South Korea's FADU will take a keynote slot at the Future of Memory and Storage (FMS) 2026 conference next week and use it to do something no company has done on a public stage yet: show a working, complete PCIe Gen 6 solid-state drive — not just the controller chip inside one. The company confirmed the appearance today, July 30, and the FMS 2026 official program lists the session for August 4, 1:50 PM PT at the Santa Clara Convention Center under the title "The Future of AI Storage: From Silicon Innovation to Hyperscale Efficiency."

The timing matters. Prior Gen 6 announcements — including FADU's own Sierra FC6161 controller reveal at FMS 2025 — have been silicon-level previews, not finished drives. ScaleFlux's FC6116 controller, announced two days ago, will not begin sampling until Q4 2026. What FADU is promising is hardware an engineer can touch: a physical drive ready to demo at the show's exhibition floor, backed by three consecutive generations of shipping Gen 5 controllers already running in hyperscale data centers.

FADU's Fabless Strategy Lands Inside the Hyperscaler Supply Chain

The physical SSD debut is the most visible moment in what has become a quietly remarkable commercial trajectory. By early June 2026, FADU reported that the total contract value of its agreements with hyperscale cloud operators had exceeded 300 billion Korean won — approximately $209 million USD at today's exchange rate of 1,436 KRW per dollar — a milestone the company attributed to expanding customer diversification across the hyperscale tier. That figure followed a return to profitability in Q1 2026 and a 4.5-fold year-over-year revenue increase in the first half of 2025.

A May 2026 contract filing disclosed a single deal worth 46.5 billion won (approximately $32 million USD) with an unnamed overseas NAND flash manufacturer for enterprise SSD controller supply — a contract widely discussed in Korean semiconductor circles as likely involving SanDisk. At the company's 10th anniversary press conference in August 2025, co-founder and then-CEO Jihyo Lee confirmed publicly that FADU had secured mass-production supply agreements with two of the world's four largest hyperscalers — the group comprising Amazon Web Services, Google, Microsoft, and Meta.

That is the structural achievement the FMS 2026 keynote is meant to demonstrate to a Western audience. A fabless Korean controller company with no proprietary NAND has now penetrated the highest tier of the hyperscale storage supply chain across three consecutive PCIe generations — something ARM-based rivals including Phison and Silicon Motion have pursued and not fully achieved at production scale with the same customer set. FADU's entry into that tier was not achieved through NAND partnerships or platform bundling: it was achieved through controller silicon and the firmware stack built around it.

What Is FADU Showing at FMS — and What Does It Run On

The hardware being unveiled at FMS 2026 is a complete enterprise SSD built on the Sierra FC6161 controller, which FADU refers to internally as "Lhotse." The controller taped out on schedule and was confirmed for chip-out in May 2026. The Gen 6 specifications are specific and independently sourced: sequential read and write speeds of 28.5 GB/s and 20.0 GB/s respectively; random read and write at 6.9 million IOPS and 1 million IOPS; controller power consumption below 7W for reads and 9.5W for writes; maximum drive capacity of 32 TB using 4kB Indirection Units, or up to 256 TB with 16kB IU configurations, according to the FADU blog's April 2026 technical overview.

Those numbers put the Sierra FC6161 marginally ahead of the Silicon Motion SM8466 (which specifies 28 GB/s sequential reads) on raw throughput, while matching ScaleFlux's FC6116's 28 GB/s target at sub-9W, as detailed at FMS 2025. The practical competitive picture is more nuanced than the spec sheet suggests: ScaleFlux's FC6116 will not begin OEM sampling until Q4 2026, placing customer qualification — typically an 18-month process — in mid-2028 at the earliest. FADU's physical SSD is at the show in August 2026.

The booth will also show the company's Gen 5 controller platform, AI storage demonstration servers, power management solutions, and liquid cooling systems. FADU is attending as an Executive Premier exhibitor, the highest sponsor tier at FMS 2026, with a dedicated exhibition floor footprint, as confirmed on the FMS 2026 official program.

