500 logo
Harnessing Intelligence with DeepInfra: Delivering the Power of AI to All of Us

2026.05.04

500 Global Team

500 Global Team

Article image

 Why we co-led DeepInfra’s $107M series B

While much attention has been given to the large capex investments to train frontier models, we have been focused on the next stage of the AI value chain: inference. Inference is the process of taking the intelligence of the models and delivering it to end users and developers. Whether we reach artificial superintelligence, or whether the models stop improving today, the demand for inference will continue.

Inference is one of the most important and fasting growing steps in the AI supply chain, . Delivering timely output tokens in response to a prompt requires optimizing speed, throughput, cost and reliability. Often these are at odds with each other and require operational tradeoffs. That is why we like investing in this space: high value with high complexity.

It is also the reason we invested in DeepInfra. The team at DeepInfra have the experience to meet the unique demands required of inference. DeepInfra’s founding team, led by co-founders Nikola Borisov and Yessen Yessenzhar Kanapin, built and scaled global, real-time systems at messaging app imo, which serves 200M+ users around the world. They are one of the few teams that have experience running distributed, high-throughput infrastructure at massive scale.

We first invested in DeepInfra back in 2024. Since then, we have witnessed firsthand the team executing with operational and technical excellence.  Across markets, tokens per day - a key metric for inference - increased significantly, with Microsoft reporting a 5× year-over-year increase in token processing on Azure, and OpenRouter growing from 10 trillion to over 100 trillion tokens per year by mid-2025; business metrics followed, with enterprise spending on AI inference APIs more than doubling from $3.5 billion to $8.4 billion in the first half of 2025 alone.  Moreover, our market hypothesis only grew: open source models continued to compete with frontier models and drive a surge in demand for inference.

We are proud to co-lead DeepInfra’s Series B funding round alongside existing investors, including our friend and co-investor Georges Harik, co-founder of humans&. We are grateful to Nikola and Yessen for this unique opportunity to deepen our partnership. Together, they have built a small but mighty team on a mission to harness the power of AI and deliver intelligence to all of us.

Legal Notices and Disclaimers This article is intended solely for general informational or educational purposes only. 500 Startups Management Company, L.L.C. and its affiliates (collectively “500 Global”) makes no representation as to the accuracy or information in contained herein and while reasonable steps have been taken to ensure that the information herein is accurate and up-to-date, no liability can be accepted for any error or omissions. All third party links in this post have not been independently verified by 500 Global and the inclusion of such links should not be interpreted as an endorsement or confirmation of the content within. Under no circumstances should any content in this post be construed as investment, legal, tax or accounting advice by 500 Global, or an offer to sell or solicitation of interest to purchase any securities advised by 500 Global. Prospective investors considering an investment into any 500 Global fund should not consider or construe this content as fund marketing material. The views expressed herein are as at the date of this article and are subject to change without notice.