The Rising Demand for AI Infrastructure: Insights from Together AI’s Recent Developments
The introduction of DeepSeek-R1 initially raised concerns within the sector regarding its potential to simplify advanced reasoning with minimal infrastructure. However, evidence suggests that this is not wholly accurate.
According to Together AI, the emergence of DeepSeek and deepseek-r1-outshines-openais-o1-with-unmatched-processing-power-and-cost-efficiency/” title=”Unleashing Innovation: How DeepSeek-R1 Outshines OpenAI’s o1 with Unmatched Processing Power and Cost Efficiency!”>open-source reasoning has, in fact, heightened infrastructure requirements rather than diminished them. This increasing demand is accelerating the expansion of Together AI’s platform and services.
Funding Success and Business Growth
Today marks a significant milestone for Together AI as they secured $305 million in a Series B funding round led by General Catalyst with Prosperity7 contributing as co-lead investor. Established in early 2023, their goal has been to make enterprise usage of open-source large language models (LLMs) easier. By 2024, they introduced the Together enterprise platform designed for deploying artificial intelligence within virtual private clouds (VPCs) and on-premises setups. Progressing into 2025, new features like reasoning clusters and agentic AI functionalities are set to enhance their offerings further.
The company boasts over 450,000 registered developers utilizing its deployment platform while experiencing an impressive sixfold annual growth rate. Their clientele includes various enterprises along with innovative startups such as Krea AI, Captions, and Pika Labs.
“We now support models across diverse modalities including text-based reasoning alongside images, audio clips, and video streams,” noted Vipul Prakash, CEO of Together AI during his conversation with VentureBeat.
DeepSeek-R1’s Impact on Infrastructure Needs
The launch of DeepSeek-R1 was transformative for multiple reasons; key among them was its suggestion that deploying an advanced open-source reasoning model could require less infrastructure compared to traditional proprietary models.
Prakash clarified that to accommodate the escalating demands linked to DeepSeek-R1 tasks; Together AI has indeed increased its infrastructural capabilities significantly.
“Running inference on this model incurs substantial costs due to its size—comprising 671 billion parameters—which necessitates distribution across numerous servers,” he explained. “Moreover, given its superior output quality—a higher caliber means greater demand at peak performance levels—significant capacity is essential.” He added that requests related to DeepSeek-R1 often have prolonged durations ranging from two to three minutes further compounding these needs.
Elevating Organizational Capabilities through Reasoning Models
Together AI observes tangible applications involving reasoning models across various domains:
- Coding Assistance: Reasoning models effectively decompose complex challenges into manageable steps.
- Mistake Reduction: Involving reasoned analysis helps validate outputs minimizing errors—critical for functions demanding high accuracy levels.
- Enhancing Non-Reasoning Models: Clients are focused on refining non-reasoning designs by leveraging insights from these advanced systems.
- Pioneering Self-Improvement: The integration of reinforcement learning allows ongoing self-enhancement without extensive reliance on manually annotated datasets.
A Surge in Demand Driven by Agentic AI Workflows
The trend towards adopting agentic workflows at organizations using Together AI’s technology contributes significantly towards increased infrastructural needs.
“In scenarios where one user request can trigger thousands of API interactions necessary for completing a task,” elaborated Prakash about agentic operations driving computational burdens within our system.”
//…additional content continues…