RISC-V and a Pipelined Control Plane Are Why FADU Controllers Run Cool

The performance-per-watt figures that have attracted hyperscaler interest trace back to a controller architecture that FADU built from the ground up rather than adapting from an existing design. Where most enterprise SSD controllers — including those from Silicon Motion and Phison — run their Flash Translation Layer and NVMe command processing on ARM CPU cores, FADU's controller ASIC uses four 64-bit RISC-V cores augmented by a programmable control plane built from parallel RISC Processing Units (PPUs), as documented by Blocks & Files.

The distinction matters for power and scalability. In an ARM-based design, scaling throughput typically requires adding more ARM cores — each of which carries licensing costs and proportional power consumption. FADU's PPU pipeline breaks NVMe commands into discrete stages, with each stage assigned to a dedicated PPU core. That pipelining enables multiple commands to be processed concurrently at different pipeline stages simultaneously, increasing effective throughput without requiring a proportional increase in compute power. FADU claims the architecture allows each pipeline stage to be "optimized independently," so performance improvements between PCIe generations — from Gen 5 to Gen 6 — do not require a ground-up redesign.

The Gen 6 Lhotse controller adds meaningful upgrades over its Gen 5 predecessor (the K2) beyond the raw speed increase. The NAND interface moves to ONFI 5.1 at 3,600 MT/s, up from ONFI 5.0 at 2,400 MT/s. DDR5 support increases to 5,200 MT/s. The number of virtual functions doubles from 63 to 128, enabling more efficient SR-IOV partitioning in multi-tenant AI inference deployments. NVMe compliance advances to version 2.2 with NVMe-MI 2.0, and OCP compliance moves to version 2.6. Flexible Data Placement support expands to accommodate up to 16 initially isolated and 16 persistently isolated Reclaim Unit Handles in a single Reclaim Group — a capability directly relevant to AI workloads where write amplification control translates to lower cost-per-token in inference operations, per FADU's technical overview.

Gen 5 Already in Production — the Track Record That Sells Gen 6

The keynote's credibility rests on a Gen 5 production history that distinguishes FADU from companies still in silicon preview. As of April 2026, FADU had shipped millions of enterprise SSDs globally, with Gen 5 controller unit shipments exceeding one million in 2025 alone, according to FADU's technical blog. The company projects Gen 5 controller shipments of three million in 2026 and over five million in 2027.

Known OEM customers include Western Digital, whose SN861 enterprise SSD runs a FADU controller and holds NVIDIA GB200 NVL72 certification for AI compute environments, and SK hynix, as reported by Blocks & Files. White-label SSD supply reaches Swissbit and SMART Modular, the latter of which supplies Lenovo. Meta's hardware engineering team appeared alongside Jihyo Lee during the FMS 2025 keynote when the Sierra FC6161 silicon was first announced, cementing the company's identity as an early Gen 6 adopter even before volume commitments were publicly confirmed.

The FMS 2026 Stage and Who Else Will Be There

The co-presenters at FADU's August 4 keynote are Ehyun Nam, FADU's co-founder and CEO, and Ross Stenfort, Distinguished Engineer for SSD Architecture at SanDisk and OCP Storage Project Lead, as listed on the FMS 2026 official program. Stenfort's appearance connects FADU's Gen 6 SSD to the OCP (Open Compute Project) infrastructure standards that govern hyperscaler storage procurement — a pairing that signals FADU's SSDs are being evaluated within, not merely alongside, the frameworks hyperscalers actually use for qualification.

FMS 2026 is the conference's 20th anniversary edition, now owned by Terrapinn following its acquisition from Conference Concepts Inc. earlier this year. Organizers expect more than 3,500 attendees from over 1,500 organizations. The 2026 program spans AI and machine learning applications, CXL memory pooling, NVMe standards updates, high-bandwidth memory, and computational storage — a scope that reflects how thoroughly AI infrastructure has reshaped the scope of what was originally a NAND flash conference.

For FADU, the keynote at the Mission City Ballroom represents more than a product unveiling. It is the company's most direct pitch to the Western storage industry — an opportunity to show engineers and procurement leaders that its contracts with hyperscalers are translating into shipping hardware. As co-founder and President Jihyo Lee put it in the announcement: FMS is "an important stage for making FADU's technology known within the big tech and hyperscaler ecosystem and simultaneously seeking to strengthen partnerships with global partners and acquire new customers."

What Engineers Should Check Before and After the Show

Several questions that the keynote announcement does not address will matter to engineers evaluating FADU for Gen 6 qualification.

