- Solutions
ENTERPRISE SOLUTIONS
Infuse new product development with real-time intelligenceEnable the continuous optimization of direct materials sourcingOptimize quote responses to increase margins.DIGITAL CUSTOMER ENGAGEMENT
Drive your procurement strategy with predictive commodity forecasts.Gain visibility into design and sourcing activity on a global scale.Reach a worldwide network of electronics industry professionals.SOLUTIONS FOR
Smarter decisions start with a better BOMRethink your approach to strategic sourcingExecute powerful strategies faster than ever - Industries
Compare your last six months of component costs to market and contracted pricing.
- Platform
- Why Supplyframe
- Resources
Booming sales of AI servers are fueling a recovery in the memory market, with signs appearing of increased prices and the potential of reduced availability for high-end DRAM parts.
SK Hynix became the latest semiconductor supplier to signal the start of recovery for the memory industry, driven by the demand for high-performance memory required for large-scale AI processing in data centers. The South Korean memory firm reported that its second-quarter revenue plunged 47% year-on-year. This announcement follows news that fellow memory supplier Micron suffered a 57% revenue drop during its previous quarter and identified AI as the driver of a nascent market rebound.
The Growing Need For Advanced DRAM in AI Applications
Significantly, SK Hynix stated that its DRAM products’ blended average selling price (ASP) increased during the second quarter, a remarkable turnaround following a lengthy period of declining prices. However, the increases were entirely due to rising ASPs for high-end memories, including HBM3 and DDR5. The price boost for these devices offset declines in older memory types, such as DDR4.
AI training requires vast amounts of data to be funneled to processors at the fastest possible rates. Advanced AI models have an enormous number of parameters, with the GPT-3 model that powers ChatGPT incorporating 175 billion parameters. This total is up from 1.5 billion for the previous version of the model: GPT-2. GPT-4 is rumored to have more than 1 trillion parameters.
Meanwhile, AI servers are increasing their average number of processors, multiplying the memory demand of each system.
AI server vendors are building systems using huge quantities of cutting-edge DRAMs to meet the growing memory speed and bandwidth challenges. AI servers integrate as much as eight times the amount of DRAM as regular servers, according to Micron. These factors are expected to contribute to a rise in DRAM sourcing activity.
An Inflection Point for The DRAM Market
The Commodity IQ Demand Index for DRAM has deteriorated since August 2021. However, the index is projected to take a turn to the positive in September and will remain there for the rest of the year. Despite this, the index is still expected to remain well below the baseline level, indicating that sales will continue to slide.
The shift toward increase after over two years of decline represents a critical inflection point for the DRAM market. DRAM buyers, particularly those sourcing memory for servers, should be prepared for increased prices and diminished availability for advanced memory types like HBM3 and DDR5. These buyers should consider negotiating with memory makers to secure long-term supply deals for advanced memories.
For more insights like this, delivered to your inbox every week, subscribe to the IQ Insider newsletter today!