AI export bet: Here to stay?

MALAYSIA has never exported so much, or depended so heavily on one sector to do it.

In September, shipments hit RM215.1bil, up 54.5% from a year earlier. Electrical and electronics (E&E) products accounted for RM110.23bil of that, a monthly record and 51.2% of total exports, according to trade data from the Investment, Trade and Industry Ministry and the Statistics Department.

MBSB Research calculates that E&E delivered about 59% of the month’s growth, led by integrated circuits and storage devices. Orders from Taiwan jumped 175.2% and those from the United States rose 89.9%.

Behind those numbers is one theme: the global rush to build artificial intelligence (AI) infrastructure. Now the money funding it is looking nervous.

China’s tech-heavy STAR50 index has fallen about 40% in under four months, erasing roughly US$480bil.

Last Sunday, Elon Musk warned on X that more than 200 “neoclouds” – firms that buy chips and rent out computing capacity – could face “guppy-level death and consolidation”.

“Debt markets have almost entirely shut off access to money for any company that doesn’t have substantial revenue independent of AI data centres,” Musk wrote.

Neoclouds buy the chips and servers that Malaysian plants package and test, so they sit at the demand end of the chain.

This raises the question: Is the bubble bursting?

Peter Lim Tze Cheng, chief research officer at Trident Analytics, says no.

“If you look at the share price aside, the fundamentals for the semiconductor and E&E are still very strong,” he tells StarBiz 7. “I think the AI boom has barely started.”

His evidence is revenue. Anthropic’s second-quarter revenue rose 1,368% to US$11.6bil, according to its S-1 initial public offering (IPO) prospectus filed with the US Securities and Exchange Commission.

Lim says this separates the current cycle from the dot.com bubble, when “there’s no revenue, there’s no business model”.

Demand is physical too, he says. “Are we seeing a volatility in demand for the chips? The answer is no,” Lim adds.

Memory is in short supply as producers shift capacity to high-bandwidth memory, and “the current shortage is mind-boggling”.

He cites a US Semiconductor Industry Association forecast of global chip sales rising from about US$800bil to US$1.5 trillion.

China’s slide, he says, reflects very high starting valuations and little comparable AI revenue, rather than a collapse in demand.

On Musk’s warning, “neocloud” is too broad a label. What separates these companies is their financing, and the weak spot is operators whose loans depend on the future resale value of their graphics processing units.

Malaysia does not design or fabricate the chips. That is the work of firms such as Nvidia Corp and Taiwan Semiconductor Manufacturing Co Ltd (TSMC).

Its role lies in packaging and testing, through listed outsourced semiconductor assembly and test (OSAT) firms Inari Amerton Bhd, Malaysian Pacific Industries Bhd and Unisem (M) Bhd , as well as makers of test equipment.

“Our exposure is indirect,” Lim explains.

The 175.2% jump in exports to Taiwan, home to TSMC, shows how deeply Malaysian plants are embedded in that supply chain, but it also ties the country’s fortunes to a small cluster of customers.

The counterpoint is that scarcity doesn’t last. Tradewind chief executive officer (CEO) Ng Zhu Hann does not dispute current demand. He questions how long the shortage will last.

Chip users may widen their sourcing beyond Samsung, SK Hynix and Micron to include ChangXin Memory Technologies (CXMT), the Chinese memory maker.

“If so, the capacity constraint will then be easily overcome,” he says.

Following a domestic IPO on the Shanghai Stock Exchange in July 2026, CXMT briefly became China’s most valuable listed company, capturing roughly 10% of the global dynamic random-access memory market by mid-2026.

That matters because scarcity is the heart of the bull case.

“It doesn’t mean, especially in a technology space, that the cost will remain high,” Ng says. “Breakthroughs will happen and within a very short span of time, things can also change.”

He also calls the sector’s share-price run-up “a bit dangerous”, since valuations may already have priced in strong earnings.

Ng is speaking about equities, not export orders. But the two views are less opposed than they seem.

Lim says demand is strong and prices are noisy. Ng says demand is strong and prices may assume it stays that way.

The what-if situation

Danny Wong, CEO of Areca Capital, frames it as a question of how long, not whether.

