Morgan Stanley: AI Investment Enters Slowdown, but Chip Value Still Rising

TL;DR
·Morgan Stanley expects global major cloud providers' capex to approach $1.6 trillion by 2028, but year-over-year growth may slow to about 12%.
·However, AI semiconductors will not decelerate in sync with overall capex: global 2.5D advanced packaging capacity could still grow about 50% year-over-year in 2028.
·By 2030, Morgan Stanley expects the global semiconductor market to reach $1.5 trillion, with AI semiconductor TAM around $753 billion, nearly half of the entire industry.
·The next phase of growth is no longer just NVIDIA GPUs. Custom ASICs like Google TPU and AWS Trainium are expanding rapidly, with growth spreading to advanced process nodes, packaging, testing, and memory.
·China's AI demand is expanding, but chip capacity remains a key constraint; even including CXMT's new capacity, Morgan Stanley expects overall DRAM supply-demand gaps of about 17% and 15% in 2026 and 2027.

Capex Begins to Slow, but AI Semiconductors Have Not Peaked in Sync

Over the past two years, the most direct way for the market to gauge AI momentum has been to watch whether Meta, Google, Microsoft, and Amazon are still willing to keep raising capital expenditures.

In 2026, this indicator remains very strong. Morgan Stanley data shows that second-quarter capex for the four major cloud providers—Amazon, Google, Microsoft, and Meta—grew 87% year-over-year, while capex/EBITDA has risen above 70%.

But such high growth clearly cannot continue forever.

Therefore, this report extends the horizon to 2028. Morgan Stanley expects that by then, capital expenditures of the world's top 14 listed cloud service providers will approach $1.6 trillion, but overall year-over-year growth may have declined to about 12%.

Looking at this figure alone, it is easy to conclude that the AI investment cycle is beginning to cool. But the report immediately provides another set of data: global 2.5D advanced packaging capacity is still expected to grow about 50% year-over-year in 2028.

Morgan Stanley expects global 2.5D advanced packaging capacity could still grow about 50% year-over-year in 2028

This is precisely the most noteworthy part of the entire report.

The slowdown in overall cloud provider capex does not mean AI semiconductors have peaked in sync.

In the past few years, AI chip growth has been driven mainly by expansion in total capital expenditure: the more data centers built, the more GPUs purchased.

But entering the next phase, even if CSP capex growth falls from 50% or 80% to 10%-20%, as long as AI's share of overall capital expenditure continues to rise, GPUs, ASICs, HBM, advanced process nodes, and advanced packaging can still outperform total capex.

In other words, the market is shifting from a "total volume story" to a "structural story."

Morgan Stanley's long-term market size forecast also reflects this. The report expects that by 2030, the global semiconductor industry could reach about $1.5 trillion, with AI semiconductor TAM around $753 billion, nearly half of the entire market. Its supply-chain-driven bull case even projects that the cloud AI semiconductor market could reach $485 billion in 2026. It should be noted that $485 billion is explicitly labeled as a bull case in the report, not a base-case forecast.

Meanwhile, the divergence between AI and non-AI semiconductors is widening.

The report expects that if memory and NVIDIA AI GPU revenue are excluded, non-AI semiconductor growth could actually decline in 2026. This means that this cycle can no longer be simply framed as a traditional "broad semiconductor industry recovery."

The real high-growth area is an independent industry chain formed by AI's continued absorption of wafer, memory, advanced packaging, and testing resources.

Growth Logic Shifting from "Buy More GPUs" to Higher AI Value Content

If the core of the first phase of AI investment was "buy more GPUs," the more important question in the second phase is: how much more expensive and complex semiconductor content does each AI server actually require.

TSMC is the most direct example.

Morgan Stanley expects AI semiconductors may already account for more than 30% of TSMC's 2026 revenue. Meanwhile, TSMC's N2 family capacity is expected to achieve a compound growth rate of about 70% between 2026 and 2028; N3 capacity continues to increase, while N5 may begin to decline from 2027.

AI semiconductors' share of TSMC revenue continues to rise and may already exceed 30% in 2026.

The logic behind this is not complicated.

AI chips are not only increasing in quantity, but each individual chip is also becoming more expensive. Moving from 5nm to 3nm and 2nm, from standard packaging to CoWoS and SoIC, and adding more HBM—each generation of AI accelerators is increasing the wafer, packaging, and memory value content per chip.

Therefore, even if server unit growth gradually slows in the future, semiconductor value content per server may continue to rise. This is why advanced packaging growth can far outpace overall capex.

Meanwhile, the source of AI chip growth is also beginning to spread from NVIDIA to custom ASICs.

Morgan Stanley believes that even as NVIDIA continues to launch more powerful GPUs, large CSPs still need their own custom chips. Google has TPU, Amazon has Trainium, and companies like Meta are also advancing their own ASIC projects.

