AI
Small States, Big Choices: Singapore’s Approach to Sovereignty in the Age of AI
How Singapore redefines AI sovereignty for small states—not as self-reliance, but as a spectrum of strategic postures across the AI stack.
When the world’s largest AI summit wrapped up in New Delhi last week, it produced the expected pageantry: 88 nations signing the New Delhi Declaration, heads of state taking photographs with Silicon Valley CEOs, and the familiar rhetoric about “democratizing AI.” Yet beneath the declarations, a far more candid conversation was unfolding in the corridors of Bharat Mandapam. As the TIME magazine observed, delegates from “middle powers” wrestled with an uncomfortable truth: the overwhelming majority of global AI compute, data, and frontier talent remains concentrated in the United States and China. For most nations, the gap between aspiration and capability is not just wide—it is structurally embedded.
Singapore, a signatory to the New Delhi Declaration and one of the summit’s quietly influential voices, understands this gap better than most. A city-state of 5.9 million people with no natural resources and a land area smaller than Los Angeles, Singapore has no plausible path to AI autarky. And yet, in the weeks surrounding the New Delhi summit, it unveiled one of the world’s most coherent national AI strategies—not by racing to build the biggest models or hoard the most chips, but by adopting a carefully differentiated set of postures across each layer of the AI stack.
This distinction matters enormously. For small, open economies navigating the age of AI, Singapore’s approach offers a template that is both intellectually serious and practically executable.
The Autarky Trap: Why the Sovereignty Debate Is Asking the Wrong Question
The concept of AI sovereignty has a seductive simplicity to it. Who owns the data? Who trains the models? Who controls the compute? In the mainstream framing—visible in the rhetoric of both Washington and Beijing—sovereignty is essentially synonymous with dominance. The nation that leads in AI leads the world.
This framing works reasonably well as geopolitical shorthand for the United States, which commands extraordinary concentrations of frontier AI infrastructure, and for China, which has matched that ambition with state-directed industrial policy on a massive scale. The EU, for its part, has staked its claim on regulatory sovereignty—shaping AI governance through the AI Act in ways that larger markets can afford to enforce. But for the vast majority of nations—including nearly all of Southeast Asia, the Middle East, Africa, and Latin America—the “race for self-reliance” framing is not merely unrealistic. It is actively misleading.
AI sovereignty, properly understood, is not a destination. It is a capacity: the ability of a state to make meaningful choices about how AI is developed, deployed, and governed within its borders and in its name. That capacity does not require building everything from scratch. It requires building in the right places, partnering wisely in others, and maintaining enough institutional coherence to keep choices in domestic hands.
Singapore’s National AI Strategy 2.0 (NAIS 2.0), launched in 2023 and now mid-implementation, offers what may be the clearest articulation of this alternative model in the world. Rather than pretending to compete with hyperscalers on their own terms, Singapore has asked a more precise question: where across the AI stack must we build sovereign capacity, and where can we safely depend on trusted partners?
Singapore’s Layered Strategy: Sovereignty Across the AI Stack
Understanding Singapore’s approach requires examining the AI stack not as a monolith but as a series of distinct layers—each with its own strategic logic, its own risk profile, and its own implications for sovereignty.
| AI Stack Layer | Singapore’s Posture | Key Initiatives |
|---|---|---|
| Compute | Selective self-sufficiency + trusted partnerships | NAIRD Plan; GPU clusters at NUS/NTU; ECI cloud partnerships ($150M) |
| Data | Domestic control with cross-border access frameworks | Privacy-Enhancing Technologies (PETs) R&D; unlocking government data |
| Foundation Models | Strategic independence via niche capability | SEA-LION multilingual LLM; international model collaboration |
| Applications | Broad deployment across key sectors | National AI Missions in manufacturing, finance, healthcare, logistics |
| Governance | Global standard-setting leadership | AI Verify toolkit; Project Moonshot; US-Singapore Critical Tech Dialogue |
Compute: Selective Self-Sufficiency
Singapore is not trying to build a domestic semiconductor industry. That race belongs to Taiwan, South Korea, and increasingly the United States and China. What Singapore is doing is ensuring it maintains adequate sovereign compute capacity for research and government use—while securing deep partnerships with global cloud providers for everything else.
The S$1 billion National AI Research and Development (NAIRD) Plan, running from 2025 to 2030, includes dedicated GPU infrastructure operated for the Singapore research community. Alongside this, Computer Weekly reports that a $150 million Enterprise Compute Initiative facilitates SME access to cutting-edge cloud AI tools through trusted commercial partners. This is not autarky—it is calibrated dependency: maintaining sovereign research capacity while leveraging global infrastructure for commercial scale.
Prime Minister Lawrence Wong was direct about this posture in his Budget 2026 speech: “Our advantage does not lie in building the largest frontier models.” Singapore is instead focused on deploying AI faster and more coherently than larger countries—a form of competitive advantage that requires institutional strength rather than raw technological scale.
Data: Domestic Control, Global Connectivity
Data sovereignty is the layer where small states arguably have the most to gain and the most to lose. Singapore’s approach here is nuanced: it is investing heavily in Privacy-Enhancing Technologies (PETs) that allow data to be used for AI training without being exposed or transferred, while simultaneously advocating for trusted cross-border data flows as a global norm.
This dual posture reflects Singapore’s economic reality. As a financial, logistics, and biomedical hub, Singapore processes an extraordinary volume of sensitive data from across Asia and the world. Restricting data flows would damage its economic model. Failing to protect data sovereignty would expose it to the kind of dependency that compromises meaningful agency. PETs offer a potential third path—allowing participation in global AI ecosystems without surrendering control over the underlying information.
Models: Strategic Independence Through Niche Capability
Singapore is one of the few small states to have invested in developing its own large language model. The SEA-LION (South-East Asian Languages in One Network) model, developed through IMDA, addresses a critical gap: Southeast Asian languages are dramatically underrepresented in global foundation models trained primarily on English-language data. This is not merely a cultural concern—it has concrete consequences for healthcare AI, legal AI, and government services across the region.
SEA-LION represents a specific kind of sovereign capability: not competing with OpenAI or Google on frontier reasoning, but ensuring that AI applications serving Singapore and the broader region reflect local languages, contexts, and values. It is sovereignty by differentiation rather than by scale.