“Malaysia’s exports are doing very well, especially electronics and computer chips, attributed to the AI boom,” he says. “But the big question is how long this growth can last.”

While the shortage and demand persist, exports would be supported, “albeit normalise on its growth”.

If the boom stalls, the damage would probably come in stages. Wong reads the STAR50’s fall as “a warning sign that the boom may be losing steam”, but not a sign that production is slowing.

If big companies spend less on AI, though, “Malaysia could face fewer export orders, slower factory production and even job losses, but not as severe as other high-end production countries”.

The second stage is a supply response. If Ng’s scenario plays out, prices could ease and revenue growth slow even without a fall in volumes. The question is whether that would hurt OSATs or help them through higher volumes.

The third is arithmetic. September’s growth beat forecasts of about 40% by a wide margin, according to Hong Leong Investment Bank Research, so comparisons will only get steeper.

“We should expect the growth to normalise,” Wong says.

None of the three expects a collapse.

“We must not focus just on AI-related demand. There are other growth drivers for our E&E sector, such automotive/electric vehicle, industrial automation, power/energy transition, consumer electronics/telecommunications/aerospace, to name a few,” Wong adds.

For a country whose exports have never leaned so heavily on one sector, the answer will determine how prepared it needs to be.

“The AI boom could slow down at some point, and Malaysia must be ready for it,” he says.

Already a subscriber? Log in

Get 20% OFF The Star Digital Access

Cancel anytime. Ad-free. Unlimited access with perks.

RM 13.90/month RM 11.12 /month

Billed as RM 11.12 for the 1st month, RM 13.90 thereafter.

RM 12.33/month RM 9.87 /month

Billed as RM 118.40 for the 1st year, RM 148 thereafter.

Thank you for your report!

Dialog tanking up for growth

Beyond dividends – a profit push

Banking on value-creation

马来西亚的出口规模从未如此之大，也从未如此严重地依赖单一行业。

9月份，马来西亚的商品出口额达到2151亿令吉，比上年同期增长54.5%。根据投资、贸易和工业部以及统计局的贸易数据，其中电子电气产品出口额为1102.3亿令吉，创下月度新高，占出口总额的51.2%。

MBSB Research 计算得出，电子电气设备行业贡献了当月约 59% 的增长，其中集成电路和存储器件的增长最为显著。来自台湾的订单激增 175.2%，来自美国的订单增长 89.9%。

这些数字背后隐藏着一个主题：全球竞相建设人工智能（AI）基础设施。如今，为之提供资金的各方似乎有些紧张。

中国科技股为主的科创板50指数在不到四个月的时间里下跌了约40%，市值蒸发了约4800亿美元。

上周日，埃隆·马斯克在 X 上警告说，超过 200 家“新云”（购买芯片并出租计算能力的公司）可能会面临“像小鱼苗一样的死亡和合并”。

马斯克写道：“债务市场几乎完全切断了任何没有独立于人工智能数据中心的实质性收入的公司获得资金的渠道。”

Neoclouds 购买马来西亚工厂包装和测试的芯片和服务器，因此它们处于供应链的需求端。

这就引出了一个问题：泡沫是否即将破裂？

Trident Analytics首席研究官Peter Lim Tze Cheng表示否定。

“抛开股价不谈，半导体和电子电气行业的基本面依然非常强劲，”他告诉《星报商业7》。“我认为人工智能的繁荣才刚刚开始。”

他的证据是营收。根据Anthropic公司向美国证券交易委员会提交的S-1首次公开募股（IPO）招股说明书，该公司第二季度营收增长1368%，达到116亿美元。

林表示，这使当前的周期与互联网泡沫时期截然不同，当时“没有收入，也没有商业模式”。

他还表示，需求也体现在实物上。“我们是否看到芯片需求出现波动？答案是否定的，”林补充道。

由于生产商将产能转向高带宽内存，内存供应短缺，“目前的短缺情况令人难以置信”。

他引用了美国半导体行业协会的预测，即全球芯片销售额将从约 8000 亿美元增长到 1.5 万亿美元。

他表示，中国的下滑反映出估值过高和可比人工智能收入很少，而不是需求崩溃。

正如马斯克所警告的，“新云”这个标签过于宽泛。这些公司之间的区别在于它们的融资方式，而薄弱环节在于那些贷款依赖于其图形处理器未来转售价值的运营商。

马来西亚并不设计或制造芯片。这项工作是由英伟达公司和台积电（TSMC）等公司完成的。

其作用在于封装和测试，通过上市的外包半导体组装和测试 (OSAT) 公司 Inari Amerton Bhd、Malaysian Pacific Industries Bhd 和 Unisem (M) Bhd，以及测试设备制造商来实现。

