Amid China woes, AI prayers, global semiconductor industry to grow 16% in 2025!

The global semiconductor forecast for 2025 has been updated to +16% (+/- 1 percent). That is $731.609 billion, with bear at $725.318 billion, and bull at $737.544 billion. This was disclosed by Malcolm Penn, Founder and CEO, Future Horizons, UK. He was speaking at the IFS 2025 in London, UK.

2026 outlook for global semiconductor industry has been currently at +12% (+/- 6%). That will be $813.104 billion, with bear at $764.538 billion, and bull at $863.121 billion. 

Trump 2.0’s ‘America First’ agenda now comes first! US is no longer a champion of free trade, global security, or gold standard on governance. US tariffs and protectionism have now been raised to the highest level since 1930s. 

Trump 2.0 shockwaves are nearly all over, with the negotiating pattern now understood. Tariffs and trade barriers are largely absorbed, and taken in stride, for now. Uncertainty levels are less unsettling, than in Jan. 2025.

We are also seeing disavowed champion of post war Washington consensus and leadership. There is total disregard for red lines on Central Bank, judicial, and statistic agency independence. There is 180-degree pivot from free trade and global security to sharp focus on national interest.

The post-WW2 drive for sustainable global peace is currently in tatters. We now have Benjamin Netanyahu's and Vladimir Putin’s blatant geopolitical aggression in Qatar and Poland. Xi Zinping has heightened the threat to Taiwan.

We are also seeing the rise of China as new economic global superpower. China, North Korea and Russia, plus India, and the emerging SE Asia, are set to rival the US global dominance. There’s no going back! USA is no longer the world’s sole superpower.

We also have heightened geopolitical tensions with Russia, over Ukraine / NATO), and Israel, over Gaza / Qatar. The overall economic fundamentals remain fragile and weak.

Mixed background and 2026 outlook
After two years of AI-ASP upcycle, mainstream products cautiously entering the fray. Historically, the boom-bust cycle persists beneath insane Nvidia success. We have 10-quarter 2H-2022 downturn vs. normal three-to-five quarter crash. There was 2020-22 Covid-19-fueled chip shortage, and consequent overstocking in the classic upcycle. Downcycle was prolonged due to LTA-enforced customer shipments, without demand.

Own-goal alienated customers, leading to stuffed bloated inventories, and undermined the next upturn. The upturn is now imminent, but severe over-capacity means strong rebound won't be likely. Growth is also vulnerable to US government tariff flip-flopping and shipment pull forwards.

Supplier / distributor / customer inventory visibility has been poor, but is declining slowly. It is unclear whether shipments are yet aligned with real market demand. Also, inventories might have depleted, but the end market demand is soft. China, the world's largest market, has increased domestic capacity and supply. We also have a lurking fear: what if the AI data center boom slows down or crashes?

We are quite unclear which way the wind blows! We have a current conundrum -- 3/12 and 12/12 are both stubbornly refusing to fall! Are we going have a re-run of 2018’s extended period of uncertainty?

If we look at non AI-chip companies revenue growth and guidance, Q2- vs. Q1-2025 had 6% vs. 8% overall market. It was 22% growth for ICs, while opto and discretes were both negative. Average Q3 growth is in 5-9% range.

Looking at the annualized semiconductor growth rates by sector, opto and discretes growth turned positive in Q2. As too did analog, but not yet so, micro. That's some good news, following 10 months of bad news.

China woes
China has been the main capex overspend culprit. China capex share accelerated after June 2022 market collapse. 2025 spend was 4x 2010-15 level, with decoupling, tariffs, and export controls driven. Capex spend is now slowing, but it is still 3x the market justifiable levels. China is focusing on non-leading edge, but making progress on 5/7nm with Huawei and SMIC.

China's own-grown wafer fab equipment vendors are also on the rise. Thanks to this, a big chunk of the domestic China market has now gone away. Once gone, it will ‘never’ come back.

