By Lina Irawan 
Good morning. Nicholas Gordon here, reporting from Hong Kong, where the global tech sector is increasingly preoccupied with a concept largely overlooked in Silicon Valley and Beijing: "Sovereign AI." This emerging buzzword signifies a growing global apprehension among governments and corporations about becoming overly dependent on the artificial intelligence systems developed by either the United States or China, two dominant forces in the AI landscape. The concern is multifaceted, stemming from a desire to protect national interests, foster local innovation, and ensure technological autonomy.
The precise definition of "sovereignty" in the context of AI, however, is remarkably fluid, adapting to the specific geographical and geopolitical priorities of each region. In Europe, the focus is unequivocally on data sovereignty. The continent, with its stringent General Data Protection Regulation (GDPR) and a strong emphasis on privacy, seeks to keep personal information and sensitive national data within its borders. This is a direct response to concerns about foreign surveillance, commercial exploitation, and the legal complexities of data residing under the jurisdiction of other nations, particularly the U.S. and its CLOUD Act. Initiatives like GAIA-X underscore Europe’s ambition to build a secure and sovereign data infrastructure, reducing reliance on major U.S. cloud providers and ensuring that European data is processed and stored according to European values and laws.
For nations in the Middle East and much of Asia, the drive for sovereign AI is primarily an economic one, a calculated bet on cultivating a homegrown AI sector that promises substantial dividends. This approach aligns with broader national development strategies aimed at diversifying economies, creating high-skill jobs, and positioning themselves as technological leaders. Countries like Saudi Arabia, with its ambitious Vision 2030, are pouring vast sums into AI research and development through sovereign wealth funds, aiming to reduce dependence on oil and foster a knowledge-based economy. The belief is that by developing local AI capabilities, these nations can not only meet their unique societal and business needs but also become exporters of AI technology and expertise.
Meanwhile, for smaller economies worldwide, the pursuit of sovereign AI is driven by a more existential fear: the specter of technological isolation. They worry about being cut off from essential AI tools and services by foreign suppliers, whether due to geopolitical tensions, economic sanctions, or strategic decisions made by larger powers. This concern for autonomy is profound, as AI rapidly becomes an indispensable utility, akin to electricity or clean water. To be denied access to advanced AI could cripple a nation’s infrastructure, economy, and public services.
This sentiment was eloquently articulated by Pak-Sun Ting, the visionary founder of Hong Kong-based Votee AI. "AI has become such an essential need, and so you don’t want to be tethered to anybody else who can turn it off," Ting recently told me, highlighting the strategic vulnerability that unchecked reliance on foreign AI systems can create. His concern resonates deeply within a global landscape where technological dependencies are increasingly weaponized. Ting is set to share further insights at our upcoming Fortune Leaders Forum in Macau on September 8, where he will participate in a panel discussion on how businesses can navigate AI adoption and move beyond superficial engagement, or "tokenmaxxing."
Votee AI’s flagship product perfectly embodies the spirit of sovereign AI tailored to local needs. It’s an AI model specifically designed to operate in Cantonese, the vibrant Chinese dialect spoken by millions in Hong Kong and the surrounding Guangdong province. Ting underscored the critical void this fills: "The whole AI revolution is in English and Mandarin. Cantonese is used in education, health care, and police communications. If those don’t get covered, then AI is essentially useless." This statement powerfully illustrates the inherent linguistic bias in the current AI landscape, where models are predominantly trained on data from dominant languages, leaving a vast majority of the world’s linguistic diversity underserved and at risk of digital marginalization. Without localized AI, vital public services and cultural heritage in non-dominant languages face significant barriers to modernization and equitable access to technological advancements.
Votee AI is far from alone in this endeavor. It is part of a burgeoning global movement of developers, startup founders, and even major corporations dedicated to building AI models that cater to the "rest of the world"—languages and cultures often overlooked by the Silicon Valley and Beijing behemoths. For instance, Indosat, one of Indonesia’s largest telecommunications companies, is developing Sahabat AI, a model specifically focused on Indonesian languages like Bahasa Indonesia. Given Indonesia’s vast linguistic diversity, such an initiative is crucial for ensuring that AI serves all citizens, bridging potential digital divides and empowering local communities.