“我们面临的风险是间接的，”林解释说。

对台积电总部所在地台湾的出口增长了 175.2%，这表明马来西亚工厂与该供应链的联系有多么紧密，但也表明该国的命运与一小群客户紧密相连。

另一种观点认为，稀缺性不会永远持续下去。Tradewind首席执行官吴竹瀚并不否认当前的需求，但他质疑这种短缺状况会持续多久。

芯片用户可能会将采购范围从三星、SK海力士和美光扩大到包括中国存储器制造商长信存储技术股份有限公司（CXMT）。

“如果真是这样，产能限制问题就很容易克服了，”他说。

2026 年 7 月，CXMT 在上海证券交易所进行国内 IPO，短暂成为中国市值最高的上市公司，到 2026 年年中，占据了全球动态随机存取存储器市场约 10% 的份额。

这一点很重要，因为稀缺性是牛市的核心论点。

“但这并不意味着，尤其是在科技领域，成本会一直居高不下，”吴说。“突破性进展总会发生，而且在很短的时间内，情况也可能发生变化。”

他还称该行业的股价上涨“有点危险”，因为估值可能已经反映了强劲的盈利预期。

吴先生谈论的是股票，而不是出口订单。但这两种观点之间的对立程度远没有表面看起来那么严重。

林先生认为需求强劲，但价格波动较大。吴先生则认为需求强劲，价格可能预计会持续走强。

假设情况

Areca Capital 的首席执行官 Danny Wong 将其视为一个时间长短的问题，而不是是否会发生的问题。

“马来西亚的出口表现非常出色，尤其是电子产品和电脑芯片，这主要归功于人工智能的蓬勃发展，”他说道。“但最大的问题是，这种增长势头能持续多久。”

尽管短缺和需求依然存在，但出口将得到支持，“尽管其增长将趋于正常化”。

如果繁荣停滞，损害可能会分阶段显现。黄认为，STAR50指数的下跌是“繁荣可能正在失去动力的警告信号”，但并不意味着生产正在放缓。

不过，如果大公司减少在人工智能方面的投入，“马来西亚可能会面临出口订单减少、工厂生产放缓甚至失业，但不会像其他高端生产国那样严重”。

第二阶段是供应反应。如果Ng的预测成真，即使销量没有下降，价格也可能回落，收入增长也会放缓。问题在于，这会对OSAT企业造成损害，还是会通过提高销量来帮助它们。

第三种情况是算术。根据丰隆投资银行研究部的数据，9月份的增长远超此前约40%的预期，因此未来的同比增幅只会更大。

黄先生表示：“我们应该预期经济增长会趋于正常化。”

三人均不认为会出现崩溃。

“我们不能只关注人工智能相关的需求。电子电气行业还有其他增长动力，例如汽车/电动汽车、工业自动化、电力/能源转型、消费电子/电信/航空航天等等，”黄补充道。

对于一个出口从未如此严重依赖单一行业的国家来说，答案将决定它需要做好多大的准备。

“人工智能热潮总有一天会放缓，马来西亚必须为此做好准备，”他说。

已经是订阅用户？登录

订阅《星报》数字版，立享八折优惠

随时取消。无广告。无限畅享，更有专属福利。

每月 13.90 马币 每月 11.12 马币

第一个月收费 11.12 马币，之后每月收费 13.90 马币。

每月12.33令吉 每月9.87令吉

第一年收费 118.40 马币，之后每年收费 148 马币。

感谢您的报告！

对话蓄势待发，促进增长

超越股息——利润驱动

依靠价值创造