DeepSeek-R1: Rising GPU Demand and AI Infrastructure Growth
In an era where artificial intelligence continues to reshape industries, Together AI is making waves with its groundbreaking advancements in reasoning models. Recently securing a staggering $305 million in Series B funding, the company has demonstrated that the demand for AI infrastructure is not dwindling; rather, it is escalating. This surge in infrastructure needs is largely driven by the emergence of DeepSeek-R1, an open-source reasoning model that, contrary to initial expectations, requires more computational resources than ever before. As Together AI expands its offerings to meet this growing demand, the implications for businesses and AI development are profound.
| Category | Details |
|---|---|
| Company Name | Together AI |
| Funding Amount | $305 million |
| Funding Round | Series B |
| Lead Investors | General Catalyst, Prosperity7 |
| Founded Year | 2023 |
| Current Year | 2025 |
| Registered Developers | Over 450,000 |
| Business Growth | 6X year-over-year |
| Key Customers | Krea AI, Captions, Pika Labs |
| Key Product | DeepSeek-R1 |
| Model Parameters | 671 billion |
| Infrastructure Demand | Increased due to DeepSeek-R1 |
| Reasoning Clusters | Dedicated capacity from 128 to 2,000 chips |
| Applications of Reasoning Models | Coding agents, reducing hallucinations, improving models, self-improvement |
| Agentic AI Influence | Increased API calls per task |
| Recent Acquisition | CodeSandbox |
| New Technology | Nvidia Blackwell GPU |
| Performance Increase | 2X performance, 25% cost increase |
| Competitive Landscape | Competition from Microsoft, AWS, Google, Groq, Samba Nova |
| Unique Offering | Full-stack GPU infrastructure with software layers |
The Rise of DeepSeek-R1 and Its Impact on AI Infrastructure
DeepSeek-R1 has changed the way we think about AI infrastructure. When it first appeared, many believed that advanced reasoning models would require less computing power. However, the opposite has proven true. Together AI shows that the demand for infrastructure has actually increased as more companies want to use DeepSeek-R1. This model, with its 671 billion parameters, needs a lot of servers to run effectively, creating a greater need for powerful computing resources.
As DeepSeek-R1 becomes more popular among developers and businesses, the infrastructure supporting it must grow too. Together AI has introduced new services, like reasoning clusters, to meet this rising demand. These clusters allow for dedicated capacity, ensuring that users can access the high-performance AI they require. With over 450,000 developers already registered on the Together AI platform, the growth shows no signs of slowing down.
How Together AI Supports Reasoning Models
Together AI is making it easier for organizations to use reasoning models effectively. These models help break down complex problems into manageable steps, which is especially useful in coding and data analysis. Moreover, reasoning models help verify outputs, reducing the chance of errors known as hallucinations. This is crucial in fields where accuracy is vital, such as healthcare and finance, where a small mistake can lead to significant consequences.
In addition to improving accuracy, Together AI’s reasoning models facilitate self-improvement. By using reinforcement learning, these models can learn from their experiences without needing a lot of human input. This capability allows businesses to develop smarter AI tools that can adapt and grow over time, making them even more valuable. This focus on enhancing reasoning models sets Together AI apart in the rapidly evolving AI landscape.
The Role of Agentic AI in Driving Infrastructure Demand
Agentic AI is another exciting area where Together AI is seeing increased infrastructure demand. This technology allows a single user request to trigger numerous tasks through API calls, which requires substantial computing power. As more companies adopt agentic workflows, the need for robust infrastructure grows. Together AI has acquired CodeSandbox to help manage these demands by providing lightweight virtual machines that execute code efficiently.
With the integration of CodeSandbox technology, Together AI can reduce the time it takes to connect agentic code with the necessary models. This improvement enhances the overall performance of agentic workflows, making them faster and more reliable. As businesses look for ways to streamline operations and improve efficiency, Together AI’s ability to support agentic AI is likely to attract more users eager to leverage this powerful technology.
Nvidia Blackwell: Advancements in AI Performance
Nvidia’s latest chip, the Blackwell GPU, is making waves in the AI industry by significantly boosting performance. As Together AI deploys these new chips, they are able to provide faster processing power, essential for running complex models like DeepSeek-R1. Although the Blackwell chips come at a higher cost, the 2X performance increase justifies the investment, helping companies achieve better results in their AI applications.
The ability of Blackwell chips to handle mixture of expert (MoE) models across multiple servers means that Together AI can support larger and more complex AI tasks. This technological advancement is critical as the demand for AI solutions continues to rise. By staying ahead with cutting-edge hardware, Together AI ensures that it remains competitive in a crowded market, ultimately benefiting its users with faster and more efficient AI capabilities.
Navigating the Competitive Landscape of AI Infrastructure
The AI infrastructure market is highly competitive, with major players like Microsoft, AWS, and Google all offering their own platforms. In addition to these tech giants, new startups such as Groq and Samba Nova are emerging, each vying for a piece of the lucrative AI market. Together AI distinguishes itself with a full-stack offering, combining powerful GPU infrastructure and a user-friendly software platform that enables easy access to open-source models.
By focusing on research and optimization, Together AI develops improved runtimes for both training and inference. This commitment to performance shows in their service capabilities, where they can process models much faster than competitors. For example, their DeepSeek-R1 model operates at 85 tokens per second, far surpassing Azure’s 7 tokens per second. This significant difference in performance allows Together AI to provide better cost-effective solutions for their customers.
Frequently Asked Questions
What is Together AI and what do they do?
Together AI is a company focused on simplifying the use of open-source large language models (LLMs) for enterprises, helping them deploy AI in private and on-premises environments.
How has DeepSeek-R1 affected GPU demand?
DeepSeek-R1 has increased GPU demand instead of reducing it, requiring more infrastructure to handle its complex tasks and higher performance needs.
What are reasoning clusters?
Reasoning clusters are dedicated capacities that Together AI provides, ranging from 128 to 2,000 chips, to efficiently run demanding models like DeepSeek-R1.
What benefits do reasoning models provide?
Reasoning models help break down complex problems, reduce hallucinations for better accuracy, and enable models to self-improve without needing extensive human data.
Why is agentic AI important for infrastructure?
Agentic AI increases infrastructure demand because it creates many API calls from a single task, requiring more computing power for efficient execution.
What role do Nvidia Blackwell chips play?
Nvidia Blackwell chips enhance performance for AI tasks, offering double the capabilities of previous models, making them ideal for training and inference.
How does Together AI compete in the AI market?
Together AI competes by offering a full-stack platform with GPU infrastructure and unique optimizations, positioning itself against major cloud providers and other startups.
Summary
Together AI has announced a significant $305 million funding round to expand its AI platform, driven by rising demand for advanced reasoning models, particularly DeepSeek-R1. Contrary to initial fears, these models are increasing rather than decreasing the need for powerful infrastructure. The company now supports over 450,000 developers and boasts a six-fold growth rate. Their reasoning models enhance coding processes, reduce inaccuracies, and enable self-improvement. The introduction of new Nvidia Blackwell chips aims to boost performance, positioning Together AI competitively against major cloud providers in the evolving AI landscape.
