Why hosting US and Chinese AI giants is only step one for Singapore

The headlines write themselves. OpenAI and Google DeepMind have opened applied AI labs in Singapore; Anthropic, the maker of Claude, is advertising its first local roles in finance, product support and economic research; and Chinese champions, from Alibaba and Tencent to newer entrants such as Manus, have made Singapore their springboard into global markets. Fortune recently asked whether the country can become “Asia’s neutral AI hub”. For a nation that has spent decades selling trust and predictability, the timing looks close to perfect.

But hosting the giants is the easy part. The harder question — the one that decides whether this moment compounds into lasting advantage — is whether these firms become embedded collaborators that build local capability bearing in mind technology safety, or foster high-wage sales enclaves that drain talent and leave the economy more exposed than before.

A front office or lab is not an automatic transfer of capability

If a frontier lab opens only a commercial front office to book regional revenue, or starts a local research group, Singapore gains some jobs and tax receipts but little durable expertise: the technology is still designed, trained and governed elsewhere. What matters is what economists call diffusion — the absorption of a general-purpose technology into the everyday workings of firms. Diffusion, not the nameplate on an office tower in one-north, is where the productivity gains live, and it happens only when foreign talent works shoulder-to-shoulder with local banks, hospitals, manufacturers and universities.

There are encouraging signs. OpenAI’s Applied AI Lab — its first outside the US — is built around “forward-deployed engineers” who embed inside customer organisations, and Anthropic’s advertised regional economist is meant to work with governments and academics to study AI’s economic effects. These are collaboration-shaped roles, not merely sales. Policymakers should make such embedding a condition of entry rather than the exception — tying incentives, data access and subsidised compute to genuine co-development and open publication.

The talent that matters is interdisciplinary

The most immediate effect of the influx is on the labour market, and it is not unambiguously good. Chinese tech giants are already dangling pay packages of S$150,000 (US$116,153) to S$273,000 for AI PhDs, and the American labs pay comparably. For an individual engineer this is wonderful news. For a local bank, hospital or start-up trying to build its own AI team, it is a bidding war against the very vendors selling them software. The danger is an enclave economy: a thin, richly paid, foreign-facing layer sitting atop a domestic economy that can no longer afford to keep the same people.

The way out is to be clear about which talent creates the most value. It is rarely the pure modeller; it is the person fluent in both a domain — finance, medicine, logistics — and in AI itself. Singapore’s own strategy names this well: the goal is “AI bilingual talent”, people who can carry a frontier technique into the messy particulars of a real industry. Such people are, by definition, interdisciplinary; they emerge not from siloed computer-science departments or business schools but from institutions built across finance, computing, economics, law and policy.

This is where the local research base matters as much as the foreign arrivals — and where Singapore is moving. The Nanyang Technological University (NTU) and National University of Singapore (NUS) already rank among the world’s leading universities, and NTU has just launched the Global Institute of Finance, Technology, and Society (GIFTS) to work across exactly those disciplines, with joint PhD pathways, industry roundtables and policy research aimed at ensuring safe frontier AI and turning it into local capability rather than a foreign import. Anchors like these give the arriving labs a serious partner to co-develop with, and give Singapore a way to retain talent through mission and mobility, not by winning every contest on salary.

For financial services, the arrival of cutting-edge labs genuinely helps. The local banks are past experimentation — they already run generative-AI tools across tens of thousands of staff, and DBS and UOB are moving fast at collaborating with top AI researchers. Having Anthropic, Google’s DeepMind and OpenAI engineers in the same time zone, under the Monetary Authority of Singapore’s Veritas/FEAT governance principles, shortens the distance from a frontier model to a deployed, compliant application; the new National AI Missions target financial services directly.

The catch is the inconvenient truth that these same firms compete with the banks for both business and talent. An institution that outsources its intelligence to a single external model becomes dependent on a vendor that may also be courting its best engineers and its customers.

A recent event offered a sharp reminder of what dependency means: when Washington abruptly restricted foreign access to Anthropic’s most powerful Mythos-class models in mid-June, overseas users were cut off with almost no notice, and access was only restored on 1 July after new safeguards were negotiated. Frontier-model access can be switched off by a single government decision. Prudent institutions will hedge — several providers, open-weight options where appropriate, and enough in-house capability that they are never hostage to one lab or one capital. At the national level, sovereign capacities and carefully governed collaborations are necessary.

The US-China contest: tailwind and tail risk

Does great power rivalry help Singapore or endanger it? In the near term, it helps. Chinese researchers often cannot operate globally without state approval and Chinese firms need world capital, making Singapore a necessary staging post; American firms, in turn, want an Asia-Pacific base both sides trust. Contestation pushes business toward neutral ground.

But neutrality is thinner than it looks. When Meta tried to buy Singapore-relocated Manus, Beijing looked straight through the Singapore holding structure to the technology’s Chinese origin and forced the deal unwound; the Mythos episode was the mirror image from the American side. A Singapore address shields no one from either capital’s reach. The hedge is the one prescribed for middle powers: specialise in the layers you can own — applied research, governance and certification, trusted deployment — while diversifying across both ecosystems and building sovereign capacity, from the ASPIRE supercomputers to the coming Kampong AI district.

The balance of the bet

For all these cautions, the honest verdict is optimistic. Few places pair Singapore’s depth of capital, quality of governance and institutional trust, and its certification regimes — AI Verify, the MAS frameworks — are becoming an export in their own right. Nor is it a passive host: through its sovereign investor GIC, Singapore holds equity in Anthropic and other AI firms, which aligns their success with the country’s and softens the risk that they simply pack up and leave.

So, is the influx good for Singapore? Overwhelmingly yes — but conditionally. The prize is not the logos going up in one-north. It is whether, five years from now, a Singaporean engineer trained inside one of these labs is running AI at a local bank; whether a homegrown firm builds on what diffused in; whether a university and a frontier lab deeply collaborate on AI-related innovations and applications. The giants will come for their own reasons. Singapore’s task — and, on current evidence, its likely achievement — is to ensure that what they leave behind is something that stays, and to identify trusted world-leading experts to lead this effort.

Related: Winning the AI race without an OpenAI | Cheap, fast and everywhere: Why China leads AI adoption