Brex Benchmark: 2025 year-end report

At Brex, we’ve always believed the strongest signal in the market comes from customers themselves. And in 2025, their behavior told a powerful story — not through surveys or sentiment, but through what they actually spent money on. Across our 35,000+ customers, real purchasing patterns revealed which vendors became essential to how teams operate and grow.

For this end-of-year report, we built a ranking system using Brex card and Brex bill pay spend data to identify which vendors grew the fastest. We then weighted recent months more heavily than older ones, as momentum matters more than an early 2025 spike. We filtered out public companies and those valued at over $30 billion. We also smoothed out the noise — a vendor that went viral for one month doesn't outperform a vendor that has been consistently growing all year.

Introducing the 50 fastest-growing software vendors of 2025

AI is now infrastructure, and Cursor is the defining tool

1. Cursor (+1,000% YoY) was the fastest-growing vendor among Brex customers, with 1,000% year-over-year growth at a significant scale. Among Brex customers, spending on Cursor compounded every single month in 2025, a feat no other vendor in our data can claim. AI is making engineering teams spend more, not less. Teams are hiring more. Building faster. And allocating bigger budgets to speed.
   
   Cursor anchors a much larger movement across global business: AI as infrastructure. The rest of the top 5 fastest-growing vendors make that clear, from tools simplifying LLM model routing and deployment to automating voice, video, and search.
2. OpenRouter (+1,500% YoY) is the connective tissue for how developers access and compare frontier models, making multi-model workflows dead simple. There’s no deciding between LLMs; OpenRouter automatically routes to the best one for each job.
3. Kling AI ( +1,900% YoY) has redefined what “video production” means, letting anyone generate cinematic clips from text and image prompts. Its rapid quality improvements are pulling AI video from “experimental” into the standard creative toolkit.
4. Retell AI (+400% YoY) powers human-like customer service interactions and competes in a crowded and well-funded AI voice market. Its natural pacing and accuracy make Retell a trusted voice tool in real customer-facing workflows.
5. Perplexity (+300% YoY) is changing internet search one query at a time. At 300% YoY growth, we’ll need to verbify it soon (“just Plex it”?) because with fast, sourced answers and conversational results, it’s becoming the default way to find information.

In 2025, the shift was clear: AI moved from experiment to infrastructure. From “should we?” to “how do we?” The top 50 list is proof of that shift from different angles. Here are the 5 trends that stood out to us:

1. The AI boom starts in the code.

Developers are the biggest AI spenders. And they will spend even more next year.

These coding tools are already creating undeniable value, suggesting chat was the spark for the AI economy, but code is the engine. AI is baked into how software gets built.

What it means:

It's not about which model is better. It's about where the tool actually lives in value-added workflows. An IDE-native tool that understands your codebase and sits right there while you're coding? That's something developers will pay for, and something managers will approve spending on.

2. The early AI winners are selling shovels.

Picks and shovels are selling out in the AI gold rush.

We saw startups snapping up AI infrastructure in our July Brex Benchmark, but it’s not just about compute anymore. The real momentum is in the orchestration layer — tools that connect every part of the stack, from LLM access to inference, databases, and deployment.

Because we’re early in AI, these tools are effectively infrastructure-as-a-product, and we’re still building the railroad. We expect a shift in 2026 from building the rails to running on them, and we’re excited to see some killer apps emerge.

What it means:

Brex customers standardize the rails before they try to run trains. The moat isn't the model you use. The moat is orchestration. Access. Routing. Evals. That's where the companies building in AI for the long term are investing.

3. You’re hearing AI everywhere.

Voice AI crossed the threshold.

We highlighted this trend in our June Brex Benchmark, and it’s only accelerating. Voice AI is now capable of handling repetitive or transactional conversations in a remarkably human-like way, making it a no-brainer for customer support, virtual agents, and product interactions.

Three voice AI companies made the top 50, and Deepgram and ElevenLabs were also top-10 startup AI tools in the Brex Benchmark for most of 2025. Voice is becoming a core feature rather than a nice-to-have.

