Outlook Supplement - Flipbook - Page 6
Forecast
While all 3 major DRAM suppliers
have new fabs coming online, it appears
they will be managing how fast they
ramp these fabs in order to maintain a
longer period of profitability. Micron
has even announced that it will be discontinuing a consumer line of DRAM
in order to support the more profitable
sectors. The higher pricing will help
drive the semiconductor revenues to
new heights, possibly exceeding the
$975 billion forecast.
Additional revenue drivers are
coming from the advanced packaging
sector, where TSMC is expanding its
CoWoS capacity from 61,000 wafers
per month to more than 100,000. This
added capacity will
support AI accelerators
and other advanced
compute chip sales in
2026.
Beyond AI, the
industrial, consumer,
and automotive
sectors should see
a recovery in 2026.
Power management
DEAN FREEMAN
chips supporting both
compute systems, as well as the grid infrastructure for hyperscale datacenters,
should see a moderate increase in 2026.
WSTS forecasts the automotive segment
to recover and grow near seven percent
in 2026. Although both automotive and
consumer segments will depend on
whether the consumer can stomach the
increase in pricing of mobile phones,
PC’s, and other electronic devices due
to the increase in memory costs.
Semiconductor equipment spending
in 2026 will be led by AI logic and
memory. The hope is that these sectors
will offset the decline in revenues from
China. Wafer fabrication equipment
revenue is projected near $130 billion,
depending largely on the pace of
DRAM capacity expansion.
Ultimately, success in 2026 hinges
on continued AI growth and hyperscale
datacenters, which remain constrained
4
by available energy supply and the
workforce needed to build and operate
new facilities.
Navigating the AI Boom in 2026
ANAND JOSHI, T e c h n i c a l
F el l o w, Te c h I n sig h t s
Artificial intelligence (AI) is everywhere, and it will remain the dominant
theme in semiconductor manufacturing
well into 2026. Currently, data centers
and cloud computing are where the
majority of AI models are being run,
and the race is on to build the largest
and smartest AI model. We expect this
trend to start diverging in 2026 and the
AI models to start shrinking in size,
making them amenable to a range of
devices.
Nevertheless, edge AI chip market
ramp up is expected only after the
model size starts shrinking. Our market
research shows the smart home and
tablet edge market rely heavily on cloud
processing today due to easier upgrade
capability, faster feature addition and
monetization. This reliance on the cloud
has vastly benefited hyperscalers and
has given rise to a range of small ‘neocloud’ companies that focus primarily
on AI services.
Computing performance needs in
AI datacenters are doubling every six
months, thanks to exponential growth in
AI workloads. High power consumption
of GPUs is propelling the power semiconductor market, particularly the data
center segment, which is estimated
to reach $4 billion in 2026. As such,
gallium nitride (GaN) is gaining interest
for high-efficiency, lower-capex applications, and is expanding into data center
power supplies. As the GaN industry
adjusts to this shifting landscape, the
redistribution of manufacturing capacity
and IP control will be a critical dynamic.
2026 is shaping up as a pivotal year for
how value flows across the fabless, IDM,
and foundry ecosystems. TechInsights
expects the overall GaN market to be
north of $900 million by 2026.
| Supplement to January 2026 Semiconductor Digest
New approaches in manufacturing
To support growing demand for
high-performance AI processors,
companies are investing in advanced
semiconductor manufacturing and
packaging technologies. Majority of
the AI accelerator chips today are in 4
and 3nm node, and we expect the first
2nm-generation design to hit the market
in 2026, with anticipated production of
over 500,000 wafers.
High-bandwidth memory (HBM) has
become essential for high-performance
AI in 2025. Despite high costs and
yield challenges, we expect more HBM
packaged into each AI chip in 2026.
While semiconductor packaging
remains a hot area of investment driven
by AI chips, the limitations of traditional
organic substrates will become more
pronounced in 2026, leaving glass as
the preferred alternative
material to combat
warpage under high
temperatures, coefficient
of thermal expansion
(CTE) mismatch, and
scaling challenges.
Similarly, we expect
adoption of rectangular
ANAND JOSHI
panel-level packaging
to optimize efficiency
and throughput of advanced processes
across market segments, including AI.
To reduce the power consumption in data
centers, we expect co-packaged optics
(CPOs) to become commonplace starting
2026. CPOs have shown that they can
reduce power consumption in data center-scale AI systems by up to 3.5X.
China is catching up
In efforts to establish dominance in
the global semiconductor technology
industry, the US has introduced a
series of sanctions in recent years to
restrict China’s access to advanced
semiconductor design and manufacturing technologies. Nonetheless,
Chinese firms continue to demonstrate
notable technological progress, both
www.semiconductordigest.com