Applications: Depth Over Breadth
Budget 2026’s establishment of National AI Missions in four sectors—advanced manufacturing, connectivity and logistics, finance, and healthcare—signals a deliberate concentration of deployment effort. Rather than spreading AI adoption thinly across the entire economy, Singapore is betting on achieving genuine transformation in sectors where it has comparative advantage and where AI can address its most pressing structural challenges: a tight labour market and an ageing population.
The accompanying “Champions of AI” program offers enterprises 400% tax deductions on qualifying AI expenditures (capped at S$50,000, effective 2027–2028)—a fiscal instrument designed to lower the activation energy for SME adoption without distorting incentives toward vanity implementations.
Governance: The Most Underrated Layer of Sovereignty
Of all the layers, governance may be where Singapore’s sovereignty strategy is most original. The AI Verify testing framework and Project Moonshot—one of the world’s first LLM evaluation toolkits—represent Singapore’s bid to become a global standard-setter rather than a standard-taker in AI governance.
This matters strategically. Nations that can shape international AI norms wield influence disproportionate to their size. Singapore’s active participation in the Global Partnership on AI (GPAI), its US-Singapore Critical and Emerging Technology Dialogue, and its contributions to the UN High-Level Advisory Body on AI have established it as a trusted interlocutor across geopolitical divides—a position that larger powers, constrained by rivalry, cannot easily occupy.
The newly formed National AI Council, chaired by PM Wong himself and spanning six ministries plus private sector representatives, is designed to ensure that this whole-of-stack strategy is coordinated from the top. As Intracorp Asia noted: Singapore is aiming to make AI “a practical instrument of competitiveness, not a slogan.”
Comparative Lessons: Switzerland, Estonia, and the Limits of the Singapore Model
Singapore is not the only small state grappling intelligently with AI sovereignty. Switzerland has leveraged its neutrality and institutional quality to attract international AI governance bodies and frontier AI research (EPFL’s contributions to open-source AI are globally significant). Estonia, with its pioneering digital government infrastructure, has demonstrated that sovereignty in the application layer can be achieved independently of frontier model capabilities—its X-Road data exchange platform remains one of the most sophisticated sovereignty-preserving digital architectures in the world.
But Singapore’s approach has features that distinguish it from both. Unlike Switzerland, it is operating in a geopolitically contested neighborhood—ASEAN sits at the intersection of US-China strategic competition in ways that Europe does not. Unlike Estonia, it is an economic hub rather than a digital governance laboratory, which means its AI strategy must simultaneously serve commercial competitiveness, national security, and regional influence.
Singapore’s “balanced posture”—maintaining deep technology partnerships with American hyperscalers and defence partners while refusing to shut out Chinese technology firms entirely, and building Southeast Asian-specific capabilities that serve neither Washington nor Beijing’s AI agenda exclusively—is inherently fragile. It requires constant diplomatic management and a credibility that is earned, not inherited.
The risk, as geopolitical tensions intensify, is that this balance becomes harder to maintain. US export controls on advanced semiconductors, Chinese pressure on supply chains, and the broader de-globalization of AI infrastructure all create pressure on small states to pick sides. Singapore’s answer, at least for now, is to make itself too valuable as a neutral hub to be squeezed out entirely.
Economic and Geopolitical Implications: Agency Without Illusions
What does Singapore’s model mean in practice for its economic competitiveness and global influence?
On the economic side, the gains are potentially substantial. Singapore’s generative AI market is forecast to grow at over 46% annually through 2030, reaching US$5 billion. The NAIRD Plan’s investment in applied AI across nine priority sectors—from climate modelling to drug discovery—positions Singapore to capture high-value economic activities at the frontier of what AI can do. The AI Park at One-North, announced in Budget 2026, is designed as a physical ecosystem where startups, research institutions, and multinationals can co-develop applications—a model of deliberate clustering that Singapore has used successfully in biomedical sciences and fintech.
On the geopolitical side, Singapore’s influence will be felt most through standard-setting and norm entrepreneurship. If AI Verify and Project Moonshot achieve international adoption—particularly across ASEAN and the Global South, where governance capacity is weakest—Singapore will have shaped AI deployment practices for a significant portion of the world’s population. This is soft power of a meaningful kind: not projecting values through cultural influence, but building technical infrastructure that embeds particular governance choices.
The risks are real too. Concentration of AI infrastructure in the hands of a handful of global hyperscalers—most of them American—creates a form of dependency that no partnership agreement fully resolves. Singapore’s cloud compute partnerships come with terms of service, export compliance requirements, and geopolitical conditions that are ultimately set elsewhere. And the race to attract AI investment means competing with much larger jurisdictions—Saudi Arabia, the UAE, India—that can offer cheaper power, larger data markets, and, in some cases, fewer regulatory constraints.
Singapore’s edge in this competition is not scale; it is quality: of institutions, of rule of law, of talent density, and of the kind of trustworthiness that makes sensitive AI deployments in finance, healthcare, and government feel safe. That edge is real, but it requires constant investment to maintain.
Conclusion: Agency Over Autarky—A Model for the World
The New Delhi Declaration’s endorsement by 88 nations, including Singapore, reflects a genuine global desire for a different kind of AI future—one not defined purely by the strategic competition of the two superpowers. But declarations are not strategies. The gap between aspiring to AI sovereignty and achieving meaningful AI agency is where most nations will struggle.
Singapore’s approach suggests a more useful framework for small states confronting this challenge. The core insight is that sovereignty is not a binary condition—you either have it or you don’t—but a portfolio of strategic postures calibrated to each layer of the AI stack. You defend your sovereignty where the risks of dependency are highest (sensitive data, critical applications, governance norms). You embrace interdependence where the gains from collaboration outweigh the risks (frontier compute, foundation models, global research). And you invest relentlessly in the institutional quality that makes your choices credible to partners and rivals alike.
For policymakers in small and medium-sized economies—from Nairobi to Bogotá, from Tallinn to Kuala Lumpur—Singapore’s model offers not a blueprint to copy but a logic to adapt. The question is not whether your country can achieve AI self-sufficiency. It almost certainly cannot. The question is whether you have the institutional coherence, the diplomatic agility, and the strategic clarity to make AI work for you on your own terms.
