As AI investors fret over ROI, these startups attracted serious cash from customers in 2025

Brex analyzed credit-card and bill-pay transactions from more than 35,000 anonymized customers to assemble a weighted list of 2025’s 50 fastest-growing software vendors, excluding public companies and firms valued above $30 billion. The data shows extraordinary customer spend growth in AI-related tooling: Cursor ranked No.1 with ~1,000% year-over-year spending growth (spend compounding monthly), OpenRouter was No.2 with ~1,500% YoY growth, and video/visual AI tools such as Kling.ai (No.3), Ideogram (No.17) and Runway (No.44) also featured, while coding tools (Windsurf, Replit, CodeRabbit, StackBlitz) and infrastructure vendors (Vast.ai, Groq, Supabase, Sentry) dominated the list — a signal that customers are approving real budgets for no-friction, high-ROI AI solutions.

Analysis

Market structure: The Brex data (Cursor +1000% YoY, OpenRouter +1500% YoY across 35k customers) signals durable, demand-driven adoption for frictionless developer and visual-AI tools. Winners: AI infra (GPU/accelerator suppliers), developer tooling vendors, and niche vertical visual-AI providers that convert free trials to paid spend; losers: long-procurement legacy SaaS and generic LLM incumbents that can’t demonstrate fast dollar-based ROI. Expect upward pricing power for best-in-class tools and higher utilization on cloud GPU capacity, putting revenue upside pressure on NVDA-like suppliers and cloud IaaS margins.

Risk assessment: Tail risks include regulatory intervention in open model marketplaces, acute GPU supply shocks, or rapid enterprise budget reallocation if macro tightens — each could erode vendor spend by 30–60% within a quarter. Near term (days–weeks) look for confirmatory spend momentum; short term (3–6 months) monitors are vendor dollar-retention and MoM spend deceleration; long term (12–24 months) risks center on consolidation and margin compression as incumbents replicate functionality. Hidden dependency: Brex signals developer-led spending which can reverse faster than enterprise contracts when cost control returns.

Trade implications: Tactical: buy AI-infra exposure (NVDA) via 6–9 month call spreads to capture continued GPU demand; establish 1–2% long position in GOOGL (AI infra + model marketplace leverage) as a 12-month trade. Allocate 2–4% of risk capital to private/secondary exposure in paid coding-tool winners (Cursor/Replit) or funds that can access them. Defensively trim/avoid mid-cap public SaaS with >50% pilot revenue and EV/NTM revenue >10x; convert those weights into infra and PE/credit names like TPG (0.5–1% tactical long) benefiting from M&A activity.

Contrarian angles: Consensus overweights feature general-purpose generators; I’d prefer niche vertical tools that lock workflows (coding, video editing) where switching costs rise quickly — this suggests underappreciated optionality in small private vendors and in infra contract providers (spot capacity marketplaces like Vast.ai proxies). The market may underprice counter-cyclical demand resiliency: developer tools bought on cards indicate stickiness but also volatility — require two consecutive months of decelerating Brex-spend (<20% MoM) to flip to defensive posture. Historical parallel: 2010s SaaS froth corrected when ARR was found thin; avoid re-rating risk in companies that mix pilots/one-offs into ‘ARR.’

Key Decisions for Investors

• Establish a 1–2% portfolio long in NVDA via a 6–9 month 1x call spread (buy 15% OTM, sell 40% OTM) to capture continued GPU demand tied to developer tooling; exit or re-evaluate if NVDA implied vol falls >30% or earnings guidance misses by >5%.
• Add a 1–2% 12-month long position in GOOGL (or buy 12–18 month LEAP calls) to play model marketplace and cloud infra leverage; trim to zero if quarterly AI/cloud revenue growth decelerates to <10% YoY or aggregate capex guidance is cut >10%.
• Reallocate 2–4% of risk capital into private/secondary exposure to paid AI developer tooling (Cursor, Replit, Windsurf-type assets) via venture funds or secondaries; target companies showing ≥30% MoM paid-spend growth and >60% dollar retention before committing.
• Reduce/trim public mid-cap AI-native SaaS exposure by 30–50% where EV/NTM revenue >10x and >50% of sales are pilots/one-offs; redeploy proceeds into infra (NVDA/GOOGL) and PE credit (TPG ticker exposure 0.5–1%) that benefit from M&A and consolidated demand.
• Use a strict 30–60 day signal: cut risk to these AI trades if vendor-level real spend from Brex-like panels decelerates to <20% MoM for two consecutive months or if GPU spot prices fall >25% from current levels, indicating demand collapse.