However, turning capex into capacity is proving 'challenging'. $50-100 billion was lost even before a single wafer was made. Jiangsu Advanced Memory Semiconductor (AMS) turned bankrupt. Wuhan HSMC saw no chips produced. QXIC's 14nm-class plans never realized. GlobalFoundries Chengdu was halted In 2020. And, Woodson Shanghai went bankrupt.

This is a case of money not well spent! There has been lack of expertise, overly ambitious goals, and restricted access to key tools. They simply could not complete their tools list.

India going for broke
Now, India is going for broke too! If you don’t believe it can be done, it won’t be done. Now, India is believing in themselves! They are putting in the money where the mouth is! Let's see!

Forecast 2025
Back in Jan. 2025 forecast, Future Horizons had made many statements. Monthly annualized growth rates were trending steadily down. Excess capacity will plague the industry throughout 2025. China has real potential to flood the world with mature node technology. We have an increasingly closed China market due to embargoes and Made In China.

The global semiconductor forecast for 2025 has been updated to +16 percent (+/- 1 percent. That is $731.609 billion, with bear at $725.318 billion, and bull at $737.544 billion.

The annual growth rates are still strong, and ASP driven. Discretes, opto, and analog markets are showing early signs of recovery. Excess inventory and over-capacity is still undermining the industry growth. China has the potential to flood the world with mature node technology. 

Also, watch out for a slowdown or hiccup in AI the infrastructure demand. We also need to keep praying that the AI boom doesn't crash and burn!

2026 outlook for global semiconductor industry has been currently at +12 percent (+/- 6 percent). This will be $813.104 billion, with bear at $764.538 billion, and bull at $863.121 billion. 

There are too many variables for anything, other than a swag. There are assumptions that US economy continues to defy gloomy expectations. Geopolitics, tariffs, and trade restrictions do not trigger global slowdown. AI data center growth does not slowdown, crash, or burn. Finally, IC unit growth resumes in 1H-2026. The next supercycle is not likely before 2029 earliest, with China / capex dependent.

Tech roadmap
There are tech roadmaps through 2039 that have potential roadmap extension. TSMC is looking to turn the roadmap into real production. 2026 will see A14, and A2 will happen by 2036. We will also witness device and material innovations, and context-aware interconnects. 

GAA transition was never going to be easy. The last nodes never die! N3 is a family of different chip types for different applications. N3 is currently manufactured at Fab 18 in Taiwan, with capacity planned in the USA. Now, we have a four-horse GAA race in the offing. There are TSMC N2 (GAAFET), Intel 18A (RibbonFET), Samsung SF2 (MBCFET), and Rapidus 2nm (2HP). Rapidus is using IBM technology. We do have competition for TSMC.

Power semiconductors
What's happening with power semiconductors? Silicon historically wins, and always will! SiC and GaN made significant gains in recent years. Diamond power devices are ‘coming to market soon’ since forever! 2D materials offer many advantages. But, only once have the problems of fabricating them were solved! We also have interesting advances from several European Universities recently.

Better (or best) is irrelevant. Only ‘cheap and good enough’ matters. When you block the front door, innovation finds the back window. (This was clearly shown by DeepSeek, in China, as an example). Well, there are no ‘high volume’ niches in semiconductors!

China had invested more in GaN fabs, than SiC. TSMC announced that it would be exiting GaN wafer processing by 2027. Wolfspeed was the first of the several SiC firms to face Chapter 11 / bankruptcy.

Makimoto's wave
Next, imec had integrated GAN on SoI. Some years ago, imec developed an interesting GaN-on-SoI technology capable of integrating many devices on a single substrate. Even now, most modules utilize separate die for switching devices, drivers, and diodes.

It is not clear why there hasn’t been greater adoption of this technology. It is certainly on the radar screen.