In South Korea, a nation with a highly advanced technological infrastructure, companies are actively participating in the "AI Squid Game," a government-sponsored competition designed to accelerate the development of the best homegrown AI model. This nationalistic approach aims to foster domestic talent, intellectual property, and technological self-sufficiency, positioning Korea as a leader in sovereign AI. Just yesterday, a significant development emerged from Saudi Arabia, where Humain, an AI company backed by the powerful Public Investment Fund (PIF), unveiled Humain M3. This frontier Arabic-language model, notably developed by the Chinese AI developer MiniMax, signals a strategic pivot by Saudi Arabia to collaborate with diverse global partners, moving beyond exclusive reliance on Western tech and bolstering its own regional AI capabilities for the vast Arabic-speaking world.
It is crucial to recognize that the languages targeted by these sovereign AI initiatives are by no means "small" in terms of speakers. Korean and Cantonese each boast approximately 80 million speakers, a population size comparable to many European nations. Bahasa Indonesia is spoken by over 200 million people. Yet, despite these significant speaker numbers, these are often categorized as "low-resource languages" in the AI world. This designation stems from the scarcity of large, high-quality digital text corpora available for training sophisticated AI models, unlike the immense datasets readily available for English and Mandarin Chinese. This lack of data presents a significant technical hurdle, making it more challenging and costly to develop robust and accurate LLMs for these languages.
However, the ambition of sovereign AI faces formidable challenges beyond data scarcity, primarily economic. The financial investment required is substantial: AI processors, particularly high-performance GPUs like Nvidia’s H100s, are incredibly expensive and in high demand globally. Establishing and maintaining the massive data centers needed to house and power these processors incurs astronomical operational costs. Moreover, the global competition for top-tier AI talent—data scientists, machine learning engineers, and researchers—drives up salaries, making it expensive to build and sustain a skilled workforce.
Pak-Sun Ting of Votee AI offered a pragmatic perspective on managing these costs. He pointed out that government entities, often the primary customers for sovereign AI models, do not necessarily require the most powerful, cutting-edge, general-purpose models. Their needs are often more specific and domain-restricted. This allows for "restrained ambitions," meaning models can be trained with smaller datasets and less computational power, significantly reducing expenditure without compromising utility for their intended applications. Ting revealed that Votee AI trained its Cantonese model for approximately $250,000. While not a trivial sum, this figure is dramatically lower than the tens of billions of dollars routinely invested by global AI giants like Anthropic and OpenAI in developing their foundational models. This demonstrates that strategic focus can lead to cost-effective solutions.
The path to AI sovereignty, therefore, appears to be less about achieving complete technological autarky and more about strategic diversification and intelligent integration. Countries and companies can source their AI models, semiconductors, and processing power from a variety of international suppliers, thereby mitigating the risk of over-reliance on any single nation or vendor. They can then add their own "local spin" – fine-tuning, cultural contextualization, and domain-specific knowledge – to the AI’s output to achieve superior, more relevant results for their specific populations and industries. A significant enabler of this strategy is the growing availability of powerful Chinese open-source models. These models, often developed with extensive government backing and catering to a massive domestic market, are increasingly being released for free download and use, offering a viable alternative to Western proprietary or open-source offerings. They provide a crucial resource for nations looking to build their own AI capabilities without incurring the immense development costs of building from scratch.
Ultimately, the evolving landscape suggests that "sovereignty" in AI might be less about absolute control and more about empowering strategic choice. It’s about building resilience through diversification, fostering local innovation through targeted investment, and ensuring that AI serves the unique linguistic, cultural, and economic needs of every nation, rather than being dictated by a select few.
CEO Daily will not publish on Monday to mark the U.S. holiday of Labor Day. We’ll be back on Tuesday.
Contact CEO Daily via Diane Brady at [email protected]
Top Leadership News
OpenAI launches GPT-6 Astra
OpenAI launched GPT-6 Astra, its newest and most advanced model, featuring groundbreaking "computer use" capabilities. This functionality is intended to allow the AI to control a computer interface much like a human operator, marking a significant leap in human-computer interaction. OpenAI co-founder Greg Brockman hailed this development as the nascent stage of artificial general intelligence (AGI), a still-theoretical form of AI designed to perform a wide array of intellectual tasks at or surpassing human cognitive levels. The implications for productivity and automation across industries are immense, potentially transforming how businesses operate and interact with digital tools.
