved moderately (cost per query / task down 10-30%) 47%; Roughly flat (change within +/-10%) 22%; Worsened (cost per query / task increased) 8%; We do not track cost at this level of granularity 3%; Too early to tell 2%. No value axis, no gridlines, no legend; baseline is a thin light grey horizontal line. Bar values sit above each bar in black. No highlighted points beyond the tallest bar at 47%. Ranked panel is a dark teal rounded rectangle listing eight levers with numbers #1 to #8 in pale blue #AED3DB as shown in text. ICONIQ wordmark logo in black at bottom-left.
Slide 21
AI GTM & Economics
Revenue expansion and utilizing optimization levers has led to an increase in gross margin; margins are expected to
increase to ~60% on AI products in 2027
Gross Margin on AI Products
Average, N=287
59%
2027P
Projected
+6pp
53%
2026P
Projected
+8pp
High-growth companies are seeing higher gross margins in 2027 than non-high-growth companies (64% vs 58%)
45%
2025
Actual
+14pp total margin expansion | 2025 to 2027P
AI infrastructure / platform products have the highest predicted gross margin
2027P Average Gross Margin by Product Type
AI platforms / infrastructure 67%
Application layer products 60%
Core AI models / technology 57%
AI development tooling 55%
Consumer AI applications 48%
Source: Perspectives from the ICONIQ GenAI Surveys (May 2026) and perspectives from the ICONIQ team and network of AI leaders consisting of our community of CIO/CDOs overseeing AI initiatives in enterprises, CTOs, our Technical Advisory Board, and others in our network
ICONIQ
Private & Strictly Confidential
21
Visuals: Bubble progression chart titled Gross Margin on AI Products, subtitle Average, N=287. Three filled circles rise left to right, each with a light translucent outer ring: 45% 2025 Actual (light green #4DB96B, smallest, lowest), 53% 2026P Projected (mid green #1A7A3A), 59% 2027P Projected (dark green #033A15, largest, highest). Two pale mint pill labels, +8pp between the first and second circles and +6pp between the second and third, with dotted light grey connector lines running from circle to circle. Beneath the 59% circle is a bold-and-regular caption: High-growth companies are seeing higher gross margins in 2027 than non-high-growth companies (64% vs 58%). Right panel: dark green rounded rectangle with a horizontal bar chart titled 2027P Average Gross Margin by Product Type. Series values: AI platforms / infrastructure 67% (longest bar), Application layer products 60%, Core AI models / technology 57%, AI development tooling 55%, Consumer AI applications 48% (shortest bar). Bars are mid green #4DB96B rounded rectangles with lengths roughly proportional to value, values printed at the right end of each bar in green. Bottom summary bar reads +14pp total margin expansion | 2025 to 2027P in light green text on dark green. ICONIQ wordmark logo in black at bottom-left.
Slide 22
03
Talent &
Organization
ICONIQ
Private & Strictly Confidential
Visuals: One photograph, a generated-looking abstract illustration in a tall rounded-corner card at x 62.1%, y 11.9%, w 31.2%, h 76.2%. Subject: a large translucent dome or sphere built from fine glowing teal particles, with flowing luminous wave-like mesh lines running across its lower half and through the centre. Action: none, static abstract. Composition: portrait framing with the dome cropped at the top edge of the sphere's arc, filling the card; the card is not full-bleed and is boxed with rounded corners. Lighting: soft internal glow, bright cyan highlights on the mesh lines against darker teal. Dominant colours: teal #1A8A9A at the top, cyan #2FD0E0 highlights, deep teal #0F6F7E at the bottom. No overlay or tint. One logo: the ICONIQ wordmark, a small black sans-serif-style logotype image at bottom left, 131 by 31 pixels. One light grey number '03' at about 10% black opacity behind the section number.
Slide 23
Talent & Organizational Structure / AI-driven1 and high-growth companies tend to operate with more cross-functional structures, fewer layers, and wider spans of control / ORGANIZATIONAL STRUCTURE / AI-driven companies tend to have more cross-functional organizational structures / CURRENT ORGANIZATIONAL STRUCTURE BY AI REVENUE MIX / COMPANIES WITH CROSS-FUNCTIONAL STRUCTURE / Single-Select, N=297 / 50%+ AI revenue / 42% / <50% AI revenue / 35% / We explore more themes around the new ways that AI companies are structuring their organizations and re-defining the workplace here: / Org Charts to Outcomes: The New Operating Model for Talent / MANAGEMENT LAYERS / AI-driven companies are run on fewer layers between CEOs and most junior ICs / MANAGEMENT LAYERS BETWEEN CEO/FOUNDER TO MOST JUNIOR IC BY REVENUE MIX / COMPANIES RUN ON 1-4 LAYERS / $100M+ Companies2, Single-Select, N=156 / 50%+ AI revenue / 72% / <50% AI revenue / 56% / SPANS OF CONTROL / High-growth companies are widening first-line manager spans of control in 2026 / FIRST-LINE MANAGERS WITH 7+ DIRECT REPORTS / R&D MANAGERS WITH 7+ DIRECT REPORTS / High-Growth Companies, Single-Select, N=56 / 2025 / 21% / 2026 / 30% / GTM MANAGERS WITH 7+ DIRECT REPORTS / High-Growth Companies, Single-Select, N=43 / 2025 / 26% / 2026 / 33% / Source: Perspectives from the ICONIQ GenAI Surveys (May 2026) and perspectives from the ICONIQ team and network of AI leaders consisting of our community of CIO/CDOs overseeing AI initiatives in enterprises, CTOs, our Technical Advisory Board, and others in our network / 1 – Companies with 50%+ revenue driven by AI products; 2 – We expect earlier stage companies to be run on fewer layers and therefore, have focused this view on later stage companies / ICONIQ / Private & Strictly Confidential / 23
Visuals: Three horizontal bar-chart panels, each with paired bars on a shared track width and printed values, no axes or gridlines. Panel 1 (Organizational structure, Companies with cross-functional structure, Single-Select, N=297): 50%+ AI revenue 42% navy bar (about 42% of track filled, navy #262685) and <50% AI revenue 35% beige bar (about 35% filled, #C9C4B8 on track #ECE9E3). Panel 2 (Management layers, Companies run on 1-4 layers, $100M+ Companies, Single-Select, N=156): 50%+ AI revenue 72% navy bar (about 72% filled) and <50% AI revenue 56% beige bar (about 56% filled). Panel 3 (Spans of control, first-line managers with 7+ direct reports): R&D managers with 7+ direct reports, High-Growth Companies, Single-Select, N=56, rows 2025 21% beige bar (about 21% filled) and 2026 30% navy bar (about 30% filled); GTM managers with 7+ direct reports, High-Growth Companies, Single-Select, N=43, rows 2025 26% beige bar (about 26% filled) and 2026 33% navy bar (about 33% filled). Navy bars and 2026 values highlighted in larger blue type #3333B1; grey values #4D4D4D for comparison bars. Bar tracks are light warm grey #ECE9E3, bars are square-ended with no outline. The ICONIQ wordmark in black is the only logo, a 131 by 31 pixel image at bottom left.
Slides 24 to 47 continue in the full text file.
Download this deck
The hosted PDF of every library deck is free for readers of the SlideDeck Co newsletter. Join with your email and the viewer and download appear here.
Preparing a results presentation or business update of your own? The SlideDeck Co Standard Deck is a slide library for business reviews, updates and plans.






