Inside UltraQLC: Sandisk’s AI-Era Enterprise SSD Platform

Table of Contents

💾 Inside UltraQLC: The Enterprise SSD Platform Engineered for AI

When Sandisk’s top engineers set out to design a next-generation enterprise SSD, the goal was not incremental improvement. Instead, they aimed to build an entirely new platform from the ground up, purpose-built for the demands of AI-era data centers.

That vision took center stage this week at Future Memory and Storage (FMS), where Sandisk previewed its 256TB NVMe™ enterprise SSD, powered by the newly unveiled UltraQLC™ platform. Behind this announcement lies a multi-year engineering effort defined by architectural risk, cultural change, and a clear bet on ultra-high-capacity storage.

🎯 A Bold Mandate from Day One
#

The UltraQLC journey began in 2021, when Sandisk Chief Product Officer Khurram Ismail issued a clear directive to Vice President of Engineering Tsiko Shohat Rozenfeld:
break away from existing product lines and create a best-in-class enterprise SSD platform for the next decade.

Sandisk had enjoyed nearly ten years of success in the client market by extending a single strong platform into multiple derivatives. But Shohat understood that enterprise storage—especially at hyperscale—required a fundamentally different approach.

“Trying to build something new from scratch in the enterprise space was a huge and ambitious goal; monumental from an engineering accountability perspective.”

Shohat assembled a hand-picked team of subject matter experts from across client, consumer, and enterprise divisions. The intent was to leverage every past success—while deliberately discarding legacy assumptions that no longer scaled.

🏗️ Building a Platform from Scratch
#

Creating a new platform meant isolating top engineers from revenue-driving products and dedicating them to a future-facing effort—a risky but strategic move.

“Building a platform from scratch is a little like that first shot in golf,” Shohat explained.
“You want to get as close to the goal and as far as possible with that first attempt.”

For several years, the team debated architectural trade-offs, invented new approaches, and stress-tested assumptions. Once the silicon architecture solidified, the challenge shifted to productization—turning an ambitious design into a validated, customer-ready platform.

🧩 Productization at Scale
#

That phase was led by Ilya Gusev, Senior Director of Systems Design Engineering and UltraQLC Product Development Team Lead.

UltraQLC required coordination across hundreds of engineers spanning more than a dozen disciplines: ASIC design, firmware, hardware validation, memory systems, packaging, mass production, system architecture, and more.

“Building a platform from scratch is, by definition, painful and challenging,” Gusev said.
“It’s also about changing a mindset—breaking from legacy thinking.”

🔁 Rethinking Firmware and Team Structure
#

For Hyuk-Il Kwon, Senior Director of Firmware Engineering, the challenge was as much organizational as technical. He reshaped teams to blend talent from across Sandisk’s portfolio, ensuring firmware development aligned tightly with the new platform’s goals.

“It’s a rare opportunity to build something new from scratch,” Kwon said.
“There’s excitement—but also enormous responsibility.”

📦 The Strategic Pivot: Ultra-High Capacity QLC
#

A critical inflection point came when Sandisk decided to focus UltraQLC on ultra-high-capacity QLC SSDs.

With AI data centers now managing exabytes of data, hyperscalers increasingly view QLC SSDs as a compelling replacement for HDDs in AI data lakes.

“Focusing on QLC ultra-high capacity was the toughest but most impactful decision,” Kwon noted.

This clarity allowed teams to optimize for what mattered most:

Massive capacity
High throughput
Power efficiency
Predictable performance at scale

⚙️ Architecting for PCIe Gen 5 and AI Workloads
#

AI data lakes demand fast access to vast datasets, making PCIe Gen 5 a non-negotiable requirement.

Critical data paths were automated into hardware
ASIC and firmware were co-designed to maximize interface bandwidth
Performance-per-watt became a first-class design metric

However, ultra-high capacity introduced new challenges.

🔄 Managing Data at 256TB Scale
#

As Mike James, Senior Director of Enterprise SSD Systems Architecture, explained, scale changes everything.

“You can’t overwrite 128 terabytes every few days—it’s not effective or efficient.”

At these capacities, background operations like NAND recycling become system-level problems. The UltraQLC team developed new strategies to:

Reduce unnecessary data movement
Truncate recycling operations
Minimize performance impact during background maintenance

Rather than tuning isolated algorithms, the team rethought how I/O itself is managed.

“You can tweak a single algorithm—or you can change the entire approach,” James said.

🚀 From Platform to Product
#

UltraQLC is not a one-off design—it is a scalable roadmap. James and his team are already planning toward a 1PB SSD, extending the same architectural foundation.

Execution of the first shipping products is led by Shai Tubul, Senior Director of Program Management Engineering.

128TB UltraQLC SSDs: Customer testing begins within weeks
256TB NVMe SSD: U.2 form factor availability planned for early 2026

For hyperscalers deploying hundreds—or thousands—of SSDs per system, performance per watt has become decisive.

“UltraQLC delivers game-changing results,” Tubul said.
“But just as important, it gives us the flexibility to deliver customized solutions faster.”