Creative tools and coding platforms now count as productivity software, as AI turns specialised skills into company-wide capabilities. Fresh data tracks where dollars flow.
What’s happening: Andreessen Horowitz analysed spending data from over 200,000 Mercury customers between June and August 2025 to identify the top 50 AI applications startups are actually purchasing.
Why this matters: AI startups raised an unprecedented $33 billion in the first half of 2024 AI startups set record with $33 billion raised in H1 2024, making it critical to understand where those funds flow and which applications prove their worth beyond initial testing phases.
The artificial intelligence spending patterns of early-stage companies look remarkably different from what web traffic data suggests. While consumer-facing metrics track curiosity and experimentation, financial data reveals commitment.
Andreessen Horowitz partnered with Mercury, the fintech provider serving over 200,000 startup customers, to analyse three months of spending data from June through August 2025. The resulting top 50 ranking identifies which AI-native application companies startups are genuinely paying for, offering a real-time signal of where AI delivers measurable returns.
The methodology deliberately excluded cloud services providers like Azure, GPU vendors such as Coreweave, and infrastructure tools, focusing instead on application-layer companies where AI directly impacts products and workflows. The data captures transactions via ACH, card spend and wires through Mercury accounts.
Horizontal tools lead
Horizontal applications, which boost productivity across entire companies rather than targeting specific roles, claimed 60 per cent of the list. OpenAI secured the top position, followed by Anthropic at number two and Perplexity at number 12. Optus partners with Perplexity to offer eligible mobile customers free 12-month access to advanced AI productivity tools Optus expands AI access for businesses with Perplexity Pro subscription, signalling mainstream business adoption of these general assistants.
The race for AI workspace dominance remains wide open. Notion ranked 10th while Manus appeared at 33rd, suggesting users haven’t settled on a single winner for accessing large language models within their existing file systems. Users appear to switch between different interfaces and models depending on specific needs.
Meeting support tools represented another major horizontal category. Notetaking applications Fyxer (number seven), Happyscribe (36), Plaude (38), Otter AI (41) and Read AI (49) all made the ranking. The category is expanding beyond transcription, with Cluely at number 26 offering real-time, in-meeting feedback.
Creative tools emerged as the single largest category with ten representatives. All-in-one suite Freepik claimed fourth position, whilst text-to-speech generator ElevenLabs secured fifth place. Image applications Canva, Photoroom and Midjourney appeared alongside video tools Descript, Opus Clip and Capcut. Avatar platforms Arcads (47) and Tavus (50) rounded out the category.
The data suggests AI has democratised previously specialised skills. Marketing and design departments once monopolised creative tools, whilst engineering teams dominated coding platforms. Now these applications spread across entire organisations, with employees in any role accessing capabilities once limited to specific functions.
Vertical apps augment, not replace
Vertical applications, which target specific roles, accounted for 40 per cent of the list. Of the 17 vertical companies identified, 12 focus on supercharging human employees rather than replacing them entirely.
Five aim to serve as AI employees completing workflows end to end: Crosby Legal (27) operates as an agentic law firm, Cognition (34) provides AI engineering, 11x (37) offers automated go-to-market employees, Serval (39) runs AI IT service desks, and Alma (42) delivers AI-powered immigration law services.
The researchers expect more end-to-end agentic products to emerge, particularly amongst Mercury’s customer base. New startups avoid being locked into multi-year contracts with expensive service providers, instead choosing to hire AI from the outset.
Three vertical categories dominated the rankings. Customer service tools included Lorikeet (eight), Customer.io (14), Ada (40) and Crisp (46). Sales and go-to-market applications featured Instantly (13), Clay (25) and 11x (37). Recruiting and human resources platforms comprised Micro1 (nine), Metaview (19) and Applaud (43).
Two companies fell into a broader operations category: Delve (11) for compliance automation and Combinely (29) for accounting functions.
Vibe coding goes enterprise
Four companies targeting AI-powered application building made the list: Replit, Cursor, Lovable and Emergent. Beyond the top two model providers, Replit claimed third position overall as an agentic product development tool.
The revenue comparison between Replit and Lovable proved particularly revealing. On the consumer top 100 list based on web traffic, Lovable reached the top quarter whilst Replit ranked lower. The enterprise spending data reversed this order dramatically, with Replit generating approximately 15 times more revenue than Lovable amongst Mercury customers.
The disparity reflects different product approaches. Lovable focuses on rapid user interface and component generation, providing a lower barrier to entry for pure consumers. Replit extends beyond front-end design to enable development of enterprise-grade, fully functional applications, agents and automations. Its Agent can run autonomously for hours, whilst built-in cloud services including databases, authentication and secure publishing operate directly within the platform. Enterprise controls make Replit’s feature set more suitable for company deployments.
The evolution of vibe coding remains uncertain. The space might fragment through platforms developing different application types, or it could converge with one winner dominating the enterprise realm.
Consumer products move up
Nearly 70 per cent of companies on the list allow individual adoption without requiring enterprise licences. Twelve companies appeared on both the enterprise spending ranking and the most recent consumer top 100, which tracks primarily business-to-consumer AI product traffic.
Of these 12, eleven started as individual products before evolving to offer team or enterprise functionality. Several continue generating consumer-majority revenue, including Cluely (26) and Midjourney (28). OpenAI followed this pattern, with 75 per cent of revenue coming from consumers as of October 2024, though recent estimates suggest a shift closer to a 50/50 split.
AI-powered consumer products deliver unprecedented power whilst serving enterprise use cases simultaneously. The urgency around AI adoption to boost employee efficiency pulls these tools into enterprise faster than previous software generations.
Excelling at product-led growth and moving upmarket has become viable within a company’s first year or two, contrasting sharply with prior software eras. Future editions of the enterprise spending list will likely feature more companies starting consumer-first.
The spending data reinforces a pattern: startups don’t just test AI tools, they commit budgets to platforms demonstrating clear returns. As the AI sector nears 2024 full-year funding with 4,835 deals AI sector nears 2024 full-year funding with 4,835 deals, understanding where those investments flow provides crucial insight into which applications move beyond hype to deliver genuine business value.
Keep up to date with our stories on LinkedIn, Twitter, Facebook and Instagram.
