Samsung has started distributing samples of its LPDDR6X memory to the Qualcomm. Essa technology still under development promises capacities exceeding 1 TB in chips dedicated to artificial intelligence. The movement occurs even before the mass production of LPDDR6, scheduled for the second half of 2026.
The partnership focuses on Qualcomm’s AI250 accelerator, successor to AI200, aimed at inference in data centers. The choice of LPDDR instead of HBM memories reflects a strategy of balancing performance, cost and energy efficiency.
These samples allow for early validations on real hardware. The initiative demonstrates the industry’s acceleration to meet the growing demand for AI processing at scale.
LPDDR6X Technical Advancements
The LPDDR6X represents a direct evolution of the LPDDR6, with specifications still being finalized by the JEDEC standard. The base version of LPDDR6 offers initial speeds of up to 10.7Gbps and efficiency gains of 21% over LPDDR5.
Optimized versions achieve 14.4 Gbps per pin. Esses numbers guarantee greater bandwidth and lower consumption on mobile devices and servers.
The LPDDR6X should raise these parameters even further. The technology maintains compatibility with intensive AI workloads, especially in inference.
Qualcomm strategy in the AI market
The Qualcomm adopts LPDDR as an alternative to the HBM memories used by NVIDIA and AMD. HBM solutions deliver higher speeds, but present high costs and greater energy consumption.
The LPDDR approach reduces manufacturing complexity and facilitates scalability. The AI200 chip already supports up to 768 GB of memory.
The AI250 successor should exceed 1 TB with the integration of LPDDR6X. Essa capacity serves inference applications in data centers with a focus on efficiency and cost-benefit.
Competition in the memory sector
Other manufacturers are advancing in the same segment. SK Hynix plans to present details of its 16 Gb LPDDR6 with 14.4
These developments reinforce the competition for leadership in providing memories for AI. Samsung seeks to consolidate its position by anticipating samples for strategic partners.
The company has completed the development of the LPDDR6 standard. Mass production begins in the second half of 2026, setting the stage for widespread adoption.
Advantages of LPDDR in AI applications
LPDDR memory offers lower random access latency compared to traditional server options. Esse feature benefits inference tasks that require fast responses.
- Reduced production costs compared to HBM
- Lower energy consumption in continuous operations
- Ease of integration into existing platforms
- Scalability to high capacities without proportional increase in complexity
These factors make the technology attractive for midsize data centers. Qualcomm positions its accelerators as competitive options in specific scenarios.
Development chronology
The LPDDR6 standard received official certification from JEDEC in mid-2025. Fabricantes began validations immediately after publishing the specifications.
Samsung highlighted its version during the CES 2026 preview, receiving recognition for innovation. The transition to mass production occurs gradually.
Early distribution of LPDDR6X samples indicates planning for commercial launch between late 2027 and early 2028. Empresas adjust roadmaps to incorporate the technology into future products.
Impact on mobile devices
While the initial focus will be on AI accelerators, the LPDDR6X will benefit premium smartphones. Greater energy efficiency extends battery life in intense tasks.
Devices with advanced processors gain the ability to run AI models locally. Essa functionality reduces dependence on cloud connection.
Manufacturers are preparing platforms to take advantage of the performance leap. The combination of speed and low consumption defines a new generation of mobile devices.
Differences between LPDDR and HBM
HBM memories stack chips vertically to maximize bandwidth. Essa architecture supports large model training, but significantly increases costs.
LPDDR uses traditional configuration with a focus on efficiency. The approach simplifies integration and reduces thermal consumption.
- HBM: Ideal for Large-Scale AI Training
- LPDDR: optimized for inference and embedded applications
- Cost: HBM significantly more expensive
- Consumption: LPDDR has an advantage in low power scenarios
The choice depends on the type of workload. Qualcomm prioritizes efficient inference in data centers.
Industry preparation
The global DRAM shortage has influenced strategic decisions. Fabricantes seek viable alternatives to maintain stable supply.
Samsung invests in advanced production lines. The company plans to expand capacity to meet future demand.
Partnerships like the one with Qualcomm speed up validations. The process ensures compatibility before widespread commercialization.
Outlook for servers
The integration of LPDDR6X into accelerators expands options in the server market. Empresas seek solutions balanced between performance and operational cost.
Qualcomm’s AI250 competes in specific inference niches. Capacity greater than 1 TB allows processing of complex models locally.
Midsize data centers are gradually adopting these technologies. The transition prepares infrastructure for the next wave of AI applications.
Samsung has begun distributing LPDDR6X memory samples for the Qualcomm prior to the commercial launch of LPDDR6. Essa technology under development targets artificial intelligence chips with capacities exceeding 1 TB. The partnership accelerates validations for the AI250 accelerator, scheduled for 2027.
