February 5, 2025
JAKARTA – The emergence of DeepSeek’s eponymously named chatbot significantly lowers barriers to AI adoption and could drive the technology’s use in emerging economies like Indonesia, analysts say, as its open-source large language model (LLM) reportedly offers good performance at a lower cost than currently available solutions.
They also caution that the country still needs substantial investment in high-performance graphics processing unit (GPU) infrastructure and local talent, however, to narrow the gap in AI research with powerhouses like China and the United States.
Founded in 2023, the Chinese artificial intelligence company is rapidly making waves in the AI sector.
Last December, it launched V3, a powerful LLM model that rivals OpenAI’s GPT-4o and Anthropic’s Claude 3.5 in performance but was reportedly trained at just a fraction of the vast sums its competitors spent.
DeepSeek followed up this year with the Jan. 20 release of its reasoning model R1, stirring up excitement among researchers for its capabilities said to be on par with OpenAI’s latest LLM offering, OpenAI o1.
Irzan Raditya, CEO of local conversational AI company Kata.ai, suggests that DeepSeek’s lower computing costs could open new doors for homegrown AI developers, as it gives them a chance to experiment, build prototypes and create AI solutions tailored to local needs.
But he cautions that AI technology and LLMs are evolving at a breakneck pace with major updates almost monthly, so instead of focusing on specific models, Irzan advises local start-ups to prioritize cases of real-world use.
“End users don’t ask what LLM you use, just like they don’t ask what cloud service you run on. They only care if the solution meets their needs,” he told The Jakarta Post on Thursday. “So, my advice is to bet on the interface or application layer.”
Sofian Hadiwijaya, chief technology officer at AI virtual chat developer AvataraLabs, echoed his sentiment, noting that the rise of DeepSeek had dramatically lowered barriers to AI adoption.
According to Sofian, OpenAI slashed its prices 80 percent compared to 2023, while Google is offering access to its Gemini API (application programming interface) at half OpenAI’s cost, while DeepSeek’s offering is just a quarter of OpenAI’s price.
“Based on our calculations, it costs just around Rp 30 (less than 1 US cent) per chat for our users,” Sofian told the Post on Friday.
“Nvidia CEO Jensen Huang once said he wanted AI to have zero computation costs, and we’re getting closer to that reality,” he added.
In response to DeepSeek’s aggressive expansion, OpenAI launched on Friday its own cost-efficient model, o3-mini, at just half the price of its o3 model.
Read also: How DeepSeek is shaking up US tech heavyweights
Dearth of AI infrastructure
Despite DeepSeek’s potential to help accelerate AI adoption in Indonesia, Sofian remains skeptical that the cost-efficient, open-source LLM can boost local AI research and development.
He highlighted the lack of high-performance GPU infrastructure, which limits researchers’ ability to train and refine AI models, as the country’s key challenge.
Sofian suggested the country to imitate Saudi Arabia’s investment in at least 3,000 Nvidia H100 chips at the King Abdullah University of Science and Technology.
These chips, or GPUs, are designed specifically for generative AI and cost around US$40,000 each.
Despite its cost-efficient approach, even DeepSeek still needed 2,048 Nvidia H800 chips to train its V3 model.
Its founder Liang Wenfeng revealed in a 2023 interview that DeepSeek had gradually amassed over 10,000 Nvidia GPUs, making it one of the largest holders of computing power among Chinese AI start-ups, South China Morning Post reported.
“Earlier this year, the communications and digital minister inaugurated the AI Center at Brawijaya University in Malang [East Java]. However, I noted that details on how many [AI-grade] GPUs are available at the facility and what types remain unclear,” Sofian added.
The communications ministry was not immediately available to comment on a request from the Post about those details.
Irzan of Kata.ai acknowledged that DeepSeek’s open-source LLM was a game changer for local AI development, but emphasized that alone would not close the gap with China and the US without significant investments in infrastructure, start-up funding and talent development.
“AI research requires sophisticated computing power. It has to be large-scale, affordable and fueled by clean energy,” he said. “This is where the government and industry consortia should step in to provide resources.”
Read also: DeepSeek and the path to Indonesia’s AI future
Roshan Raj, Southeast Asia partner at Redseer Strategy Consultants, said DeepSeek had democratized access to high-quality AI tools, and that this should serve as a wake-up call for emerging countries.
“It appears possible to achieve similar technological breakthroughs by only having high-quality tech teams [that] can work around capital and infrastructure constraints,” Raj told the Post on Friday.
“It gives greater credence to the ability of countries like Indonesia to compete in the AI landscape, versus what was thought possible before.”
To turn potential into reality, however, Raj noted that Indonesia must position itself as an attractive AI research and investment hub.
One suggestion would be to forge partnerships with countries more advanced in AI development, as well as investing in local tech talent, “including enticing talented Indonesians from abroad to come and contribute to local AI development”, he said.