📑 Contents (5 sections)
  1. Four Pillars
  2. What It Bought This Year
  3. Institutions Established Locally in Singapore
  4. Rules It Set for Itself
  5. Two Tracks

· Singapore AI Observatory · Analysis  · 5 min read

Temasek Annual Report 2026: AI from 6% to 15%

Temasek's 2026 annual report devotes a full chapter to AI: AI-related investments currently account for 6% of the portfolio, with a target of 10–15% by March 2031. This article maps out the four pillars, this year's acquisition list, and several institutions based locally in Singapore.

On 8 July 2026, Temasek released its annual report, Temasek Review 2026. Within its 143 pages lies a dated, ranged commitment to AI: AI-related investments currently represent 6% of the entire portfolio, with a target of 10–15% by 31 March 2031.

Converted at this year’s portfolio scale: as of end-March 2026, Temasek’s portfolio was worth S$518 billion, so 6% is roughly S$31 billion, and 15% would be 2.5 times the current level. Over the year Temasek invested S$51 billion and divested S$31 billion; the gap between 6% and 15% is about the size of one year’s total investments.

This 6% carries a footnote: it excludes AI investments made independently by Singapore companies in which Temasek holds stakes (such as DBS, Singtel, and ST Engineering). In other words, the 6% figure refers only to AI assets that Temasek actively acquires on the global market.

Four Pillars

Temasek established a cross-functional AI team in 2019 (originally called an AI pod). This year’s annual report codifies years of accumulated work into a standalone chapter (§3.3 Our AI Strategy), with four pillars:

AI-Enabling Ourselves (implementing AI internally). All employees are equipped with multi-model generative AI environments; investment teams have custom tools spanning pre-investment due diligence, investment committee materials, and post-investment tracking; the investment committee is additionally staffed with an AI assistant tasked with providing independent perspectives and identifying blind spots. All employees attend prompt engineering workshops, and the annual CEO Challenge encourages staff to use AI to solve their own work challenges. Some high-value processes are already running autonomous task-executing AI (agentic AI).

AI-Proofing Our Portfolio (bringing portfolio companies up to speed). Assess the AI transformation progress of each Singapore company in which Temasek holds stakes, and co-develop with them a Workforce AI Fluency Playbook—an operational manual that teaches entire workforces how to adopt AI. Overseas holdings are subject to continuous screening for AI-related risks and opportunities.

Scaling Our AI Exposure (scaling up investments). That’s the 6% to 10–15% mentioned at the beginning. Investment scope spans five layers: energy infrastructure, semiconductors, cloud services, foundation models, and AI applications and software infrastructure. Four themes: Scaled AI Winners (platforms that have already emerged as major players), AI Innovators (AI-native challengers), AI Factories (data centres and energy—these kinds of “AI factories”), and AI Transformation (traditional companies undergoing AI-driven transformation).

Supporting AI Diffusion (spreading the capability). Pushing AI beyond its own portfolio companies: a third-party testing firm, leadership training, and executive study trips to Shanghai and Hangzhou to look at the local AI industry.

What It Bought This Year

Three foundation-model companies: Anthropic, OpenAI, xAI (later incorporated into SpaceX), and also invested in SpaceX. Two application-layer companies: CuspAI uses machine learning to discover new materials, PhysicsX applies AI to engineering simulation. One supporting company: co-invested with Warburg Pincus in Park Place Technologies, which provides third-party maintenance and repair for data centre hardware. For chips: increased holdings in ASML, Broadcom, and Nvidia, and newly acquired Lam Research. After the fiscal year ended: led the investment in TeraHop, the overseas subsidiary of Zhongji Innolight, producing optical modules—components for data transmission between data centres.

We previously compiled on the site what GIC and Temasek invested in AI, a list assembled from individual transaction news stories. This annual report for the first time articulated the selection criteria behind these transactions: five layers, four themes, and one dated proportion range.

Institutions Established Locally in Singapore

The part of the report most directly tied to Singapore sits outside the investment list—the institutions Temasek has set up around AI are all based in Singapore:

  • Temus’s AI Foundry. Temus is a digital services company established by Temasek; the AI Foundry cultivates local AI talent in Singapore and carries out AI transformation projects for Temasek-linked companies.
  • Aicadium. An AI engineering company established in 2021. The annual report provides a concrete example: it helped portfolio company gategroup use AI for quality inspection of airline meals.
  • Resaro. Established in 2023, it conducts third-party independent testing for critical AI systems. In October 2024, when Singapore’s Cyber Security Agency (CSA) released AI system security guidelines, the accompanying paper discussing AI security risks was co-authored with Resaro (see the SICW speech recorded on this site).
  • Temasek International Institute. In March 2026, it co-organized an AI Leadership Programme with Stanford Graduate School of Business—five days, over 70 participants including Temasek executives, portfolio company executives, and Singapore government officials. This was the first project of this new platform.

Rules It Set for Itself

The risk section outlines two internal governance mechanisms: the Responsible AI Use Committee, which reviews various AI use cases; and the Agentic AI Security and Governance Framework, which specifically manages risks from AI capable of autonomous task execution. This can be compared with IMDA’s recently updated Model AI Governance Framework for Agentic AI—the government sets rules for the nation; the sovereign fund sets rules for itself.

Two Tracks

Singapore’s AI funding operates on two tracks. This site has been tracking the fiscal track: NAIS 2.0, the 400% tax incentive in Budget 2026, and RIE2030’s S$37 billion total research budget (see the investment intensity dimension on the AI Dashboard). For the sovereign capital track, we could previously only piece it together from individual transaction announcements. This annual report has now provided the complete details of the second track: how much to invest, what to invest in, how to use it internally, and what institutions to establish in Singapore.

The next verifiable checkpoint is 31 March 2031. Two things to check then: whether AI-related investments have entered the 10–15% range, and what Temus, Aicadium, and Resaro have delivered in Singapore.

Sources

Share:
Back to Opinion

Related Posts

View All Posts »

What GIC and Temasek have invested in AI: the full public record

In May 2026 Anthropic closed a $65 billion Series H with GIC as co-lead and Temasek as a significant investor — Singapore's two sovereign funds on the same frontier AI company's cap table. This piece surveys every publicly known AI investment by GIC and Temasek: frontier models, AI infrastructure, data platforms, each with date, scale, and source tier.

Singapore tops the world in Claude usage intensity: what 5.53 means

Anthropic's March 2026 Economic Index ranks countries by Claude usage intensity, normalized by population, and Singapore comes first at 5.53. The number measures per-capita adoption intensity — not absolute volume, and not how advanced the usage is. Singapore's top spot rests on a sustained policy push and on its base as a small, rich, English-speaking, knowledge-worker-dense city-state.

From 50 People to 5,700,000 — One AI-native Architecture, Two Scales

Two AI-native experiments are running in parallel in 2026 — 50-person companies and a 5.7-million-person city-state. Put them side by side and you see an overlooked fact: AI-native is not a matter of scale — it is an architecture. The real bet of Singapore's Budget 2026 is to use the entire country as a "wrapper layer" for the AI-native transformation of its enterprises.