The introduction of advanced reasoning models, particularly DeepSeek-R1, has significantly altered the landscape of artificial intelligence infrastructure demand. Despite initial industry speculations that these models would reduce the need for extensive computational resources, the proliferation of DeepSeek-R1 and similar models has instead driven a heightened need for robust infrastructure. Together AI’s impressive $305 million Series B round of funding highlights a substantial investment in ensuring and expanding the infrastructure required to support these advanced AI models.
The Rise of Advanced Reasoning Models
DeepSeek-R1 and Industry Concerns
When DeepSeek-R1 was launched, there were widespread concerns in the industry that advanced reasoning models, especially open-source ones, would reduce reliance on extensive computational infrastructures. However, the reality has been quite the opposite. As these models grew in capability and usage, they required even more substantial infrastructure support. Together AI seized this opportunity, capitalizing on the increased demand for AI infrastructure to drive its growth and market presence. The introduction of DeepSeek and similar models has underscored the importance of investing in AI infrastructure to meet the growing computational needs.
The rise of DeepSeek-R1 has demonstrated that advanced reasoning models often demand robust and scalable infrastructure, rather than reducing dependency on it. The industry’s initial expectation that open-source reasoning models would operate with less infrastructure has been challenged by the actual operational requirements. As DeepSeek-R1 comprises 671 billion parameters, it requires distribution across multiple servers to function effectively. This extensive infrastructure ensures the delivery of high-quality outputs, which further boosts demand and necessitates increased capacity. Together AI has responded proactively to these demands, solidifying its position in the competitive AI landscape.
Together AI’s Rapid Growth
Founded in 2023 with the mission to simplify the enterprise adoption of open-source large language models (LLMs), Together AI has experienced rapid growth. The company expanded its offerings in 2024 to include AI deployment in virtual private cloud (VPC) and on-premises environments. This strategic expansion reflects Together AI’s commitment to supporting the computational needs of advanced reasoning AI and its ability to adapt to evolving market demands. By 2025, the company aims to further enhance its platform to incorporate reasoning clusters and agentic AI capabilities, underscoring its dedication to innovation and scalability.
Together AI’s impressive growth metrics include a developer base exceeding 450,000 registered users and a year-over-year business growth of sixfold. These figures illustrate the increasing market demand for robust AI deployment platforms and highlight the company’s ability to attract a diverse clientele. From large enterprises to emerging AI startups like Krea AI, Captions, and Pika Labs, Together AI’s platform is catering to a wide range of customers. The company’s rapid scaling and strategic expansions are a testament to its effectiveness in meeting the growing computational needs of advanced reasoning AI.
Meeting the Demand for Infrastructure
Developer Base and Clientele
Together AI’s growing developer base and diverse clientele underscore the market’s escalating demand for robust AI deployment platforms. With over 450,000 registered users, the company’s platform has become a vital resource for developers seeking to harness the power of advanced reasoning models. The clientele ranges from large enterprises to innovative AI startups like Krea AI, Captions, and Pika Labs, showcasing the platform’s versatility and broad appeal. This diverse user base is not only a testament to Together AI’s ability to meet varied market needs but also a clear indication of the increasing importance of scalable AI infrastructure.
The company’s year-over-year business growth of sixfold further highlights the surging demand for AI deployment platforms. Together AI’s platform has been instrumental in helping businesses implement advanced reasoning models, driving efficiency and innovation. This growth is a reflection of the company’s strategic focus on developing and enhancing infrastructure capable of supporting the extensive computational requirements of modern AI applications. As organizations continue to adopt advanced reasoning models, the demand for scalable and robust AI infrastructure is expected to rise, positioning Together AI for continued success.
Resource-Intensive Operations
Running inference on DeepSeek-R1 is notably resource-intensive, necessitating a robust infrastructure capable of supporting its extensive computational requirements. The model, which comprises 671 billion parameters, requires distribution across multiple servers to function effectively. This immense demand is further compounded by the model’s ability to deliver higher quality outputs, which invariably attracts more significant demand and necessitates increased capacity. As a result, the infrastructure supporting DeepSeek-R1 must be scalable and robust to accommodate these heightened demands.
