While chip stocks have just experienced a significant 'earthquake', Bank of America Merrill Lynch has chosen this moment to release a report that is extremely bullish on the semiconductor sector.
On June 23rd, the very same day global semiconductor stocks suffered a heavy blow—with the Philadelphia Semiconductor Index plunging 7.9% in a single session, Micron Technology (MU) dropping 13%, and South Korea's KOSPI index triggering a circuit breaker—analysts including Vivek Arya at Bank of America Merrill Lynch published a report titled "State of the US Semiconductor Industry: Raising Estimates, AI Extends Visibility to 2028."
The report concludes that the visibility of AI-driven demand for chips now extends to 2028, with memory supply not expected to exceed demand before that time, while semiconductor equipment spending is poised for a major surge in 2028.
The Semiconductor Industry's Path to Another $1 Trillion in Revenue
The chip industry took roughly 50 years to achieve its first $1 trillion in annual sales. However, the bank's framework suggests AI is expected to help the industry add another $1 trillion in the next five years.
In numerical terms, the forecast for the total size of the global semiconductor market has been raised from a previous $2.3 trillion to $2.7 trillion (by 2030), implying a compound annual growth rate of 28% from 2025 to 2030.
2026 is projected to be the breakout year: total semiconductor sales are expected to grow 103% year-over-year, with memory chip sales growing at an even higher rate of 298%—DRAM is forecast to grow 309%, and NAND 295%.
Key Themes Driving This Growth
The AI data center systems market: Expected to reach approximately $1.7 trillion by 2030, starting from about $273 billion in 2025.
The strength and longevity of memory chips: Long-term supply agreements (LTAs) provide 2-3 years of visibility on supply, demand, and pricing, with Micron Technology's (MU) partnership with Anthropic being a recent example.
Semiconductor equipment/advanced packaging/EDA: Benefiting from extended supply agreements and increasing chip complexity.
Analog chips: Benefiting from rising power demand driven by AI.
Server CPUs: Agentic AI presents an opportunity in the server market worth approximately $170 billion.
Memory Chip Outlook: No Oversupply Before 2028
Analysts judge that the supply-demand sufficiency ratio for DRAM and NAND will remain above 110% throughout the forecast period, avoiding the severe oversupply scenarios seen historically. Regarding pricing, DRAM and NAND spot/contract prices are expected to remain robust throughout 2026, with no quarter-on-quarter declines expected before 2027.
Why is supply elasticity so low? The reasons are multifaceted:
Capital expenditure is primarily focused on building fabrication facilities, not buying equipment. Micron Technology (MU) guides its fiscal 2026 capital expenditure to exceed $25 billion (compared to $13.8 billion in fiscal 2025), but most of the increase is for constructing cleanroom space. The actual equipment investment needed to produce chips won't be deployed on a large scale until 2028.
Building a cleanroom and expanding production capacity are two separate things. The former is costly and time-consuming; the latter is what truly affects supply. This means that the significant capital expenditure in 2026-2027 is essentially paving the way for capacity expansion in 2028.
Micron Technology's (MU) new Idaho plant is expected to begin initial output in mid-2027, with volume ramp-up in 2028; its Singapore HBM advanced packaging plant is expected to start contributing in 2027 and be fully operational in 2028.
Geopolitical factors, packaging capacity, and power constraints are also limiting the speed of supply expansion.
HBM (High Bandwidth Memory) is the most closely watched segment within memory chips. The HBM market size is projected to grow from approximately $35 billion in 2025 to about $246 billion by 2030, a compound annual growth rate of 34%. The HBM capacity per AI accelerator is expected to grow from about 187GB in 2025 to approximately 464GB by 2030, an 18% growth rate.
NVIDIA's (NVDA) latest Vera Rubin system requires 288GB of HBM4 memory per accelerator.
Semiconductor Equipment: Spending to Surpass $250 Billion in 2028
The forecast for wafer fabrication equipment/semiconductor equipment (WFE) spending is another major upward revision in this report.
The 2028 WFE spending forecast has been significantly raised by 23%, from a previous $203 billion to $250 billion (representing 32% year-over-year growth). The 2027 forecast has also been raised from $183 billion to $190 billion (up 31% year-over-year).
Overall, WFE is expected to achieve a compound annual growth rate of 20% from 2025 to 2030.
Why 2028? Three key drivers:
First, cleanroom capacity is set for large-scale release around 2028. Current heavy capital expenditure is for building facilities; equipment purchases will be concentrated after the facilities are completed.
Second, technology node transitions are increasing equipment intensity. The mass production of 2nm Gate-All-Around (GAA) processes is accelerating from 2026 to 2028. The initial low yields and low equipment utilization rates of new processes actually require higher equipment investment per wafer. The introduction of High-NA EUV lithography machines is also a significant driver.
Third, technological upgrades in memory chips. The transition from HBM3 to HBM4/5 involves more layers and more complex packaging; the migration of NAND from 300 layers to 400 layers also requires more equipment investment.
Calculations show that if historical patterns for equipment investment per wafer are followed, the implied WFE for 2027 would be about $193 billion and about $245 billion for 2028, closely aligning with the $190 billion/$250 billion forecasts.
It is noteworthy that one metric could be misleading: the ratio of WFE to semiconductor sales (WFE intensity) is expected to decline from a historical norm in the mid-teens percent to about 11%. However, this is primarily because the denominator (sales) is expanding rapidly due to soaring memory chip prices, not because equipment demand is shrinking. A more meaningful metric is equipment investment per wafer, which is continuously rising.
HBM: From $35 Billion to $246 Billion
HBM is the core driver of the current memory supercycle.
Forecasts indicate the HBM market size will grow from approximately $35 billion in 2025 to about $246 billion by 2030, a compound annual growth rate of 34%. The HBM capacity per AI accelerator is projected to increase from 187GB in 2025 to 464GB by 2030, a compound annual growth rate of 18%.
NVIDIA's (NVDA) latest Vera Rubin system (to be launched in the second half of 2026) features 288GB of HBM4 per accelerator—a figure that itself indicates the direction.
HBM price forecast: Approximately $17.5/GB for 2027-2028, higher than the $14.3/GB in 2026.
Based on this comprehensive industry assessment, the bank's analysts have significantly raised price targets for several semiconductor stocks and shifted the valuation base year from 2027 to 2028.
In memory chips, analysts raised the price target for Micron Technology (MU) from $950 to $1,500, maintaining a Buy rating. For semiconductor equipment, the price target for Applied Materials (AMAT) was raised from $540 to $720, with targets for several other companies like MKS Instruments also raised. In AI infrastructure chips, the report is positive on Marvell, Credo, Intel (INTC), and others.
Consumer Electronics and Smartphones: On the Sidelines of the AI Feast
Not all market segments are sharing in the AI bounty.
Smartphone chips are expected to decline 13% year-over-year in 2026, PC chips to fall 9%, and consumer electronics overall to drop 7%. These markets face structural pressures from declining shipment volumes, in stark contrast to the explosion in AI data centers.
The automotive and industrial markets are in a phase of moderate recovery: automotive chips are expected to grow 4% in 2026, and industrial chips 18%, mainly benefiting from demand replenishment after inventory digestion is complete and the continued increase in chip content per vehicle.
Comments