What it means:

Here's what's happening: Teams are taking workflows that used to require humans handling customer support calls, qualifying leads, taking orders and instead running them through voice agents. Not perfectly. But well enough. Well enough that it's cheaper and faster than the alternative. That's the real shift. And it's only accelerating.

4. The dev teams that see more, ship more.

Developers move fast, but only if they can see what’s working and what’s breaking.

Modern development teams are locking in a baseline stack of analytics, monitoring, deployment, and project management to keep the code, the team, and the product in sync. The top solutions in this stack give developers instant feedback so they can fix issues and ship faster.

What’s cool is how much these tools are shaping developer culture. They tend to bake in best practices and guide teams toward how they should work rather than letting everyone reinvent the wheel.

What it means:

Teams aren't loyal to legacy tools anymore. They're switching because the new stack is purpose built to help teams ships faster. The old way: pick five different tools, spend months wiring them together, hope they talk to each other.

The new way: pick tools built for speed and interoperability from day one. They're designed to get out of your way. They're opinionated about how fast teams work. And they don't require engineering effort just to make them function together. That they're built to help you ship, and that wins spend.

5. There’s a whole meeting slop economy now.

Everyone’s trying to fix meetings, or at least make them hurt less, with AI summaries.

Three companies just proved something: you can build a billion-dollar business by making meetings actually productive.

Fireflies captures everything so you don't have to choose what to remember. Granola surfaces the insights so nothing important gets buried. Fathom tracks commitments so action items don't die in Slack.

The old way: meetings were a tax on productivity. The new way: meetings become searchable, reviewable, actionable. Teams get the value without the waste.

What it means:

The inflection point for voice AI wasn't when the tech got good. It was when teams realized it wasn't about saving money. It was about handling more conversations better. Letting humans focus on getting things done, not taking notes.

Crossing the threshold: From exploration to acceleration

What this year’s data makes undeniable is that AI has crossed a threshold, and it’s becoming the foundation for how companies build, scale, and operate as well as manage spend.

The vendors in this top 50 represent that shift from every angle: the tools embedded directly in the coding experience, the orchestration layers tying together increasingly complex AI stacks, the voice systems taking on real customer interactions, and the platforms turning meetings and analytics into actionable intelligence.

Brex customer spend shows a clear pattern: instead of cautiously exploring AI, companies are reorganizing their workflows around it. And once teams start relying on AI to move faster, see more, and ship continuously, there’s no going back to the slower way of working.

2025 set the groundwork. In 2026, we’ll see who learns to run at full speed.

...

Ranking methodology

We developed a scoring system utilizing Brex spend data to forecast which software vendors are likely to thrive in 2026 based on their growth in 2025.

Four principles power it:

1. Recency weighting: Last quarter matters more than last year (recent data gets 100% weight, a year ago just 25%).
2. Sample size skepticism: A vendor with five customers showing 90% retention? We pull that toward industry baseline until we see more data.
3. Signal vs. noise: We smooth monthly volatility to surface real trends, not random blips.
4. Customer-adjusted impact: We distinguish between vendor quality and customer quality. A vendor retaining customers who churn often from new products is worth more than those retaining customers who typically lock in to purchases for the long term.

The system blends observable metrics (spend, growth) with hidden impact (retention lift), weighting each by data volume, and gives us the ability to see ahead of the curve on who will continue to win spend.

Exclusions: We excluded vendors that are publicly traded, more than $30B in value or over $1B in revenue as of 10/31/25.

Methodology and privacy: All analysis is based on anonymized and aggregated Brex customer spend from 1/1/25 until 10/31/25. Rankings reflect real card and bill pay transactions across Brex’s customer base. We never share customer‑identifiable information. For questions about how we use data in anonymized or aggregated form for trend reporting, email privacy@brex.com.

Sumeet Marwaha is the Head of Data at Brex, supporting Brex in understanding how customers spend, adopt tools, and grow their businesses.