AI Economics · · 4 minute read
In five days, a reported US$50 billion pledge to Malaysia, a US$15 billion round for DeepSeek and a memory maker heading for a 78 per cent margin. The money behind Asian AI is no longer venture money, and that changes the risk an operator is buying.
Two weeks ago I wrote that the business of AI in Asia is a bundling business: the money follows cloud, commerce and ads, not the model. This week added a second layer. The people paying for Asian AI are increasingly not venture funds. They are conglomerates, carmakers, battery makers and governments, and they invest the way infrastructure owners do.
Three stories from the past five days show it.
On Saturday 24 October, Prime Minister Anwar Ibrahim said a global investor planned to put US$50 billion into AI, new technology and data centres in Sarawak and Peninsular Malaysia, calling it "one of the largest in the region" (The Star, 26 October 2026). CNBC reported the investor as SoftBank, with Masayoshi Son choosing sites in Kuching and the peninsula after a video call with Anwar (CNBC, 26 October 2026). Sarawak's premier said the Kuching portion was likely to sit at Tanjung Embang, planned as a low-carbon AI campus.
For scale, MIDA counts RM385.7 billion (about US$94 billion) of data-centre investment in Malaysia from 2021 to mid-2026, per the same CNBC report. The new figure is more than half of that in a single pledge, and roughly a tenth of national GDP (Yahoo Finance, October 2026).
Note who announced it. A head of government, not the company. That is how large AI capital arrives in Southeast Asia now: as a state-level deal about land, power and water, with the AI part decided later.
So what: A pledge is not capacity. If your plans depend on cheaper local compute in Johor or Kuching, track power connections and permits, not headlines. The West Java halt of BDx's 640MW campus over missing permits last month is the reminder (TechRepublic, September 2026).
On 22 October, Bloomberg reported that DeepSeek had raised about 100 billion yuan (roughly US$15 billion), led by Tencent and CATL, with Geely, Midea and GigaDevice among the backers. The company raised its target at least twice during talks, and reports put the valuation near US$74 billion (AI Weekly, citing Bloomberg and TechCrunch, 22 October 2026).
The round landed two days after the US Department of Homeland Security published an order removing DeepSeek products from federal civilian agencies, including third-party products built on its technology (BleepingComputer, October 2026).
Read the cap table as a customer list. A battery maker, a carmaker, an appliance maker and a chip designer are not buying a lottery ticket on a chatbot. They are securing a model supplier for their own factories and products, and a domestic one that Washington cannot switch off for them.
So what: For a Chinese or China-facing business, DeepSeek just became a safer long-term bet, not a riskier one. For anyone selling into US government supply chains, it became untouchable. Many Asian operators sit in both camps. Map which of your products and clients fall on which side before your vendor choice is made for you.
SK hynix reports third-quarter results on Wednesday 28 October. Consensus expects revenue of KRW98.23 trillion and operating profit of KRW76.96 trillion, a margin of about 78 per cent. The company said in July it had signed multi-year supply deals with about ten customers, and Alphabet is reportedly seeking memory contracts that run to 2035 (TrendForce, 26 October 2026). Samsung, Advantest and Tokyo Electron also report this week, which Bloomberg calls the AI trade's biggest test after the recent sell-off (Bloomberg, 25 October 2026).
A ten-year memory contract is a utility contract. The largest buyers are locking in supply a decade out, which means the smaller buyer gets whatever is left, at whatever price is left.
So what: This is the same story as the AI bill going up. Do not assume your AI costs fall every year. Budget a range, and prefer vendors who will commit to price for more than twelve months.
Venture money asks whether a product will find customers. Strategic money already is the customer, and state money wants jobs, land use and sovereignty. This week, Asian AI was financed by all three except the first.
That makes the region's AI build-out sturdier than a bubble and slower than a startup. It also means the decisions that shape your tools are being made in ministries and boardrooms you are not in. You cannot steer those. What you can steer is whether your own people understand the work well enough to switch vendors, models or bundles when the ground moves. That is the part no balance sheet buys for you.
Keep reading
AI Economics
Asia's biggest AI earners are not selling AI as a product. They sell it as a feature of something customers already pay for, and that changes how an operator should buy it.
AI Economics
Asia makes the chips, pours the concrete and writes the code for the AI boom. The profits sit with the first two. Most of the region's businesses are still on the paying side.
AI Economics
For three years operators budgeted AI as something that only gets cheaper. The first week of October says otherwise, and this region is both selling the scarce inputs and paying the higher price.