The rise of AI multihoming, according to Ramp data

Every year, tens of billions in spending, from tens of thousands of businesses, flow through Ramp’s platform. Today we’re releasing our Fall 2024 Spending Report highlighting the latest trends. 

‍

AI spending continues to grow exponentially as companies infuse models and chat subscriptions into their operations and products. In Q3, Ramp customers spent over $20M across the 1500+ AI vendors we track, representing a 38% growth in card expenses year over year. Not only are more companies purchasing, but they’re also spending more: mean spending via cards and AP rose 18% and 31% respectively.  

‍

This quarter, we took a fresh look at the two largest vendors Ramp customers spent on: OpenAI and Anthropic. The data reveal how companies are responding to rapid advances in foundation models, experimenting with GenAI chat, and more. 

AI flexibility is the new competitive advantage 

In 2024, choosing an AI vendor isn’t a zero-sum game. 

‍

OpenAI is still the industry titan, but companies have seen the pace of innovation in AI and are expecting faster, cheaper, and more powerful models to continue to come onto the market. As a result, they are borrowing from the multicloud playbook and embracing a multihoming approach to AI to ensure maximum flexibility. They are setting up infrastructure to adopt multiple models and swap them with ease.  

‍

Our data on the number of OpenAI customers spending with Anthropic show just how quickly companies are moving to this new approach. At the start of 2024, only 3% of OpenAI customers used Anthropic, but this figure jumped to 13% in March after Anthropic released Claude 3, a trio of models with scaling capabilities and price points. Adoption spiked again after Claude 3.5 Sonnet rolled out in June and has only grown higher since: in September, 22% of OpenAI customers also used Anthropic.

‍

OpenAI customers’ adoption of Anthropic

‍

Cutting-edge models quickly gain market share

Companies are gravitating to Anthropic because its models are affordable and powerful, especially Claude 3.5 Sonnet. As more employees use AI for tasks like document summarization, businesses increasingly prefer models that can do this work at a lower cost, allowing them to save ​​higher-end, expensive models for more complex tasks. In response, OpenAI has also introduced cheaper models like GPT-4o. 

‍

Our data show Ramp customers quickly shifting their usage to Claude 3.5 Sonnet and GPT-4o. Since its launch four months ago, Claude 3.5 Sonnet has already reached 12% of spend across all OpenAI and Anthropic models, according to line item data in Ramp. In contrast, Sonnet’s more powerful—and more expensive—sibling, Claude 3 Opus, has hovered at 4% of spend, even though it’s been available longer.

‍

GPT-4o’s growth has been even more incredible. Since launching in May, the model has catapulted to an astounding 44% market share, while its predecessor, GPT-4, saw its share plunge from 70% to 23% over the same period.

‍

Foundation model spending on Ramp 

‍

‍

We can see the impact of lower-cost Anthropic models on customers’ spending. Back in September 2023, customers using both OpenAI and Anthropic models spent $749 on average with Anthropic. After Claude 3 came on the scene with its cheaper models, that amount dropped by more than half to $324 in May. The launch of Claude 3.5 Sonnet in June—which came with a huge upgrade in performance—has since caused spending to rise slightly and level out at around $460, as people migrated from competitor models like GPT-4. 

‍

Anthropic spending for Ramp customers using both OpenAI and Anthropic

‍

‍

Competition is heating up. Just last week, Anthropic announced a powerful new version of its cheapest model, Claude 3.5 Haiku. And open-source models like Nvidia’s Nemotron are introducing even more options for businesses. 

‍

By building for flexibility, companies can easily test new releases and switch to models that offer better performance and cost savings. 

Chat subscription spending grew 6x year over year

Not only are companies flocking to OpenAI and Anthropic to turbocharge their products with models, but they’re also using them to boost employee productivity with GenAI chat. Tools like ChatGPT and Claude’s chatbot are quickly gaining a foothold in the workplace as employees adopt them for everything from data analysis to image generation. According to a 2024 Glassdoor report, the percentage of workers using GenAI chat in the workplace has more than doubled since 2023. 

‍

We see this growth in Ramp data as well. The number of Ramp customers using chat subscriptions has risen 133% in the past year, and spending has grown even faster (616%).

‍

Chat subscription spending on Ramp

‍

The majority of the fastest-growing software vendors on Ramp are AI 

AI is fundamentally changing the way businesses run and grow—and companies powering this transformation are growing the fastest. 

‍

In Q3, six of the 10 fastest-growing software vendors on Ramp were AI tools, supporting Stripe’s recent finding that AI startups are growing 5X faster than past waves of SaaS companies. In particular, AI-assisted coding platforms—like Cursor, Supermaven, and Replit—are gaining traction. 

‍

Developer tools that have added AI capabilities to their services are also seeing rapid growth. Long-time virtualization platform Parallels saw over 7X customer growth after announcing new tools to help companies test AI models.

‍

‍

A look at the most popular AI vendors rounds out this picture. Unsurprisingly, OpenAI and Anthropic top the list of popular tools, while Flow.ai (part of customer engagement platform Khoros) saw explosive growth, indicating a newfound interest in support chatbots. 

‍

‍

With developer tools​​ gaining steam, engineering AI spending has risen the most year over year, as seen in our new cut of AI spending data by employee department. Also notable is the rise in AI spending with Human Resources (HR) and Finance teams, driven by productivity tools like Fireflies.ai and Otter. These areas may be especially hampered by repetitive work and ripe for AI assistants.  

‍

AI spending by department

‍

Come back tomorrow to see how businesses are controlling another big area of spend: travel. Later this week, we’ll also explore how AI is playing a role in retail ahead of the holiday shopping season. Check out more industry cuts on our updated spending microsite. 

‍

Methodology: Data come from millions of aggregated, anonymized transactions on Ramp cards and invoices paid through Ramp Bill Pay, as well as trusted third-party sources. For year-over-year comparisons, our sample comprises customers who have been active with Ramp over that entire 12-month period. Quarter-over-quarter comparisons comprise customers who have been active with Ramp over the two quarters analyzed.