You're offline - Playing from downloaded podcasts
Back to All Episodes
Podcast Episode

NVIDIA Asks Samsung to Skip Quality Tests as Memory Becomes AI's Critical Bottleneck

February 5, 2026

Audio archived. Episodes older than 60 days are removed to save server storage. Story details remain below.

NVIDIA has made an unprecedented request for Samsung to ship HBM4 memory chips before completing final quality testing, highlighting how memory shortages have become the critical constraint for the global AI buildout. Korean memory giants Samsung and SK Hynix now control over eighty percent of the HBM4 market, giving them enormous leverage in the AI supply chain.

NVIDIA's Unprecedented Request

In a dramatic reversal of typical semiconductor industry dynamics, NVIDIA has asked Samsung Electronics to begin shipping its next-generation HBM4 memory chips even before completing final quality verification testing. The request underscores the severity of memory constraints facing the artificial intelligence industry, where high-bandwidth memory has become the critical bottleneck for AI chip production.

Korean Memory Giants Rise to Power

The development marks a stark shift in bargaining power. Just one generation ago, Samsung was anxiously awaiting approval from NVIDIA's stringent quality tests for its previous HBM chips. Now, Korean memory manufacturers have ascended to what analysts describe as "super supplier" status, effectively controlling the chokepoint of the global AI industry.

According to market research firm Counterpoint Research, SK Hynix and Samsung are projected to hold fifty-four percent and twenty-eight percent of the global HBM4 market respectively, giving them a combined share exceeding eighty percent. Analysts have compared this position to Taiwan Semiconductor Manufacturing Company's dominance in advanced chip fabrication.

Supply Constraints Drive Urgency

The request comes as Samsung prepares to begin mass production of sixth-generation HBM4 in February, with initial shipments destined for NVIDIA's upcoming Vera Rubin AI accelerator platform. Morgan Stanley has stated that both Korean memory companies have entered "an unprecedented stage of supply constraints," with the entire year's memory production already sold out.

Both companies have begun what the industry calls "risk production" of HBM4, putting wafers into production lines before customer certification is completed. This approach allows them to build volume ahead of formal approval, though it carries risks around reliability and yield consistency.

Market Implications

The memory shortage reflects a broader industry transformation as manufacturers redirect capacity toward high-bandwidth memory for AI servers. HBM commands higher margins but consumes roughly three times the wafer capacity of standard memory per gigabyte. Analysts project memory prices to rise forty to fifty percent through the first half of the year, with relief not expected until new manufacturing facilities reach volume production in twenty twenty-seven.

Published February 5, 2026 at 1:14am

More Recent Episodes