While the United States and China fight for control over the future of artificial intelligence, the United Arab Emirates is steadily expanding its own AI capabilities. In its latest achievement, the Gulf nation has made headlines with the release of an AI model known as K2 Think — an open-source model that can compete with models designed by companies like OpenAI and DeepSeek at a fraction of the size and cost.
What Is K2 Think?
K2 Think is a lightweight, open-source model developed by Emirati AI firm G42 and the Mohamed bin Zayed University of Artificial Intelligence. The model exhibits advanced reasoning, rivaling models built by AI companies like OpenAI and DeepSeek despite its smaller size and price tag.
The release of K2 Think reinforces the UAE as a legitimate player in the global AI landscape, and expands the AI race beyond the U.S. and China. However, concerns remain around the UAE’s motives for breaking into the AI industry and whether working with the country could pose national security risks to the United States.
What to Know About the UAE’s Latest AI Model
K2 Think is an AI model jointly developed by AI company G42 and the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), both of which are located in the United Arab Emirates. As an open-source model, K2 Think is available on Hugging Face under the Apache 2.0 license. However, what makes this model special is its ability to perform at a high level while leveraging a limited number of parameters and a lower price tag.
“K2 Think is a 32 billion parameter, open source reasoning model that punches well above its weight,” MBZUAI wrote in a post on X. “It’s built for advanced logic, math, and science reasoning, delivering frontier-class performance while being remarkably efficient.”
For reference, DeepSeek-V3 possesses 671 billion parameters, making K2 Think’s results all the more impressive. According to a paper published by researchers at MBZUAI, K2 Think holds its own against industry heavyweights like OpenAI’s o3, and outperforms DeepSeek-V3.1 and DeepSeek-R1 in math composite performance. This has its creators calling it the “world’s fastest open-source AI model,” which begs the question of how they pulled it off.
How Does K2 Think Work?
MBZUAI’s researchers list several techniques as essential to K2 Think’s advanced reasoning, particularly long chain-of-thought (CoT) supervised fine-tuning and test-time scaling.
CoT is a technique that large language models use to solve a complex problem by following a series of steps. In long CoT, LLMs break from this linear process, revisiting previous steps, correcting errors along the way and considering a range of possible solutions instead of just one. The researchers also used fine-tuning, which tailors a pre-trained model for a specific task, and supervised learning, where a model is trained on labeled data, to teach K2 Think tasks that require complex, step-by-step reasoning.
But the key ingredient behind K2 Think’s impressive reasoning capabilities is test-time scaling. This method provides additional computational resources during the AI inference phase, where an AI model makes predictions after analyzing unfamiliar data. Basically, test-time scaling gives an AI model more time to process or ‘think,’ resulting in richer insights, improved reasoning abilities and more accurate outputs.
Together, these aspects make K2 Think a lightweight model that can surpass flagship reasoning models up to 20 times larger in size, according to a press release from MBZUAI. It’s accomplishments like these that are propelling the UAE onto the global AI stage and putting the rest of the industry on notice.
How K2 Think Fits Into the UAE’s Plans for AI Relevance
The United Arab Emirates has been quietly launching lean, open-source LLMs for years, with K2 Think being the most recent example. MBZUAI has already played a major role in building Arabic LLM Jais, Hindi LLM Nanda and Kazakh LLM Sherkala. Even before the release of DeepSeek-R1, the UAE’s Technology Innovation Institute released Falcon 3, which rivaled competitor models like Meta’s Llama 3.1 and Mistral.
Inspired by DeepSeek and its disruptive impact on the AI landscape, the UAE is more determined than ever to tilt the scales of the AI competition in its favor. The country holds a unique advantage with its wealth of natural resources, which can be used to fuel the data centers and other AI infrastructure needed to train the next wave of LLMs. But the open-source nature of the UAE’s models suggests a more collaborative approach, rather than hoarding resources or vying to replace the U.S. and China as AI powers.
And the UAE might just get its way: Having witnessed the Gulf nation repeatedly produce smaller models that can compete with those made by major tech companies at a fraction of the cost, the U.S. and China are leaning more heavily on the UAE as a crucial partner in their push for AI supremacy.