That is what sovereignty actually requires. Not the biggest model. Not the most chips. But the wisdom to know which choices are yours to make, and the capacity to make them well.
Discover more from The Economy
Subscribe to get the latest posts sent to your email.
The atmosphere at the India AI Impact Summit in New Delhi this February 2026 made one reality unavoidably clear: the US China AI race is no longer a straightforward sprint to a singular finish line. Instead, we are witnessing the entrenchment of an asymmetric bipolarity. For global economists, corporate strategists, and policymakers, the AI competition US China has evolved from a theoretical technology battle into a grinding, multipolar war over supply chains, energy grids, and the economic allegiance of the Global South.
To understand the true stakes of US vs China AI supremacy, we must discard the simplistic, moralizing narratives of Cold War 2.0. As an analyst watching the tectonic plates of the global economy shift, the reality is far more nuanced. The question of AI leadership US China is not merely about who builds the smartest chatbot; it is about who controls the underlying thermodynamics of the future economy.
In this comprehensive analysis, we will demystify the geopolitics of AI race dynamics, cutting through the hype to examine the real-time tradeoffs, capital constraints, and data-driven realities defining 2026.
The Illusion of a Single Finish Line in the US China AI Race
Western media often frames the US China AI race as a zero-sum game of frontier models. However, Time’s recent February 2026 analysis correctly notes that there are, in fact, multiple overlapping races. While the United States continues to dominate closed-source, highly capitalized frontier models, China has pivoted toward a radically different theory of value: rapid, low-cost diffusion.
The AI competition US China shifted permanently with the “DeepSeek shock” and the subsequent surge of open-source models. When Alibaba released Qwen 2.5-Max—surpassing 1 billion downloads globally—it proved that Chinese developers could achieve near-parity with US models at a fraction of the computational cost. As CNN reported in February 2026, China’s AI industry is utilizing algorithmic efficiency to circumvent hardware limitations.
This dynamic explains the pragmatic, if politically fraught, decision in January 2026 to loosen US export controls on Nvidia H200 chips. The move was a stark acknowledgment of global interconnectedness: starving China of chips entirely risks accelerating their indigenous semiconductor ecosystem while severely denting the bottom lines of American tech champions. In the battle for US vs China AI supremacy, capital requires market access just as much as it requires compute.
Key Divergences in the AI Competition US China
- US Strategy (Innovation & Capital): High-end chips, hyperscale data centers, closed-source models (OpenAI, Anthropic), and massive capital concentration.
- Chinese Strategy (Diffusion & Application): Open-source models (DeepSeek, Qwen), industrial deployment, legacy chip scale, and aggressive pricing to capture emerging markets.
The Core Battlegrounds: Compute, Chips, and Energy Bottlenecks
You cannot discuss the geopolitics of AI race dynamics without discussing thermodynamics. Artificial intelligence is, fundamentally, electricity transformed into computation. Here, the US vs China AI supremacy narrative takes a politically incorrect but entirely substantiated turn.
The US undeniably leads in compute. According to the Federal Reserve’s late-2025 data, the US commands a staggering 74% global share of advanced compute capacity. Furthermore, as Reuters reported, US AI investments are projected to hit $700 billion in 2026. However, American capital advantages face a severe domestic bottleneck: regulatory holdups and grid limitations. Building a hyperscale data center in the US requires navigating localized zoning, environmental reviews, and grid interconnection queues that can take years.
Conversely, China’s state-controlled model enables faster scaling of physical infrastructure. While the Brookings Institution’s January 2026 report highlights the contrasting energy strategies, the raw numbers are sobering. By 2030, China is projected to have 400 GW of spare energy capacity, heavily subsidized by state directives (Bloomberg, Nov 2025).
The Asymmetric Matrix: US vs China Advantages
| Strategic Domain | United States Advantage | Chinese Advantage |
| Silicon & Compute | 74% global compute share; unmatched dominance in leading-edge architecture and design. | Overwhelming scale in legacy chip manufacturing; highly optimized algorithmic efficiency to bypass hardware bans. |
| Model Ecosystem | Dominates closed-source, reasoning-heavy frontier models (e.g., GPT-4o, Gemini). | Dominates lightweight, open-source models (DeepSeek R1, Qwen) tailored for global diffusion. |
| Energy & Grid | Massive private capital influx ($700B) for next-gen nuclear and SMRs, but hindered by grid regulations. | State-backed grid expansion; projecting 400 GW spare capacity by 2030 to power decentralized industrial AI. |
| Capital & Scaling | World’s deepest capital markets driving astronomical firm-level valuations. | State industrial policy suppressing tech valuations but rapidly building real, physical productive capacity. |
The Geopolitics of AI Race: Courting the Global South
The geopolitics of AI race extends far beyond Silicon Valley and Shenzhen. As highlighted at the New Delhi summit, the Global South is actively refusing to be relegated to mere consumers in the US China AI race.
For middle powers and developing economies, the AI leadership US China paradigm offers a stark choice. US closed-source models are highly capable but computationally expensive and heavily paywalled. In contrast, China is weaponizing open-source AI as a form of geopolitical diplomacy. By flooding the Global South with highly capable, free, or hyper-cheap models like Qwen and DeepSeek, Beijing is embedding its digital architecture into the foundational infrastructure of developing nations.
As Foreign Affairs noted in its February 2026 “The AI Divide” issue, this dynamic creates a new non-aligned movement. Countries like India, Saudi Arabia, and the UAE are hedging their bets. They purchase US hardware where possible but eagerly adopt Chinese open-source models to build “sovereign AI” capabilities. To win the geopolitics of AI race, the US cannot simply sanction its way to the top; it must offer a compelling, cost-effective alternative to Chinese digital infrastructure.
Capital Flow vs. Regulatory Bottlenecks: A Politically Incorrect Reality
To truly understand US vs China AI supremacy, we must look at how each system translates capital into productive capacity. A recent CSIS geoeconomics report provides a sobering multiperspective analysis: the US is optimized for a pathway dependent on high-end chips and continuous model scaling, heavily indexed to stock market expectations.
In the AI competition US China, America’s greatest strength—its free-market capital—is concurrently its Achilles’ heel. Trillions of dollars in market capitalization rely on the promise of Artificial General Intelligence (AGI) and sustained productivity gains. If regulatory holdups prevent the physical building of power plants to support this compute, the capital bubble risks deflating.
