Sovereign Australia AI has announced an ambitious plan to develop a local alternative to ChatGPT, aiming to create a culturally aware AI model for under $100 million. The initiative arises from concerns over cultural nuances and data security, particularly how American AI systems may misinterpret Australian expressions and comply with foreign data laws.
The company’s co-founder and CEO, Simon Kriss, highlighted the need for an AI that understands local vernacular, stating, “We wanted a model that understands the difference between a pair of togs and a pair of thongs, and knows that thongs aren’t underwear.” This project aims to address the cultural blind spots found in existing AI models, particularly for Australian users.
Infrastructure and Development Plans
To support this initiative, Sovereign Australia AI has ordered 256 Nvidia Blackwell B200 GPUs, marking the largest sovereign AI hardware deployment by an Australian company. These GPUs will be housed in NEXTDC’s Melbourne data centre. The company plans to develop two models: Ginan, a smaller research model to be open-sourced, and Australis, a flagship model with 700 billion parameters designed specifically for understanding Australian slang, legal frameworks, and cultural values.
Kriss emphasized the security implications of developing a local AI solution, particularly in light of the US Cloud Act. This act allows American companies to access data from servers worldwide without notification, potentially compromising Australian data privacy. “If it’s a US identity, you’ve got zero protection,” he stated, indicating the importance of a sovereign approach to AI development.
Competition and Market Landscape
Despite these efforts, OpenAI is not retreating. The company, which opened its first Australian office in March 2024, considers Australia one of its most significant markets. According to Oliver Jay, OpenAI’s managing director for international operations, user growth in Australia has increased by 2.5 times over the past year. OpenAI is aggressively hiring local talent and collaborating with major Australian companies such as CommBank and Canva.
Jay acknowledged the merits of local investment in AI capabilities, arguing that OpenAI’s advanced models could still meet Australian needs through localization. He noted, “Countries investing in their own AI capabilities is a good thing – it builds resilience and sovereignty.”
However, Kriss remains skeptical of this approach. He pointed to political pressures in the United States that could influence AI development, such as directives from former President Donald Trump regarding the content of AI models. “Is that what we want for Australia? We have our own values here,” he remarked, underscoring the importance of developing AI that aligns with Australian principles.
Challenges and Opportunities Ahead
Sovereign Australia AI has allocated $10 million to compensate copyright holders as part of their commitment to fostering local content. The company is also collaborating with UNSW and Deakin University to establish benchmarks for assessing the Australian-ness of their AI model.
Despite the enthusiasm for developing a homegrown AI, some industry leaders express caution. Mike Cannon-Brookes, co-founder of Atlassian, argues that Australia should focus on applying AI solutions rather than building foundational models. He believes that the country’s strengths lie in leveraging its abundant renewable energy resources for AI data centres and smart technology adoption across various sectors.
Technology expert Craig Dargusch, chief data officer at Cotality, echoed this sentiment, emphasizing that creating a purely Australian foundational model presents substantial challenges. “If we gathered every word ever written by Australians throughout history, it still wouldn’t come close to the data required to train a purely Australian foundational model,” he said.
Dargusch proposed that fine-tuning existing models for local context could be a more practical approach. Yet, he also recognized the potential for creating something uniquely Australian, integrating cultural narratives and values into AI.
As the AI landscape continues to evolve, Sovereign Australia AI is not alone in its efforts. Melbourne-based Maincode recently introduced its AI model, Matilda, at the SXSW Sydney event, though its CEO has distanced the company from sovereignty rhetoric.
With estimates suggesting AI could contribute $142 billion annually to Australia’s economy by 2030, the stakes are high. Kriss warned that if foundational AI models remain controlled offshore, Australia risks losing control over how its cultural values are represented in digital technologies.
The journey towards developing a sovereign AI model is fraught with challenges, but it also represents a significant opportunity for Australia to carve out its place in the global AI landscape.