Makimoto's wave
That brings us to Makimoto's wave, published by Electronics Weekly, 1991. It is the cyclic nature of semiconductor industry in which the “age of standardization” and the “age of customization” alternately appear approximately every 10 years. The semiconductor pendulum has been swinging away. The semiconductor industry is still holding good through 2037, at least. We are seeing more advanced CPUs and GPUs, and moved further on to chiplets.

Next, packaging has moved from begrudgingly to enabling! There is battle for survival supremacy --- between ASAT vs. wafer foundry. Advanced packaging doesn’t scale up that easily. And, this is the fact that gives ASATs a huge competitive advantage. Both sides are fighting hard, and going for it. We have to wait and see!

Next disruptive innovation?
So, what's going to be the next disruptive innovation? Clue -- it's not AI! We have had several disruptive innovations in the past. It started with the invention of transistor (Shockley, Bardeen, Brattain: 1947), invention of IC (Kilby: 1958 / Noyce: 1959), microprocessor development (Faggin, Shima, Hoff, Mazor: 1971), and quantum computing (IBM / Google / Microsoft / China).

It will, once we learn how to do it!

AI, as we currently know, is just an improved Windows / Office / Google! It has been doing what Moore’s Law’s been doing for six decades – more bang / same buck.

We also have the AI data center madness! We have projects, such as  Float on Canals in the Netherlands, 10GW in Switzerland / Hong Kong, and Under The Sea (China). A 1GW plant supports 1 million people. It is similar to AI ‘Hyperdrive FTL’ Poker Game “I’ll Call You for US$ 2.4 Trillion”. Who is going to pay this money back?

There are AI smart glasses, but, the least said, the better! It's just a mess, right now.

In the 1970s, there was billion-dollar watch disaster, with TI, FSC, and Intel losing billions. We need to pay heed to lessons of history. We can't have square pegs and round holes. Ego always trumps wisdom! We have four more decades to go, to 2060s!

A good news is here: we have the first AI killer app -- a camera-less smart phone. There are things like: why capture moments on camera when you can just invent them? There is FriendCastAds! Who wouldn't buy products if your friends and mum recommended it? There is the iWant app that buys stuff and delivers, before you knew you even wanted it! If that camel turns up on your doorstep, be happy, and pleased!

Summary
So, what's driving the chip market? It's the same as always! In technology, we have Moore’s Law and learning curve. Legislation leads to conservation/safety. Structural -- relentless A to D conversion/geographic diversification.

All these are combining to do what the chip industry does best! It entertains, removes drudgery, or enables something previously impossible -- either technically, practically, or, commercially.

All killer apps take a long time to surface. Transistor took eight years, as did the IC. DRAM took 10 years, while the MPU took nine years. While PCs came in 1971, the Netscape Navigator happened 23 years later in 1994. Flash EEPROM took nine years. Can Chat GPT 2022 be the first killer AI application in 2030? We need to proceed with extreme caution. 

There is near-term ASP boost, but structural risks remain. The global semiconductor industry should grow 16% in 2025, with bull at 17%, and bear at 15%. We have to be prepared well for the future.

Rocky roads ahead
On the subject of the Intel, Softbank and Arm collaboration, Penn described that as rocky! There are two dinosaurs, and both rested on their laurels, but now, in catch up mode. That’s a bad place to be. It will take a minimum of 10 years to execute, and that’s if the investors don’t lose patience.

Next, how long can the US tariffs last, and can there be any change?  He said tariffs could stay for as long as Trump wants them. That is at least three plus years, and, yes, with lots of changes enroute. And, who knows what the next president might do?

We also discussed Section 232, and where will it be used?  Now that's going to be the big catch! All, depending on what Trump decides, is ‘national security interest’.

Finally, what's the future of the various Chips Acts from the EU and USA? He said that subsidies will prevail, but the objective of massive reshoring was never going to happen. It takes ‘three years’ to build a new fab by which time Trump’s reign is over. Plus, neither Act is being driven by the end users. They’re not really interested, especially, as its going to increase their costs.