Meanwhile, China’s industrial policy suppresses firm-level valuations (to the detriment of its stock market) but excels at embedding AI into its leading industrial sectors, such as robotics and electric vehicles. As the Council on Foreign Relations (CFR) emphasized late last year, China’s approach guarantees that even if its frontier models lag by a few months, its factories will not. The US China AI race is therefore a test of whether America’s financialized innovation can outpace China’s state-directed diffusion.
The Path Forward: Redefining AI Leadership US China
The AI leadership US China debate is ultimately about resilience. The global supply chain is too interconnected to fully de-risk. America relies on TSMC in Taiwan, which relies on ASML in the Netherlands, to produce the chips that fuel the US China AI race.
For the United States to secure long-term AI leadership US China, it must transcend a purely defensive posture of export controls and tariffs. True US vs China AI supremacy will belong to the power that not only innovates at the frontier but scales those innovations globally. As Forbes analysts have routinely pointed out, democratic techno-alliances must move beyond rhetorical agreements and start co-investing in physical compute infrastructure, energy grids, and open-source ecosystems tailored for the Global South.
The AI competition US China will define the economic hierarchy of the 21st century. But victory will not be declared in a single moment of algorithmic breakthrough. It will be won in the trenches of grid interconnections, the boardrooms of middle powers, and the quiet diffusion of productivity across the global economy.
Next Steps for Democratic Alliances: To maintain relevance and leadership, Western coalitions must prioritize “compute diplomacy”—subsidizing energy-efficient AI infrastructure and accessible models for emerging markets, rather than ceding the open-source landscape entirely to Beijing. Would you like me to dive deeper into the specific policy frameworks the US could use to counter China’s open-source diplomacy in the Global South?
Discover more from The Economy
Subscribe to get the latest posts sent to your email.
Analysis
Are Anthropic’s AI Work Tools a Game-Changer? How Adaptable Plug-Ins Stack Up Against Bespoke Solutions for Lawyers and Consultants
On February 3, 2026, global markets witnessed what analysts are now calling the “SaaSpocalypse”—a single-day wipeout of approximately $285 billion in market value triggered by an unassuming GitHub release. Anthropic unveiled a legal plugin that helps customize its large language model Claude for legal tasks such as document review, sending public legal software stocks into a spin Legal IT Insider, with Thomson Reuters plummeting 16% and LegalZoom crashing 19.2% eWEEK. The culprit? A suite of open-source plugins that promised to democratize AI capabilities once locked behind expensive, specialized platforms.
The market’s violent reaction raises a fundamental question for knowledge workers: are Anthropic’s adaptable AI work tools genuinely game-changing, or do they represent yet another false dawn in the ongoing quest to automate professional judgment? For lawyers billing $800 per hour and consultants commanding similar premiums, the answer carries existential weight.
The Adaptability Offensive: What Anthropic Is Really Selling
Unlike previous AI tools that functioned as glorified chatbots, Claude Cowork can plan, execute and iterate through complex, multi-step workflows Legal IT Insider. Launched in January 2026, Cowork represents a philosophical shift from AI-as-assistant to AI-as-colleague—an autonomous agent capable of managing file systems, drafting documents, and executing specialized tasks without constant human supervision.
The real innovation lies in Anthropic’s plugin architecture. Skills are reusable instruction sets that teach Claude specific workflows, standards, and domain knowledge, such as brand style guidelines, email templates and task creation in tools like Jira and Asana Axios. By releasing 11 open-source plugins spanning legal, sales, marketing, and data analysis, Anthropic has essentially commoditized functionality that bespoke providers spent years—and billions in venture capital—developing.
For legal professionals, the implications are stark. The legal plugin can review documents, flag compliance risks, triage NDAs, and track regulatory changes—tasks that Harvey AI, the $11 billion legal tech darling, has built its entire business model around. The question becomes: why buy a tool that is no better than the legal plugin available from Anthropic? Artificial Lawyer
Yet beneath this seemingly straightforward value proposition lurks a more complex reality. Anthropic’s tools offer breadth; bespoke solutions promise depth. The distinction matters more than Silicon Valley’s venture capitalists—who’ve poured $300 million into Harvey AI in 2025 alone—would like to admit.
The Bespoke Advantage: When Specialization Still Matters
Harvey AI didn’t achieve 700 clients across 58 countries by accident. Top law firms and in-house legal teams trust Harvey to elevate their craft and navigate complexity Harvey, with two-thirds of Harvey customers reporting measurable benefits within 90 days, and nearly a third seeing impact within 30 days Legal IT Insider. The platform’s strength lies not in generic contract review—which Anthropic’s plugin handles adequately—but in highly customized workflows that integrate with a firm’s precedent database, understand jurisdiction-specific nuances, and learn from a decade of partner annotations.
Consider a scenario: A multinational law firm needs to review merger agreements under Delaware law while cross-referencing EU competition regulations and incorporating proprietary negotiation playbooks developed over 15 years. Anthropic’s legal plugin can identify standard risk factors. Harvey AI, custom-trained on the firm’s historical deals, can predict which specific clauses will trigger pushback from this particular opposing counsel based on patterns invisible to a general-purpose model.
The consulting world presents similar dynamics. McKinsey’s Lilli, which synthesizes over 100 years of proprietary knowledge across more than 100,000 documents and interviews Substack, doesn’t just answer questions—it embeds the firm’s institutional wisdom into every recommendation. Since its rollout in 2023, over 70% of McKinsey’s 45,000 employees utilize Lilli approximately 17 times per week, reportedly saving consultants up to 30% of their time Plus. BCG’s GENE and Deloitte’s Zora AI offer comparable advantages, each trained on decades of case studies, frameworks, and client engagements that no open-source plugin can replicate.
This specialization gap explains why Accenture is training approximately 30,000 professionals on Claude Accenture rather than simply handing them the plugins and calling it a day. Professional services firms understand that AI tools are multipliers, not replacements—and the multiplication factor depends entirely on what you’re multiplying.