Canva’s productivity push gains traction
Design powerhouse Canva announced impressive milestones in its expansion into workplace productivity tools. The Australian startup reports that users have created an astounding 4.5 billion presentations and 1.4 billion documents on its platform, demonstrating its growing influence beyond graphic design. This surge in adoption, particularly in Southeast Asia, is largely driven by users on mobile devices, indicating a successful strategy to capture the burgeoning mobile-first workforce. Canva’s move positions it as a direct competitor to established players like Microsoft and Google in the enterprise productivity software market.
Higher bond yields pose a midterm problem for Trump
Rising bond yields are exerting upward pressure on borrowing costs for both consumers and businesses across the U.S. This economic shift is creating significant headwinds for President Donald Trump as the midterm elections approach, placing increased strain on household budgets and corporate investment. With voters increasingly focused on affordability and the cost of living, the rise in the 10-year Treasury yield is a critical economic indicator that could sway public sentiment and political outcomes, adding to the challenges faced by the incumbent administration.
The Markets
S&P 500 futures are up 0.05% this morning, following a strong close of 1.06% in the last session. The STOXX Europe 600 was flat in early trading, reflecting a cautious sentiment across the continent. The U.K.’s FTSE 100 was down 0.04% in early trading. In Asia, the Nikkei 225 surged by 1.26%, while South Korea’s KOSPI saw a robust gain of 1.64%. China’s CSI 300 experienced a slight dip of 0.10%, contrasting with Hong Kong’s Hang Seng, which climbed by 1.74%. India’s NIFTY 50 posted a modest gain of 0.26%. Bitcoin is currently trading up at $81,000, continuing its strong performance.
Around the Watercooler
Crypto exchange Gate enters prediction market space with white-label builder for businesses by Camila Grigera Naó
Crypto exchange Gate is expanding its offerings by launching a white-label builder, allowing businesses to easily create and host their own prediction markets. This move positions Gate to capitalize on the growing interest in decentralized forecasting and betting platforms, offering a new avenue for engagement and revenue generation in the crypto space.
Google shipped four Gemini Flash models in 106 days. But its flagship frontier model is still nowhere to be seen by Wen Shao
While Google has rapidly deployed multiple "Flash" versions of its Gemini AI model, showcasing its agility in iterating on smaller, more efficient models, its much-anticipated flagship frontier model remains elusive. This highlights the intense competition and developmental challenges in building truly groundbreaking, large-scale AI systems, even for a tech giant like Google.
AI visionary Ray Kurzweil is joining a Silicon Valley startup developing a brain computer system that involves snorting nanoparticles by Alexei Oreskovic
Renowned futurist and AI visionary Ray Kurzweil is making headlines by joining a Silicon Valley startup focused on an ambitious brain-computer interface system that utilizes snorted nanoparticles. This intriguing and potentially revolutionary technology aims to create a direct neural link, pushing the boundaries of human-AI integration and challenging conventional notions of computing.
AI wants electricity now. The electric grid needs years to catch up by Mia Osmonbekov
The insatiable demand for electricity from burgeoning AI data centers is putting immense pressure on global electric grids, which are struggling to keep pace. This article explores the widening gap between AI’s immediate energy needs and the years required for infrastructure development, raising concerns about energy security, sustainability, and the future expansion of AI.
Clippers fined $30 million as owner Steve Ballmer is suspended for a year following NBA investigation on behind-the-curtain deal by Joshua Hong
The NBA has handed down a hefty $30 million fine and a one-year suspension to Los Angeles Clippers owner Steve Ballmer following an investigation into a "behind-the-curtain deal." This significant penalty underscores the league’s commitment to maintaining fair play and transparency, sending a strong message about accountability for team owners and their conduct.
CEO Daily is curated and edited by Joseph Abrams, Jason Ma, Claire Zillman, and Lee Clifford.
