Reporting by NUPIAO |
Edited by | Wen Shuqi
If you’ve been tracking the storage game lately, you’ll notice Western Digital (WD, Nasdaq: WDC) is stepping up its presence in China.
Late last May, WD rolled out its very first Innovation Day tailored specifically for Chinese customers. They pulled in heavy hitters from cloud providers, autonomous driving, and a handful of other hot sectors. Honestly, looking at that attendee list basically maps out the most aggressively funded AI tracks in China right now.
Speaking of hot, WD has easily been one of the standout performers on the global stock market when it comes to AI infrastructure plays. We’re talking about a stock that’s jumped from under $50 a year ago to roughly $526 today—over ten times its previous valuation. During their Q2 fiscal 2026 earnings call, CEO Irving Tan didn’t mince words: the company’s annual manufacturing capacity is practically fully booked, with some long-term supply contracts already stretching out to 2027 and 2028.
So why the rush? Here’s the straight talk: AI data centers absolutely need storage, and when you’re talking about massive-scale deployment, nothing beats the economics of spinning magnetic disks. That’s exactly where WD’s bread and butter lies right now.
In February 2025, WD officially split from its flash memory arm, SanDisk. Post-split, WD goes all-in on HDDs, while SanDisk focuses squarely on the SSD space. The reality on the ground? The global mechanical drive market has basically shrunk to three major players: WD, Seagate, and Toshiba. Supply is incredibly tight, which means pricing power has definitely shifted to the manufacturers.
Yet, even with inventory practically flying off the shelves, WD is still actively chasing new orders from China.
We sat down with Stefan Mandl, VP of Global Sales and Marketing at WD, who let slip that he’s been making constant trips to visit Chinese partners. The reason? China’s momentum in this sector is moving at a pace you just don’t see anywhere else right now.
The numbers back him up. Forecasts from Fortune Business Insights peg China’s AI infrastructure market to grow at a staggering 32.5% CAGR between 2025 and 2032, comfortably outpacing IDC’s global average of 31%. Big names like Tencent, ByteDance, and Alibaba are consistently bumping up their capex year after year to race ahead in the AI data center arms race.
All that spending inevitably translates into massive procurement requests for storage gear. Yeah, we’ve all heard the “SSDs will eventually kill HDDs” rumor mill chatter for years. But Ahmed Shihab, WD’s Chief Product Officer, pointed out that early adopters aren’t seeing that shift happen fast. The real dealbreaker? Mechanical drives wear out incredibly slowly and come with a serious price advantage. Right now, the cost per terabyte on flash is still hovering around twenty times higher than a standard HDD.
Look, everyone in the global tech scene finally agrees on one thing: storage is the unsung hero of AI infrastructure.
As Ahmed breaks it down, AI is fundamentally a data system. You’ve got ingestion, preprocessing, model training, and inference running in cycles. Every single training run or inference pass spits out fresh data. Compute cycles can be recycled endlessly, but your dataset keeps ballooning without stopping. That relentless data bloat is exactly what’s sending storage demand through the roof.
IDC projects global AI infrastructure spend will smash past the $1 trillion mark by 2029, riding a healthy 31% five-year CAGR. And guess what? Nearly 80% of that cloud storage workload will still be leaning heavily on flash drives.
When we polled WD’s core global clients and distributors about what actually matters to the Chinese market, two things jumped out. First, keeping a tight leash on TCO while juggling massive AI training and inference workloads tops the list for over 80% of respondents. Second, rock-bottom downtime isn’t an option; reliability and uptime demands are sky-high. Think about it: if storage bottlenecks trip up a 10,000-GPU cluster, the whole rig sits idle, and the financial bleed is absolutely brutal.
There’s no sugarcoating it: the spike in storage demand has become a make-or-break variable in building out AI infrastructure.
Take modern multimodal foundation models as a baseline. A single training cycle now ingests hundreds of petabytes of images, audio, and video. As parameter counts continue to climb, we’re rapidly inching toward exabyte territory. Both Alibaba Cloud and Tencent have openly admitted their standalone clusters have already crossed the hundred-EB threshold. Crunch the math—roughly 100,000 drives per EB—and the sheer storage overhead becomes painfully obvious. That’s precisely why TCO remains the absolute North Star for Chinese buyers when they’re shopping for storage.
Knowing exactly what their Chinese partners are sweating about, WD recently unveiled a few targeted engineering plays. Leveraging a dual-track strategy that pushes both energy-assisted perpendicular magnetic recording (ePMR) and heat-assisted magnetic recording (HAMR) simultaneously, WD plans to ship drives exceeding 100TB per platter by 2029. They’ve already locked in 4TB per platter commercially, and lab tests have successfully smashed through the technical ceiling to prove 10TB-per-platter is viable.
Why does cramming more capacity onto a single disk matter so much? Simple physics. Higher density means your racks hold exponentially more data while requiring drastically fewer physical units. Less hardware translates directly to slashed electricity bills, reduced rack space, and way lighter maintenance overhead.
On top of that, WD is rolling out power-efficient HDD variants designed to trim energy consumption by 20% with only a marginal 5% to 10% hit to raw performance. Those drives are slated to enter customer validation cycles around 2027, another clear nod to shaving operational costs off the bottom line.
Beyond raw specs, supply chain resilience is a massive conversation starter with Chinese buyers. Once mechanical drives get classified as strategic AI assets, paranoia about facing the exact same export curtailment risks as advanced GPUs is totally understandable. Stefan addressed that head-on, noting they’re actively setting up long-term demand forecasting sharing protocols with local teams. “Securing the supply chain requires two to four years of visibility,” he explained. “That’s why we’re locking into weekly—even daily—syncs with our partners. It’s not just about aligning on the tech roadmap; it’s about synchronizing our supply forecasts so we can match market pull with pinpoint accuracy.”
