As the AI revolution accelerates, the demand for Graphics Processing Units (GPUs) has skyrocketed, triggering a severe shortage that shows no signs of easing.
NVIDIA, the dominant player in high-end GPUs, has seen its Blackwell series sell out through 2025, with lead times stretching into years.
Analysts predict that by 2030, GPU supply could fall short by as much as 43% of projected demand, with training costs for individual AI models reaching $1 billion by 2027. This challenge is not limited to big tech—it’s a major barrier for startups, researchers, and enterprises seeking to innovate cost-effectively.
Amid this landscape, Bangalore’s tech ecosystem has become a critical hub for innovation. Pure Storage’s R&D and global capability centre (GCC) in India is spearheading enterprise AI initiatives, including the development of the pure key-value accelerator (KVA)—a protocol-agnostic, high-performance key-value caching solution designed to optimise large language model (LLM) inference.
By persisting and reusing precomputed attention states across sessions, Pure KVA eliminates redundant computation, delivering substantial performance gains without requiring changes to the underlying model or infrastructure.
The core idea behind Pure KVA is instead of discarding the key and value tensors after each inference session, the system captures these intermediate states, compresses them, and stores them on a high-performance Pure Storage NFS or S3 backend.
When the same prompt or context is reused, the stored tensors are quickly reloaded, bypassing unnecessary recomputation and dramatically improving efficiency for AI workloads.
GCC Growth and Investment in India
Talking to AIM, Nirav Sheth, VP – WW sales & customer success engineering at Pure Storage, highlighted the importance of India for the company: “I believe we have about 25% of our R&D function here in the GCC in India. Roughly, the team size is about 500 to 600.”
He added that the GCC is treated as a true hub for product development and innovation, not just a back office, “we’re seeing a tremendous amount of innovation. This Pure KVA, a fantastic optimisation opportunity for any customer looking at AI, is being developed in India.”
Pure KVA, one of Pure Storage’s flagship innovations from India, helps enterprises reduce AI infrastructure costs by optimising GPU usage.
“A very large cost of AI is actually within the GPU, which is further compounded by GPU availability. KVA helps customers reduce the cost of AI by optimising GPU utilisation,” Sheth said, adding that it leads customers to utilise their existing infrastructure more efficiently.
On its impact, Sheth mentioned that “for inferencing, based on some of the calculations we’ve seen, it could be 20X optimisation.”
As much as 70% of Pure KVA was developed in India. The company established an AI Center of Excellence within the GCC with data scientists, prioritising AI skill sets.
Ajeya Motaganahalli, VP engineering and MD India R&D, added they incubated a Gen AI team here. “We don’t have a similar team anywhere else. This team constantly looks at ways to make things better, faster, more cost-efficient for AI users.”
AI Partnerships
Pure Storage has built strong AI partnerships and leverages open-source large language models for enterprise needs.
“We have a world-class partnership with NVIDIA. We were the first in the industry to have a reference architecture with NVIDIA back in 2017. We also partner with RunAI, Weights & Biases, vector database providers like MongoDB, and infrastructure partners like Cisco and Arista,” Sheth said.
On building models for enterprise AI, Motaganahalli explained taking existing large language models, like Llama or Claude, and tuning them to internal data.
“We’re building small-scale models, while leveraging enterprise AI infrastructure for tuning and training. Open-source models have democratised AI, allowing enterprises to adapt and train models to their specific needs,” Motaganahalli said.
Talent Strategy
Pure Storage also emphasises nurturing talent through internships, ensuring strong retention and building a skilled workforce.
Motaganahalli explained, “When you come to a deep tech company like us, you spend a lot of time understanding how the development processes work, what the code base looks like, and how to write your unit tests. Internship gives this ability.” Most interns end up joining them as employees as the interview process remains the same for both, he added.
Elaborating on the GCC hiring ecosystem, Sheth said that Pure Storage GCC is inviting internships, hiring fresh graduates, as well as staff at junior and senior levels.
Meanwhile, Pure Storage recently announced its expansion in Enterprise Data Cloud to streamline AI workflows across on-premises and cloud environments.
Key updates include Pure Storage Cloud Azure Native, enabling seamless VMware workload migration, Portworx integration with Pure Fusion for unified data management, Pure1 AI Copilot for natural-language storage management; and Key Value Accelerator with NVIDIA Dynamo to speed AI inference. Next-gen FlashArray and Purity Deep Reduce further optimise performance and efficiency.
The post How Pure Storage India R&D Centre Built Pure KVA appeared first on Analytics India Magazine.