The Productivity Promise: Data, Hype, and Reality
Anthropic’s market disruption rests on a seductive premise: why pay $10,000 per month for specialized legal AI when Claude’s $20 Pro subscription delivers 80% of the value? The economic logic is compelling—until you examine what “productivity gains” actually mean in white-collar professions.
Deloitte’s 2026 State of AI in the Enterprise report reveals that two-thirds (66%) of organizations are reporting productivity and efficiency gains from AI adoption Deloitte. Yet the same report shows that only 34% of companies are truly reimagining the business, while 74% hope to grow revenue through AI in the future compared to just 20% currently doing so Deloitte. The gap between efficiency and transformation remains stubbornly wide.
For knowledge workers, this distinction is critical. A junior associate using Anthropic’s legal plugin can draft a first-pass NDA 80% faster—but if that draft requires three rounds of senior partner revisions due to missing jurisdictional nuances, the net productivity gain approaches zero. As one McKinsey consultant shared: “My manager does not even ask me to do the task anymore. They just say ‘Get Lily to do it'” Merrative. The concern isn’t speed; it’s whether speed without judgment creates long-term value or simply faster mediocrity.
Research on AI’s cognitive effects supports this skepticism. A BCG study found that GenAI boosted performance on creative tasks but decreased performance on complex business problem-solving tasks by 23%, partly because consultants either over-trusted AI where it was weak or under-trusted it where it was strong Merrative. The risk of “prompt anxiety” giving way to “prompt dependency” looms large.
The Integration Crucible: Where Adaptability Meets Reality
Theory rarely survives first contact with enterprise IT infrastructure. Anthropic’s plugins may be open-source and “easy to customize,” but integrating them into workflows governed by compliance frameworks, legacy systems, and risk committees is anything but simple.
Compared to last year, more companies (42%) believe their strategy is highly prepared for AI adoption, but they feel less prepared in terms of infrastructure, data, risk, and talent Deloitte. The preparedness gap is widening, not narrowing. Perceptions of high preparedness have shifted down compared with last year for technical infrastructure (43%), data management (40%), and talent (20%) Deloitte.
Bespoke solutions offer a distinct advantage here: turnkey integration. Harvey AI’s partnership with Aderant delivers the industry’s first deeply connected ecosystem that unites AI-powered legal work with work-to-cash operations, bringing unprecedented transparency, accuracy, and productivity to both the front and back office Aderant. For law firms where time tracking, matter management, and billing are as critical as legal analysis, this integration isn’t a luxury—it’s table stakes.
Anthropic’s plugin architecture requires firms to build these bridges themselves. Plug-ins currently get saved locally to a user’s machine, although Anthropic says that an organization-wide sharing tool is on the way TechCrunch. Until then, enterprise deployment remains a DIY project requiring technical expertise that most legal departments and consulting practices lack.
Security concerns amplify these integration challenges. Anthropic’s own safety documentation for Cowork encourages users to monitor the agent closely and not grant unnecessary permissions, cautioning users to “be cautious about granting access to sensitive information like financial documents, credentials, or personal records” TechCrunch. Bespoke providers, by contrast, have spent years building enterprise-grade security frameworks that satisfy the most paranoid general counsels and CISOs.
The Economic Calculus: When “Good Enough” Isn’t
The cost differential between Anthropic’s plugins and bespoke solutions is dramatic. Claude Pro costs $20 monthly; Harvey AI runs into five figures for enterprise deployments. For solo practitioners and small firms, Anthropic’s offering is transformative. For Am Law 100 firms processing billions in transactions annually, the economics tell a different story.
Consider risk-adjusted value: A $50,000 annual Harvey AI subscription might seem extravagant compared to a $240 Claude Pro subscription—until a single missed compliance clause triggers a $5 million regulatory fine. A 2025 benchmark study found AI can be up to 80x faster than lawyers at document analysis and data extraction Grow Law, but speed without precision is professional malpractice dressed in silicon clothing.
The consulting market presents similar dynamics. BCG generated 20% of its $13.5 billion revenue ($2.7 billion) from AI-related advisory services in 2024, a revenue stream that didn’t exist two years ago Brainforge. These clients aren’t paying for generic AI capabilities—they’re paying for AI plus institutional knowledge, plus industry relationships, plus regulatory expertise. Anthropic’s plugins offer the first component; bespoke solutions deliver the package.
Moreover, the total cost of ownership extends beyond subscription fees. Customizing Anthropic’s plugins, training staff, managing version control, ensuring compliance, and troubleshooting failures all carry hidden costs that bespoke providers bundle into their pricing. For organizations with sophisticated AI maturity, building on Anthropic’s foundation makes sense. For those still navigating AI adoption—which includes 67% of finance leaders who are more optimistic about AI than last year, even as adoption has slowed Gartner—turnkey solutions remain attractive despite premium pricing.
The Skills Gap: The Real Bottleneck Isn’t Technology
Perhaps the most overlooked dimension of the adaptability-versus-specialization debate is human capital. The AI skills gap is seen as the biggest barrier to integration, and education—not role or workflow redesign—was the No. 1 way companies adjusted their talent strategies due to AI Deloitte. Anthropic’s plugins are only as valuable as the professionals wielding them.
Consulting firms are creating specialized AI teams: BCG’s 3,000-person BCG X division, Accenture’s plan to reach 80,000 data and AI professionals by 2026, representing the largest workforce transformation in consulting history Plus. These aren’t professionals learning to use ChatGPT—they’re hybrid talents who understand both domain expertise and AI architecture.
The skills divide creates a paradox: Anthropic’s tools are most valuable to organizations with sophisticated AI literacy, but those same organizations are precisely the ones with resources to build or buy bespoke solutions. Meanwhile, smaller firms and individual practitioners who would benefit most from democratized AI tools often lack the expertise to customize plugins effectively or the judgment to verify outputs.
This competency gap explains why McKinsey reports that 40% of its new projects now involve AI work Merrative, yet many clients remain in pilot purgatory. The bottleneck isn’t technology—it’s knowing what to ask, how to ask it, and whether the answer is correct. Bespoke solutions embed this expertise into their platforms; adaptable tools require users to bring their own.
The Regulatory Wild Card: When Compliance Meets Innovation
The market’s violent reaction to Anthropic’s plugins reflects not just economic displacement fears but regulatory uncertainty. Legal and financial services operate under scrutiny that makes “move fast and break things” a criminal liability rather than a business strategy.