When a cloud provider has sufficiently large computing demand, in-house ASICs can optimize cost, performance, and energy efficiency for specific workloads while reducing dependence on a single GPU supplier.

Therefore, the future AI chip market may not be a question of "who takes over after NVIDIA's growth ends," but rather GPUs continuing to grow while ASICs become a second growth curve.

Morgan Stanley expects Alchip's revenue from AWS Trainium to rise from about $1.8 billion in 2026 to $2.8 billion in 2027, and reach $8 billion in 2028, at which point Trainium revenue could account for about 82% of Alchip's total revenue.

The forecast for Google TPU is even more aggressive.

The report expects that under its model, MediaTek's TPU-related revenue could rise from $13.5 billion in 2027 to $43.5 billion in 2028, and further to $70 billion in 2029; TPU revenue could already account for about 65% of MediaTek's total revenue in 2028.

Morgan Stanley expects CSP in-house ASIC projects to continue increasing, with custom chips like Google TPU entering a rapid volume ramp phase.

These figures are all long-term model projections, and the ultimate realization still depends on chip production, yields, customer procurement, and cloud providers' own capital expenditures. But the direction is very clear: AI industry chain growth is expanding from pure GPU shipments to ASICs, advanced process nodes, advanced packaging, ABF substrates, and chip testing.

The report even specifically notes that Google TPU's future volume ramp potential may be constrained by ABF substrate supply. Therefore, when judging the AI cycle, "how many GPUs were sold" remains important, but it is increasingly not the whole story.

China's AI Demand Is Rising, but the Real Bottleneck Is Still Capacity

The Chinese market presents a different kind of change.

In the past, discussions of domestic AI chips focused mainly on "import substitution." But Morgan Stanley this time places more emphasis on demand creation.

The report believes that DeepSeek has demonstrated lower-cost AI inference capabilities, which could further stimulate inference demand; at the same time, China's domestic wafer manufacturing supply chain is also improving its ability to produce AI GPUs.

This means that China's AI chip logic is gradually shifting from pure "import substitution" to: lower inference costs → more applications → expanded computing demand → increased demand for domestic chips.

Therefore, in the China AI and semiconductor equipment space, Morgan Stanley highlights Iluvatar, Cambricon, Hygon, Naura Technology, and AMEC, and SMIC is also included in the AI Overweight portfolio.

But demand is not the biggest current constraint. In summarizing the major bottlenecks for global AI development, the report characterizes the U.S. issue as Energy, while China's is Chip Capacity. In other words, how much volume China's AI chips can ultimately ship depends more on how much effective capacity can be released from advanced process nodes, advanced packaging, and memory.

Memory is the most typical example.

Morgan Stanley expects global DRAM capacity to reach about 3,374 kwpm by 2028, of which CXMT accounts for about 500 kwpm, or about 15% of the global total; from 2025 to 2028, CXMT's bit shipment is expected to achieve a compound growth rate of about 40%.

Meanwhile, CXMT is also beginning to enter the HBM market.

The report expects its HBM TSV capacity to rise from 20 kwpm in 2026 to 40 kwpm in 2027 and 70 kwpm in 2028. By bit shipment, CXMT could account for about 3.8% of the global HBM market in 2026 and about 4.4% in 2027.

But this has not immediately changed the global memory tightness.

Even including CXMT's new capacity, Morgan Stanley's model shows that the global DRAM supply-demand gap could still reach 17% in 2026 and about 15% in 2027; the HBM market also remains in short supply.

CXMT's DRAM and HBM capacity is expanding rapidly, but Morgan Stanley expects global memory supply will still struggle to keep up with AI demand growth.

This is also one of the biggest differences between this memory cycle and past ones.

In the past, DRAM and NAND were more like typical PC and smartphone cyclical products: demand rises, prices increase, manufacturers expand capacity, and then oversupply follows. But in AI servers, HBM, server DRAM, and high-performance memory are increasingly becoming infrastructure bottleneck assets. They are no longer just supporting components next to GPUs, but are beginning to directly determine whether an AI server can be produced and whether an entire computing cluster can expand as planned.

Ultimately, what this 61-page report really wants to convey is not simply that "AI will keep rising." A more accurate statement is: AI semiconductors are moving from the first phase, which relied on rapid CSP capex growth, into a second phase driven by rising AI share and rising value content per server.

Therefore, to judge whether the AI hardware cycle is truly approaching an inflection point, it is no longer enough to look only at the capex growth rates of Meta, Google, Microsoft, and Amazon.

More worth watching are AI capex share, CoWoS/EMIB capacity, HBM supply-demand, 2nm utilization, and the actual shipment pace of ASICs such as Google TPU and AWS Trainium.

If overall CSP capex growth declines in the future but these indicators remain high-growth, then AI semiconductor momentum may not peak along with total capex. Conversely, only when these core indicators begin to weaken simultaneously will it be closer to a signal that the AI semiconductor cycle is truly at an inflection point.