The operational demands of DeepSeek-R1 are particularly noteworthy for their duration. The requests processed by the model tend to be longer-lived, often lasting between two to three minutes. This extended processing time further drives the need for substantial infrastructure to ensure consistent performance and reliability. Together AI has addressed these demands by investing in scalable infrastructure and innovative solutions, enabling the model to operate efficiently and effectively. This commitment to supporting resource-intensive operations underscores Together AI’s role in meeting the growing infrastructure needs of advanced reasoning models.
Innovations in AI Infrastructure
Reasoning Clusters and Agentic AI
To address the extensive infrastructure demands of advanced reasoning models like DeepSeek-R1, Together AI introduced “reasoning clusters,” a specialized service designed to optimize the performance of AI workloads. These clusters provision capacity ranging from 128 to 2,000 chips, ensuring that users have access to the necessary resources to run their AI models efficiently. This innovative solution aligns with Together AI’s goal of enabling high-performance AI deployments, providing users with scalable and robust infrastructure to support their computational needs.
Reasoning clusters have proven particularly beneficial for various key areas of reasoning model usage. For instance, coding agents utilizing reasoning models can break down larger problems into smaller, sequential steps, significantly enhancing problem-solving capabilities. This feature is invaluable for developers working on complex coding tasks, allowing them to tackle challenges more effectively. Additionally, reasoning models play a crucial role in reducing hallucinations by verifying output models, ensuring accuracy and reliability in applications where precision is paramount. The introduction of reasoning clusters underscores Together AI’s commitment to addressing the evolving needs of AI developers and users.
CodeSandbox Acquisition
To further enhance its infrastructure and address the growing computational demands of advanced AI models, Together AI strategically acquired CodeSandbox. This acquisition is pivotal in optimizing the performance of agentic AI workflows, which involve a single user request generating thousands of API calls to achieve a task. These workflows place considerable computational demands on Together AI’s infrastructure, necessitating innovative solutions to ensure efficiency and reliability. CodeSandbox’s technology, which provides lightweight, fast-booting virtual machines (VMs), plays a crucial role in reducing latency and enhancing performance.
The integration of CodeSandbox’s technology within the Together AI cloud has significantly improved the performance of agentic workflows. These VMs can execute arbitrary, secure code swiftly and efficiently, ensuring that the computational demands of agentic AI workflows are met. These enhancements are essential for supporting the increasing use cases and applications of advanced reasoning models. By incorporating CodeSandbox’s technology, Together AI has bolstered its infrastructure capabilities, ensuring that it can meet the growing demands of its users and maintain high performance across its platform.
Competitive Landscape
Performance Demands and Nvidia’s Role
The surge in demand for AI infrastructure has been accompanied by escalating performance requirements, a trend that companies like Nvidia have continuously addressed with higher-performing silicon solutions. One of Nvidia’s latest products, the Blackwell GPU, has been deployed at Together AI. Despite being approximately 25% more expensive than its predecessor, the Blackwell GPU offers twice the performance, making it particularly well-suited for training and inference of mixture of expert (MoE) models across multiple InfiniBand-connected servers. This substantial performance boost is expected to be instrumental in supporting the inference of larger AI models, further underscoring the importance of advanced infrastructure.
Nvidia’s contributions to AI infrastructure are critical in meeting the heightened performance demands of modern AI applications. The deployment of Blackwell GPUs at Together AI exemplifies the company’s commitment to leveraging the latest technological advancements to enhance its platform capabilities. This strategic alliance with Nvidia ensures that Together AI can provide its users with state-of-the-art infrastructure, enabling them to run complex AI models efficiently and effectively. As performance demands continue to escalate, the role of advanced silicon solutions in maintaining high-performance AI systems will remain crucial.
Distinguishing Factors
The rise of advanced reasoning models, such as DeepSeek-R1, has significantly shifted the demand dynamics for artificial intelligence infrastructure. Contrary to early industry predictions that these models would diminish the necessity for extensive computational resources, the widespread adoption of DeepSeek-R1 and similar models has instead escalated the demand for solid infrastructure. Together AI’s impressive $305 million Series B funding round underscores a substantial commitment to ensuring and expanding the infrastructure needed to support these cutting-edge AI models. This substantial investment not only highlights the industry’s recognition of the crucial role that robust infrastructure plays but also signals a long-term commitment to advancing AI capabilities. As these models continue to evolve and become more integrated into various applications, the necessity for continuous investment in superior computational infrastructure will only grow, ensuring these powerful AI models can function at their highest potential.