Data privacy and security tops the list of AI risks companies worry about at 73%, followed by legal, intellectual property, and regulatory compliance (50%) Deloitte. These concerns aren’t hypothetical. Deloitte was asked to issue a partial refund for a $290,000 report prepared for the Australian government that contained AI-generated hallucinations Plus. When AI makes mistakes in regulated industries, the consequences extend far beyond embarrassment.
Bespoke providers have invested heavily in building compliant-by-design systems. Harvey AI’s deployment in CMS law firm’s expansion to over 7,000 lawyers demonstrates scalability within risk-managed frameworks The Global Legal Post. These platforms undergo legal review, security audits, and compliance certifications that generic AI tools can’t match.
Anthropic’s plugins, by contrast, place compliance responsibility squarely on users. For sophisticated organizations with robust risk functions, this arrangement is acceptable. For mid-sized firms without dedicated AI governance teams, it’s an existential risk. The choice between adaptable and bespoke often reduces to: who carries liability when something goes wrong?
Looking Forward: Convergence or Coexistence?
The binary framing—adaptable versus bespoke—is likely temporary. The more probable future features hybrid approaches where foundation models like Claude provide infrastructure while specialized layers add domain expertise.
Anthropic announced that Agent Skills is now an open standard making skills portable across different tools and platforms, which means skills people create in Claude can be used in models like ChatGPT or platforms like Cursor that adopt the standard Axios. This interoperability suggests a future where professionals move seamlessly between general-purpose and specialized tools, choosing the right instrument for each task.
Yet certain professional domains will remain resistant to pure commoditization. The craft of negotiating a complex M&A deal, advising on regulatory strategy, or designing organizational transformation involves judgment that transcends pattern recognition. As economists draw comparisons to the introduction of the spreadsheet in the 1980s or the browser in the 1990s FinancialContent, we should remember that those technologies eliminated certain jobs while creating entirely new categories of expertise.
The real game-change may not be Anthropic versus Harvey or McKinsey versus Claude, but rather the acceleration of knowledge work’s evolution from information processing to strategic judgment. Tools that enhance this evolution—whether adaptable or bespoke—will thrive. Those that merely automate yesterday’s workflows will join the wreckage of disrupted business models.
For now, the answer to whether Anthropic’s AI work tools are game-changing depends entirely on what game you’re playing. For legal secretaries doing routine document review, Claude’s $20 subscription is revolutionary. For M&A partners negotiating billion-dollar transactions, Harvey’s bespoke platform remains indispensable. For mid-market firms navigating between these extremes, the choice isn’t binary—it’s strategic, context-dependent, and likely to involve both.
The SaaSpocalypse of February 2026 wasn’t an ending. It was an opening salvo in a competition that will reshape how professionals work, what skills command premium compensation, and which organizations successfully navigate the transition from knowledge hoarding to knowledge orchestration. Anthropic’s adaptable plugins and bespoke solutions like Harvey AI aren’t mutually exclusive futures—they’re different tools for different hands, and knowing which to grasp may be the most valuable professional skill of all.
Discover more from The Economy
Subscribe to get the latest posts sent to your email.
Inside Orbbec’s 3D Vision Revolution: How Howard Huang’s Technology Powers China’s Humanoid Robot Boom and Reshapes Global Markets
BEIJING — On a sweltering August afternoon in 2025, a sleek humanoid robot named Tiangong Ultra streaked across a specially modified track at Beijing’s National Speed Skating Oval, completing the 100-meter sprint in 21.50 seconds. While nowhere near Usain Bolt’s record, the achievement marked something far more significant: the first fully autonomous humanoid to win a track event at the inaugural World Humanoid Robot Games. Behind the robot’s uncanny ability to navigate, perceive depth, and avoid obstacles lay a critical component most spectators overlooked—a pair of 3D vision cameras manufactured by a Shenzhen-based company called Orbbec.
Those robotic “eyes” have made their creator, 45-year-old engineer Howard Huang (Huang Yuanhao), a billionaire. As China’s artificial intelligence and robotics sectors explode, Orbbec has emerged as the invisible backbone of the nation’s humanoid robot revolution—and Huang’s personal fortune has soared alongside it. With Orbbec’s stock surging over 315% in the past year and reaching a market capitalization of approximately $5.2 billion, Forbes estimates Huang’s net worth at $1.4 billion, making him one of China’s newest tech tycoons.
Yet Huang’s journey from MIT-trained optical physicist to billionaire entrepreneur illuminates more than just a personal success story. It reveals how China is methodically constructing dominance in the robotics sector, building an ecosystem that spans from chip design to system integration, from academic research to commercial deployment. And at the center of this strategic push sits Orbbec’s technology—the depth-sensing cameras that give machines the spatial awareness they need to function in the real world.
The Rise of Robotic ‘Eyes’: From Academic Labs to Robot Marathon Winners
Howard Huang’s path to robotics royalty began not in boardrooms but in laboratories. After earning his PhD from City University of Hong Kong, specializing in optical measurement and laser speckle interferometry, Huang spent years as a postdoctoral researcher at the Singapore-MIT Alliance for Research and Technology (SMART) Center. His academic work produced over 40 published papers on 3D scanning technology—research that would later become the foundation for Orbbec’s proprietary depth-sensing systems.
In 2013, Huang founded Orbbec in Shenzhen with a singular vision: to “create 3D vision for a 3D world.” The timing proved prescient. Intel’s RealSense cameras had validated the market for depth-sensing technology, but Huang saw an opportunity to build something more comprehensive. Where competitors purchased components and assembled them, Orbbec developed its own chips from scratch—a vertically integrated approach that would later give the company crucial advantages in cost and customization.
“My motto is: ‘Climb the highest peaks in the best of times, and tackle the toughest challenges in the most cutting-edge industries,'” Huang told City University of Hong Kong in an August 2025 interview. “Achieving global influence in the robotics industry—that is the highest peak we aim to conquer.”
The company’s 3D vision cameras work by projecting structured light or using stereo vision to create detailed depth maps—enabling robots to perceive three-dimensional space much like human eyes do, but with millimeter-level precision. Orbbec’s RGBD (Red, Green, Blue, Depth) cameras combine traditional color imaging with depth sensing, giving robots the ability to identify objects, measure distances, and navigate complex environments in real-time. This technology has applications far beyond robotics: facial recognition for contactless payments, 3D scanning for industrial measurement, and autonomous navigation for warehouse logistics.