The first is the qualification pipeline's current state. FADU's Gen 5 production traction confirms the controller ASIC ships reliably, but Gen 6 involves a new silicon tape-out and a different signaling architecture (PCIe Gen 6.2 x4). A 12- to 18-month OEM qualification timeline from drive availability to production deployment is standard, meaning that engineering teams who engage at FMS 2026 are targeting 2027 or 2028 production timelines at the earliest.

The second is the FDP (Flexible Data Placement) implementation depth. FADU's expanded FDP support — 16 initially isolated plus 16 persistently isolated RUHs — is architecturally significant for AI inference workloads where separating hot and cold write streams reduces write amplification. But whether that implementation is validated against NVIDIA's CMX (Context Memory eXtension) architecture and emerging inference server configurations will be the question procurement engineers raise in the booth.

The third is the power ceiling under real inference workloads. FADU's sub-9.5W write figure is the controller's rated typical power, consistent with industry practice. Sustained AI inference workloads — particularly those mixing large sequential writes (KV cache persistence) with high-concurrency random reads (model serving and vector database access) — stress controllers differently than published throughput benchmarks. Independent verification from qualification testing at a hyperscaler, expected to emerge in 2027, is the data point that will determine whether the power budget holds under production conditions.

Exchange rate as of July 30, 2026; conversions are approximate.

Frequently Asked Questions

What makes FADU's PCIe Gen 6 SSD different from other Gen 6 announcements?

Most Gen 6 announcements to date — from Silicon Motion, ScaleFlux, and others — have been controller-silicon reveals, not complete drives. FADU's FMS 2026 debut will show a finished, physical PCIe Gen 6 SSD built on the Sierra FC6161 ("Lhotse") controller, which taped out and reached chip-out in May 2026. The distinction matters for engineers beginning qualification timelines: a finished drive on a show floor is closer to productizable hardware than a controller chip with a sampling roadmap. FADU also has an established production track record across three consecutive PCIe generations with confirmed hyperscaler supply agreements, which differentiates it from companies showing Gen 6 silicon for the first time.

Does FADU's RISC-V controller architecture actually outperform ARM-based designs?

RISC-V and ARM are instruction set architectures, and raw ISA performance comparisons without workload context are not meaningful. FADU's architectural claim is more specific: its PPU (RISC Processing Unit) pipeline allows NVMe command stages to be processed concurrently, increasing throughput without scaling the number of processor cores proportionally. In practice, the Gen 5 ECHO controller runs under 7W at rated performance — a figure that compares favorably to enterprise Gen 5 SSDs from Samsung's PM1743, which can exceed 20W at peak load. Whether the Gen 6 Lhotse maintains that efficiency gap against Micron's 9650 (February 2026 mass production) and Samsung's PM1763 (July 8, 2026 mass production) will depend on verification under real AI workloads, not datasheet comparisons.

Who are FADU's confirmed hyperscaler customers?

FADU has not publicly named its hyperscaler customers, citing confidentiality agreements. At its August 2025 10th anniversary press conference, then-CEO Jihyo Lee confirmed that two of the four major global hyperscalers — Amazon Web Services, Google, Microsoft, and Meta — had signed mass-production supply agreements. Meta's hardware engineering team appeared alongside Lee at the FMS 2025 keynote when FADU introduced the Sierra FC6161 silicon, making Meta the most publicly visible association. The remaining two hyperscalers and the specific products being supplied have not been disclosed.

Can a South Korean fabless company actually compete with Samsung and Micron in enterprise SSDs?

FADU competes in the SSD controller market, not in NAND flash manufacturing — a distinction that changes the competitive equation significantly. Samsung and Micron are vertically integrated: they design controllers, manufacture NAND, and build finished SSDs. FADU supplies only the controller and firmware stack, meaning it depends on NAND manufacturers (SanDisk, SK hynix, and others) to build completed drives on top of its silicon. That fabless model has allowed FADU to ship production controllers in enterprise SSDs already deployed by Western Digital and SK hynix without competing with its own customers for NAND supply. The question the Gen 6 physical SSD debut answers is whether the company's own finished drives — built on the same silicon it licenses to OEMs — can compete on a single platform at hyperscale.