What the UAE’s Strategy Means for the AI Race
President Donald Trump and U.S. tech leaders have prioritized strengthening relations with the United Arab Emirates. This past May, Trump reached a deal with the UAE that ramped up investment in the Gulf region to more than $2 trillion, including an agreement that will see both nations fund data centers in each other’s territories. This culminated in the announcement of Stargate UAE, a project spearheaded by OpenAI and G42 to construct a massive AI data center in Abu Dhabi. Additional partners include Oracle, Nvidia and Cisco.
Still, the U.S. tech titan with the strongest ties to the region may be Microsoft, which first established cloud data centers in the UAE back in 2019. In 2024, Microsoft invested $1.5 billion in G42, and it has since formed a partnership with G42 and MBZUAI to champion responsible AI efforts in the Middle East.
Although K2 Think follows in DeepSeek’s footsteps by challenging larger, more costly models, the UAE doesn’t seem intent on wresting power from AI leaders. Instead, these partnerships cement the country as a critical component of the U.S.’ AI ecosystem. By bringing more AI infrastructure and tech jobs to its backyard, the UAE is becoming a hub for AI innovation. And so, rather than try to defeat an AI power like the U.S., the UAE is making itself indispensable to its success — a move that could spell trouble for the Trump administration.
Why the UAE’s AI Ambitions Could Be a Problem for the U.S.
Cozying up to the UAE could create multiple conflicts for the U.S. government. On one hand, human rights advocates worry that viewing the United Arab Emirates as a vital tech partner may encourage the U.S. and other allies to overlook the country’s spotty human rights record. Especially if projects like Stargate UAE come to fruition, losing access to this kind of infrastructure over a human rights clash may be too great a sacrifice for the U.S. government to accept.
But the more pressing concern for U.S. lawmakers is the UAE’s relationship with China. According to the UAE’s Ministry of Foreign Affairs, the country is the largest incubator for Chinese businesses, as well as the largest trade partner with China in the Arab region. In 2024, the UAE embassy even hosted a think tank forum that celebrated bilateral relations with China and touched on economic progress, trade and technology, among other topics.
An entrenched Chinese presence in the UAE has U.S. officials worried that any technology sent to the Gulf country could end up in the hands of China. This was enough for former President Joe Biden to propose export restrictions on AI chips back in January. While the Trump administration ignored these initial warnings, national security fears have cast doubt on the U.S.-UAE deal central to the Stargate project, slowed down an Nvidia chip deal with the UAE and forced Microsoft to rethink its partnership with G42.
In an escalation of anti-China sentiment, U.S. politicians have also introduced a bipartisan bill known as the No Adversarial AI Act, which would ban U.S. executive agencies from using AI models built in China, including all DeepSeek models. If U.S. officials determine the UAE to be too much of a security risk as well, it’s possible that K2 Think and other models developed there could face a similar ban.
Either way, Trump seems set on transforming the Gulf region into the next frontier on the AI battlefield. The only question is whether helping countries like the United Arab Emirates will provide a much-needed boost to the U.S. in its AI race with China or come back to bite the Trump administration in the long run.
Frequently Asked Questions
What is K2 Think, and what makes it different from other AI models?
K2 Think is an open-source reasoning model built by AI firm G42 and the Mohamed bin Zayed University of Artificial Intelligence — both located in the UAE. With just 32 billion parameters, the model has rivaled the performance of much larger models produced by American and Chinese AI powerhouses. For comparison, DeepSeek V3 contains 671 billion parameters.
What role does K2 Think play in the UAE’s AI strategy?
K2 Think enhances the UAE’s lineup of impressive AI models. Combine this with the country’s rich natural resources that can support the infrastructure needed to train large language models, and the UAE becomes an attractive partner in the eyes of the U.S. In fact, the UAE has inked deals with the Trump administration and American tech companies, establishing it as an AI hub that’s central to the U.S. and its AI efforts.
Why are there concerns about the UAE’s role in AI?
Human rights advocates worry that forming closer ties with the UAE may lead the U.S. government to ignore the country’s poor human rights record for the sake of preserving trade agreements and access to AI infrastructure. The UAE’s relations with China also raise national security fears around U.S. technologies sent to the UAE ending up in the hands of China, threatening to undermine U.S.-UAE deals.