China’s Robot Craze: The Perfect Storm for Orbbec’s Ascent
Orbbec’s explosive growth coincides with—and directly fuels—China’s unprecedented push into humanoid robotics. The numbers tell a compelling story. According to the 2025 World Humanoid Robot Games, held in Beijing from August 15-17, 280 teams from 16 countries competed with over 500 humanoid robots across 26 events. Chinese robots dominated: Unitree Robotics swept medals in the 1,500-meter, 400-meter, and relay races, while Beijing’s Tiangong Ultra—powered by Orbbec cameras—won the 100-meter sprint and had earlier become the first humanoid to complete a half-marathon.
These aren’t just publicity stunts. The games revealed the real-world capabilities that Chinese manufacturers are racing to commercialize. Robots sorted medicine, handled logistics, cleaned hotel rooms, and performed industrial tasks—all applications where Orbbec’s depth-sensing technology provides critical advantages. Ant Group’s wheeled robot R1, unveiled in September 2025, uses advanced Orbbec 3D cameras. So does state-backed X-Humanoid’s flagship Tiangong 3.0 platform, which recently achieved “the first full-size humanoid robot capable of touch-interactive, high-dynamic whole-body control.”
The broader Chinese robotics market reflects this momentum. According to the International Federation of Robotics (IFR), China installed over 276,000 industrial robots in 2023—more than 51% of the global total. The nation’s robot density in manufacturing reached 392 robots per 10,000 employees, up from virtually zero two decades ago. Statista projects the global robotics market will reach $205 billion by 2030, with China commanding the lion’s share of production capacity and an increasingly large portion of innovation.
Shenzhen’s manufacturing ecosystem has proved particularly fertile ground for Orbbec. The city’s unmatched supply chains allow rapid prototyping and scaling—advantages that Western competitors struggle to match. When Intel controversially announced in 2021 it was “winding down” RealSense camera production (the decision was later reversed, but the division was eventually spun out in July 2025), Orbbec moved swiftly to capture market share, hiring former Intel RealSense sales leader Mike McSweeney as Vice President of Sales in 2024.
Outpacing Global Competitors: The Technology War Heats Up
In the high-stakes 3D vision market, Orbbec faces formidable competitors—yet has systematically carved out advantages. Intel’s RealSense, now operating independently after its July 2025 spin-out with $50 million in funding, remains a major player. The company’s new D555 camera, featuring 5 TOPS of onboard AI compute, targets industrial robotics with advanced edge processing. RealSense claims to work with 60% of AMR and humanoid developers, including high-profile clients like Agility Robotics and Geek+.
Yet Orbbec has competitive edges that extend beyond technology. Independent testing by OpenCV researchers comparing Orbbec’s Gemini 2 L against RealSense D455 found Orbbec cameras delivered lower temporal noise (0.15mm vs. 1.41mm) and more stable depth images in challenging conditions. While RealSense’s D455 performed better in certain scenarios involving strong light reflections, Orbbec’s newer Gemini 2 XL addressed many of these limitations.
More critically, Orbbec offers superior pricing and customization. Where RealSense cameras must serve a global market with standardized products, Orbbec can rapidly develop specialized versions for Chinese robotics companies. The company’s Gemini 435Le, priced competitively around $499, claims 30-50% better depth precision than comparable alternatives while maintaining the same price point. For cost-sensitive Chinese manufacturers racing to scale production, this value proposition proves compelling.
Other competitors include Luxonis (focusing on embedded AI cameras), Zivid (industrial 3D cameras), and various Chinese players. But none combine Orbbec’s vertically integrated manufacturing, local market access, and proven track record in humanoid robotics. According to market research firm Interact Analysis, Orbbec held a commanding 72% market share in South Korea’s mobile robot 3D vision market in 2024—a testament to its competitive strength even in export markets.
The Business Model: From Facial Recognition to Humanoid Robots
Orbbec went public on Shanghai’s STAR Market (ticker: 688322) in July 2022, raising 1.2 billion yuan ($170 million) in its IPO. The company’s subsequent financial performance has been volatile but increasingly positive. In the first half of 2025, Orbbec posted net profit of 30 million yuan ($4.1 million), reversing an 81 million yuan loss from the same period in 2024. Revenue more than doubled to 436 million yuan, with 62% still coming from facial recognition modules for retail and healthcare—not robotics.
This reveals both Orbbec’s current business reality and its future trajectory. The company’s bread-and-butter revenue comes from 3D cameras used in China’s ubiquitous contactless payment systems and medical insurance verification kiosks. Jack Ma’s Ant Group, which deployed Orbbec cameras in millions of Alipay terminals, remains a critical customer. But robotics represents the growth engine. In September 2025, Orbbec announced plans to raise up to 1.9 billion yuan ($262 million) through private placement, specifically targeting “AI vision and spatial perception technology” for robotics. Management projects robotics revenue will grow at a compound annual rate of approximately 100% over the next three to five years.
Huang, who holds a 27% stake in Orbbec, has seen his paper wealth fluctuate with the stock’s dramatic swings. Trading around 103 yuan per share in mid-February 2026 (up from lows of 41.53 yuan in early 2025), the stock remains highly volatile—reflecting both investor enthusiasm for China’s robot boom and uncertainty about when humanoid robots will achieve mass-market adoption. Analysts note that with a price-to-earnings ratio exceeding 395, Orbbec is priced for perfection, betting that robotics revenue will eventually dwarf the company’s current facial recognition business.
Global Implications: The New Technology Arms Race
Orbbec’s success sits at the intersection of several global trends that extend far beyond one company’s fortunes. First, it exemplifies China’s systematic approach to achieving technological self-sufficiency. Unlike previous generations of Chinese tech companies that relied on foreign components, Orbbec designs its own photosensitive chips, depth computation algorithms, and system architectures. This vertical integration insulates the company from supply chain disruptions and potential sanctions—a strategic priority as U.S.-China tech rivalry intensifies.
