A small group of tech giants, led by Nvidia, Google, and AMD, are set to control the majority of the high-bandwidth memory market by 2027, driven by surging AI demands and manufacturing constraints.
Analyst estimates suggest that the market for high-bandwidth memory, or HBM, is becoming increasingly concentrated around a small group of buyers, with NVIDIA, Google and AMD expected to account for 85.4% of global demand by 2027, according to Morgan Stanley figures reported by The Inference. NVIDIA alone is projected to take 37.3%, Google 36% and AMD 12.1%, leaving the rest of the market with only a narrow share of supply. That level of concentration is striking at a time when several other companies are still trying to build rival accelerators for artificial intelligence workloads.
HBM sits at the centre of the AI hardware stack because it provides the high memory bandwidth needed by accelerators such as GPUs and TPUs. The projected demand split reflects how deeply the market is being shaped by a handful of large-scale AI programmes, particularly in datacentres. Reuters has repeatedly noted that the AI build-out is not only lifting demand for compute chips but also tightening the supply chain for the specialised memory those chips require.
NVIDIA’s role in the market looks especially significant. The company said in its latest disclosed commitments for the quarter ended 26 July that total supply obligations reached $279 billion, up sharply from $119 billion in the previous quarter. According to the company’s filing, much of that reflects long-term arrangements tied to memory procurement and the production capacity needed for datacentre systems. The scale of the figure does not mean NVIDIA has already spent that amount on HBM alone; rather, it represents a broader pipeline of obligations spread across several years, with roughly $267 billion of those commitments due to mature by the end of fiscal 2029.
The underlying constraint is manufacturing capacity. HBM is materially more demanding to produce than standard DDR5 memory because it consumes several times more wafer area per gigabyte. Micron has said that making HBM3E requires about three times the wafer capacity needed to produce the equivalent amount of bits in DDR5 on the same process node. That does not mean every unit of HBM directly displaces three units of DDR5, but it does mean the industry must devote far more scarce production capacity to HBM whenever demand rises.
That squeeze is already feeding through into the broader memory market. DDR5 prices have risen by almost 500% over the past 12 months, according to the figures cited by Hardware Upgrade, and Micron has warned that a reversal may take several years. The combination of aggressive AI deployment, long-term supply commitments and the heavy manufacturing burden of HBM suggests that memory availability will remain a key bottleneck for the sector well beyond the current cycle.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
