Bank of America analyst Vivek Arya projects the semiconductor market will reach $3.2 trillion in 2030, up 88 percent from $1.7 trillion this year, arguing that order books, capacity commitments and pricing show no sign of the AI investment slowdown some commentators have flagged.
Memory takes the lead
Memory is forecast to become the largest segment, growing 92 percent from $937 billion to $1.8 trillion and contributing $863 billion of absolute growth, more than half the market's total expansion. Arya notes that high-bandwidth memory, which is packaged with GPUs to cut latency, drives the surge, but ordinary DRAM and NAND prices have risen more sharply. If the broader DRAM market rebalances, SK Hynix stands to benefit more than Micron because a larger share of its revenue already comes from HBM.
Nvidia's full-stack play
On the compute side, Nvidia remains the primary supplier of chips used to train AI models, with its CUDA platform locking in early code. The company has also moved into inference through its Groq acquisition, deploying language processing units for the memory-heavy decode phase and GPUs for the compute-intensive pre-fill phase. Combined with its networking portfolio, this lets Nvidia sell complete end-to-end servers for training, inference, storage and agentic AI. Demand still outpaces capacity, and the stock trades at a forward price-to-earnings ratio of 17 times.
ASML's monopoly scales
Wafer fabrication equipment spending is expected to jump 129 percent from $156 billion to $360 billion by 2030, with ASML as the primary beneficiary. The Dutch firm holds a monopoly on extreme ultraviolet lithography, required for advanced logic chips and HBM, while its older deep ultraviolet machines handle less critical layers. New high-NA EUV systems, priced at double the standard EUV tools, are slated to drive the next leg of growth, with commitments from the three largest foundries. ASML plans to raise EUV capacity 30 percent next year and another 30 percent in 2028.
What to watch
The forecast rests on sustained AI infrastructure build-out. Any crack in hyperscaler capex plans would hit memory, compute and equipment simultaneously. For now, the order flow Arya cites suggests the pipeline remains full.