Second, the company’s growth underscores how China’s manufacturing advantages extend beyond low-cost labor to encompass entire industrial ecosystems. Shenzhen’s electronics supply chains, talent pools, and rapid iteration cycles allow companies like Orbbec to move from prototype to production at speeds unmatched in the West. When Orbbec needs a custom chip or modified optical system, local suppliers can deliver in weeks, not months.
Third, robotics raises profound questions about labor markets and economic disruption. According to IFR research, more than half of manufacturing operators will work alongside robots by 2034. In China, where youth unemployment already exceeds 15% in some regions, the robot revolution presents both opportunity and risk. Orbbec’s cameras enable automation that could displace millions of factory workers—even as they create new roles in robot maintenance, programming, and oversight.
Western policymakers are taking notice. The U.S. Department of Commerce has added various Chinese robotics and AI companies to export control lists, citing national security concerns. While Orbbec isn’t currently targeted, the company’s deep integration into China’s military-civil fusion strategy—Huang serves as a guest professor at Peking University’s robotics program—makes future restrictions plausible. For now, Orbbec sells globally: its cameras power robots in South Korea, Japan, and increasingly Europe, though the company has been cautious about expanding too aggressively into the U.S. market.
The Road Ahead: Challenges and Opportunities
For all Orbbec’s success, significant challenges loom. The humanoid robotics market remains largely speculative—full of impressive demos but limited commercial deployment outside controlled environments. Tesla, Figure AI, and other Western competitors are racing to develop general-purpose humanoids that could render specialized Chinese robots obsolete. Boston Dynamics’ Atlas, recently upgraded with electric actuators, demonstrates capabilities that still exceed most Chinese humanoids.
Orbbec also faces potential technical disruption. The rapid advance of end-to-end AI vision systems—where neural networks process raw camera feeds directly—could reduce demand for specialized depth cameras. NVIDIA’s latest robotics platforms, including the Jetson Thor module announced in 2025, increasingly handle depth perception through AI rather than hardware sensors. Orbbec is hedging by developing AI-enhanced cameras, but the company’s core value proposition could erode if pure computer vision proves sufficient.
Financial sustainability remains uncertain. Despite revenue growth, Orbbec reported negative free cash flow of -247 million yuan in the past 12 months as of late 2025. Heavy R&D spending—necessary to stay ahead of competitors—consumes much of the company’s revenue. The planned 1.9 billion yuan capital raise will provide runway, but investor patience may wear thin if robotics revenue doesn’t accelerate quickly.
Yet Huang remains bullish. At CES 2026 in January, Orbbec unveiled the ultra-compact Gemini 305 camera specifically designed for robotic arms, alongside announcements of full compatibility with NVIDIA’s Jetson Thor platform. The company is also expanding manufacturing capacity, establishing a dual-hub strategy across China and Vietnam to mitigate geopolitical risks. Partnerships with major chipmakers like MediaTek (an investor) and platform providers like NVIDIA suggest Orbbec is being woven into the broader robotics technology stack.
The Bigger Picture: What Orbbec’s Rise Tells Us
Howard Huang’s journey from MIT researcher to billionaire robotics magnate encapsulates a broader shift in global innovation leadership. China is no longer playing catch-up in advanced technologies—it’s increasingly setting the pace, particularly in fields like robotics where manufacturing prowess and scale advantages matter most.
Orbbec’s 3D cameras won’t make headlines the way flashy humanoid robots do. But they represent the unsexy infrastructure—the picks and shovels—of the robot revolution. Just as NVIDIA grew rich selling GPUs during the AI boom, Orbbec is positioned to profit regardless of which specific robot manufacturer wins the humanoid race. As long as robots need to see, Orbbec has a product to sell.
The World Humanoid Robot Games in Beijing may have seemed like spectacle—clumsy robots stumbling through sprints and football matches. But they previewed a future that’s arriving faster than most anticipated. A future where Chinese companies like Orbbec don’t just participate in global technology markets—they define them. Where a trained engineer’s vision for “3D vision in a 3D world” becomes the foundation of billion-dollar fortunes. And where the eyes of tomorrow’s robots, seeing the world with inhuman precision, bear the mark “Made in Shenzhen.”
For Western competitors and policymakers, Orbbec’s ascent poses an uncomfortable question: In the race to build the robots of tomorrow, who is really seeing clearly?
Key Sources and Further Reading
• Forbes: “How A Chinese Engineer Became A Billionaire Making Robotic ‘Eyes'” – Primary source on Huang’s net worth and Orbbec’s market performance
• Orbbec Official Website – Company technology specifications, product lines, and corporate announcements
• Global Times: “First World Humanoid Robot Games conclude” – Coverage of the Beijing robot games and Chinese robotics achievements
• The Robot Report: “After Intel exit, RealSense maps its own future” – Analysis of competitive landscape in 3D vision market
• OpenCV: “A Quick Comparison of the Orbbec and RealSense 3D Cameras” – Independent technical comparison of depth camera performance
• Statista: Global Robotics Market Forecast – Market size projections and industry growth data
• International Federation of Robotics (IFR) World Robotics 2024 Report – Authoritative data on global robot deployment and manufacturing trends
Discover more from The Economy
Subscribe to get the latest posts sent to your email.
-
Markets & Finance2 months agoTop 15 Stocks for Investment in 2026 in PSX: Your Complete Guide to Pakistan’s Best Investment Opportunities
-
Analysis2 weeks agoBrazil’s Rare Earth Race: US, EU, and China Compete for Critical Minerals as Tensions Rise
-
Investment1 month agoTop 10 Mutual Fund Managers in Pakistan for Investment in 2026: A Comprehensive Guide for Optimal Returns
-
Banks1 month agoBest Investments in Pakistan 2026: Top 10 Low-Price Shares and Long-Term Picks for the PSX
-
Asia2 months agoChina’s 50% Domestic Equipment Rule: The Semiconductor Mandate Reshaping Global Tech
-
Global Economy2 months agoWhat the U.S. Attack on Venezuela Could Mean for Oil and Canadian Crude Exports: The Economic Impact
-
Global Economy2 months agoPakistan’s Export Goldmine: 10 Game-Changing Markets Where Pakistani Businesses Are Winning Big in 2025
-
Global Economy2 months ago15 Most Lucrative Sectors for Investment in Pakistan: A 2025 Data-Driven Analysis