2026 WAIC Siteng Heli & GenstorAIge Joint Booth

 

7Month17Day,2026World Artificial Intelligence Conference(WAIC)was held in Shanghai.GenstorAIge AI90 —— FaaS (Flash as a Service) for AI and its AI SSD series productsmade a major appearance,joining hands with Siteng Helito showcase AI90a live "compute-storage convergence" solution.

GenstorAIge CEO Sun Dan exchanges with Siteng Heli partners

 

On the first day of the conference,GenstorAIge'slocated atZhangjiang Science Hall'smultiple boothsattracted a large number of professional visitors. The AI90 inference workstationrunning at the booth demonstrated to the industry, through stable operation in a real environment, the technical capability of "consumer-grade GPU also being able toefficiently handle large model inference".

2026 WAIC GenstorAIge Zhangjiang Hall Exhibition Area

 

Over the past two decades, GPU computing powerhas increased by about60,000times, while video memory capacityhas only grown by16times——. The "memory wall" leaves expensive computing chips in a common "starving" waiting state, with mainstream GPU effective computing power utilizationat only30% - 60%。

 

. Meanwhile, as large model inference capabilities and Agent intelligent agents accelerate their evolution,KV Cache the rapid expansion ofmakes this contradictioneven worse. Large model inference is gradually becoming an"exclusive privilege"of data centers and dedicated computing cards, with a high barrier to entry.

 

GenstorAIge's AI90 breakthrough approach hits the core issue:offloading KV Cache from expensive HBM to high-performance SSD, building a layered architecture of "GPUcomputing power focuses on computation,SSD handles memory", achieving a leap in GPU computing power utilization,unleashing token efficiency.。

GenstorAIge AI90 On-site Benchmark Data

On-sitereal workstationdemonstration datawas highly impactful:

 

• In Llama 3 70B model inference,a4card5090cluster paired with AI90, throughput jumped from approximatelyabout 120 tk/s in the traditional architecture to about 610 tk/s, a performanceincrease of5.1times.

 

• 64K First Token token latency dropped from27.99secondssharply to0.564seconds, a performance improvement of nearly50times.。

 

• Video memory utilization increased from30% - 40% to 85% - 95%, an improvement of2-3times.

 

• "The longer · the better" ——The longer the input sequence, the more pronounced the acceleration effect ofAI90 . Perfectly suited for complex scenarios such aslong document analysis,、code understanding,andmulti-turn conversation history.complex scenarios.

 

GenstorAIge Joins Hands with Siteng Heli: The Ecosystem Resonance of GenstorAIge'sAI90&AI SSD"Compute-Storage Convergence"

At this WAIC Zhangjiang exhibition area, another important move by GenstorAIge wasits deep booth collaboration with Siteng Heli.。

GenstorAIge'sAI90"compute-storage convergence" acceleration solution and AI SSD(PT200z) storage foundation——one solves"fast computation", the other solves"smart storage" ; deeply integrated with Siteng Heli's self-developed GPU cluster scheduling system, the offloading, scheduling, and read/write paths of KV Cache are optimized, breaking the video memory wall and truly maximizing the computing power utilization of theGPU cluster.

In the era where AI inference moves from single cards to clusters, the synergy efficiency between computing and storage directly determines the upper limit of Token throughput. Only deep coupling between the computing and storage ecosystem layers can move"compute-storage convergence"from concept to engineering practice.

 

GenstorAIgeandSiteng Heli'scollaboration validates the deep adaptability ofGenstorAIge's AI90 innovative architecture within edge-side computing infrastructure.

 

Conclusion

Large model inference is shifting from"computing power-centric"to"efficiency-centric". Data and storage are the core propositions for the next stage of AI infrastructure.

 

GenstorAIge AI90 has validated the strategic value of this direction.AI 90Through an innovative FaaS architecture, consumer-grade GPU can also efficiently handle large model inference, truly unleashing token efficiency.

 

From the real benchmarks at theZhangjiang boothto the deep ecosystem resonance withSiteng Heli,we witnessed a Chinese storage company's determination to break the "computing power privilege" through technological innovation. — WAIC 2026witnessed a Chinese storage company's determination to break the "computing power privilege" through technological innovation.

 

GenstorAIge is leveraging the core capabilities of PT200Z AI SSD , using an innovativeAI90 FaaS for AIarchitectureand joining forces with numerous industry ecosystem partners,to build a more inclusive data foundation for the intelligent world.

GenstorAIgeCEOCEO Sun Dan poses for a group photo with Siteng Heli partners

 

| About GenstorAIge:

GenstorAIge (GenstorAIge) is dedicated to building the data foundation for the intelligent world. It is the only full-stack vendor in China providing enterprise-gradeSSD、PCIe Retimer/Redriverconnectivity chips andAItraining-inference integrated solutions covering“storage,-connectivity,-and management”. Founded in2023, its core team has over20years of deep experience in the storage and server industry and has quickly entered the supply chains of several top-tier internet andOEMmanufacturers.2026InWAIC, GenstorAIge debuted itsAI90——FaaS for AIarchitecture, using an innovative“compute-storage convergence”solution to enable consumer-gradeGPUto unleash the potential of large models, drivingAIcomputing power towards inclusivity.

 

| About Siteng Heli:

Siteng Heli (Tianjin) Technology Co., Ltd., founded in2009, is one of the earliest domestic manufacturers focused onAIcomputing power infrastructure and has ranked on theIDCChina Accelerated Server ListTOP 8. The company is anNVIDIAElite Partner and a Huawei AscendAPNDiamond Partner, with over20NVIDIAcertified engineers, and has delivered thousand-card-scaleAIcomputing power clusters.