Synopsys chief executive Sassine Ghazi has delivered a blunt update to an industry already braced for disruption: the current squeeze on memory chips is unlikely to ease before 2026 and may persist into 2027. Ghazi says much of the high-end DRAM and high-bandwidth memory (HBM) being produced by the world’s leading suppliers is being absorbed by AI data centres and infrastructure projects, leaving little spare capacity for smartphones, laptops and other consumer and industrial markets.
That observation matters because Synopsys sits at the centre of the chip ecosystem. The California-based electronic design automation (EDA) firm supplies the software and IP that chipmakers and system designers use to build silicon. Its CEO’s read on allocation and capacity is therefore informed by conversations with both foundries and the large chip designers racing to supply AI accelerators and servers.
The mechanics of the squeeze are straightforward. Training and inference at scale consume vast amounts of fast, wide memory, especially HBM. Cloud providers and AI infrastructure builders have committed tens of billions of dollars to expand data-centre capacity, and much of the incremental demand for premium memory is going to those projects. Supply is concentrated: Samsung, SK Hynix and Micron still dominate high-end DRAM and HBM production, and expanding a fabrication line takes time and capital.
Ghazi notes that from investment decisions to actual production usually requires about two years — a delay that helps explain why shortages can become prolonged. Memory markets have long been cyclical, swinging between shortages and gluts, but several executives and analysts now describe the current phase as a “supercycle” driven by secular AI demand rather than the shorter-term consumer cycles of the past.
The knock-on effects are visible down the supply chain. Rising memory prices push component costs for consumer electronics higher, creating inflationary pressure that manufacturers may pass on to buyers. Other sectors — notably automotive and industrial applications that increasingly rely on bespoke memory for safety-critical functions — risk being deprived of supply as priorities tilt towards AI infrastructure.
Geopolitics and industrial economics amplify the problem. High-end memory production is capital-intensive and geographically concentrated in East Asia; export-control regimes, trade frictions and government subsidies all influence where new capacity is planned and how quickly it comes online. For companies and governments seeking resilience, the options are costly: build more fabs, diversify suppliers, or redesign systems to rely less on scarce memory types.
For investors and corporate strategists the current backdrop is a mixed opportunity. Memory majors are in a strong pricing position and have called this period a “golden era.” EDA firms and IP suppliers like Synopsys stand to benefit from sustained chip design activity. But original equipment manufacturers face margin pressure and possible product delays, while chip buyers may accelerate efforts to secure supply through long-term contracts or strategic investments.
The upshot is that the memory market is poised to remain a choke point for technology deployment. Unless producers radically accelerate capacity expansion or demand moderates, bottlenecks and elevated prices for high-bandwidth memory will likely be features of the market for the next 18–24 months, shaping the pace and cost of AI roll-outs and affecting a wide range of consumer and industrial products.