![](https://i0.wp.com/e27.co/wp-content/uploads/2025/01/Deepseek_news.png?resize=690%2C388&ssl=1)
Across Southeast Asia, from Singapore’s fintech hubs to Indonesia’s e-commerce powerhouses, startup founders are evaluating DeepSeek’s promise of democratising advanced AI capabilities at a fraction of traditional costs.
With performance levels reportedly matching industry giants like GPT-4 and Claude-3.5, DeepSeek’s emergence marks a potential turning point for regional startups traditionally priced out of cutting-edge AI development.
But as regional tech leaders and investors weigh its implications, a crucial question emerges: Could this be the catalyst that propels Southeast Asia’s startup ecosystem into AI’s next frontier?
We spoke with several VCs, AI experts, founders and top executives at AI startups to learn how DeepSeek will impact the local startup ecosystem.
Below are the insights they shared:
Mauro Sauco, CTO of Transparently.ai, an accounting fraud detection firm
DeepSeek has generated buzz for a good reason. Its innovative approach to AI model development is already influencing how large language models are built and could fundamentally reshape our future in AI.
That said, I tend to be cautious when technology is surrounded by so much hype. While DeepSeek’s approach is innovative and impressive, I’m still on the fence about who will truly benefit from it in the long run.
Also Read: DeepSeek: The smart disruptor in the AI race
Here are some of my thoughts:
The impact on the ecosystem:
DeepSeek’s advancements have advanced the state of the art in LLM development. However, major frontier model developers are likely to quickly adopt these innovations, which might diminish DeepSeek’s lasting competitive edge.
Who stands to benefit:
Frontier model developers are likely to gain the most from incorporating this innovative approach to AI model development, which will lower costs and drive further innovation.
a) Large enterprises: Companies with specialised requirements might fine-tune DeepSeek to better suit their needs.
b) Regulated industries: Organisations that need to operate models within a controlled, private infrastructure could find value in DeepSeek.
c) Startups: For most startups, the direct benefits appear limited. However, any potential benefits might come indirectly through cost reductions passed down from the savings that mainstream frontier model providers eventually deliver.
Practical considerations for startups:
I’ve personally tested DeepSeek on metal (referring to the model itself, not the service) and had my team run some preliminary tests. For startups like ours, the practical benefits—especially in terms of cost savings and performance improvements—don’t seem substantial:
a) Infrastructure demands: Running DeepSeek locally (or on VMs) requires significant memory and GPU resources to achieve acceptable latency.
b) Production complexity: Setting up a production-grade system means managing redundancy, availability, and global distribution.
c) Operational costs: The overall costs and operational burdens can add up quickly.
d) Cloud provider offerings: Although some cloud providers are now offering DeepSeek via hosted endpoints, the advantages are minimal. DeepSeek R1 might be on par with, or only slightly superior to, existing reasoning models. Given that mainstream frontier models are likely to integrate these advancements in the near term, the effort and cost of refactoring software—including prompts and evaluations—to accommodate DeepSeek don’t seem justified.
Advantages of mainstream frontier models:
One major advantage of using established frontier model providers is their continuous improvement. These companies invest heavily in enhancing their models, meaning that as users, we benefit from ongoing enhancements with minimal effort on our part. DeepSeek might struggle to match this, especially when pitted against the giants in the field.
In conclusion, DeepSeek marks a significant advancement in the LLM ecosystem with its innovative approach to AI model development. While it’s clear that DeepSeek has reshaped the way we will approach AI model development in the future, its broader impact is still unfolding. For startups, in particular, sticking with established API services remains the more practical and cost-effective choice given the continuous improvements and financial backing of major frontier model providers.
Alvin Toh, co-founder of Straits Interactive, a data protection startup
DeepSeek’s affordability and accessibility democratise artificial intelligence (AI), making it attractive to ASEAN startups. Its low-cost, high-performing models enable developers in companies to rapidly prototype and integrate it into their technology stack and allow startups to innovate in various fields without prohibitive infrastructural investments.
However, there are concerns surrounding DeepSeek’s higher level of hallucination compared to other models in the market (notably, in following rules, writing, creativity, and persuasiveness) and certain biases from China in its outputs.
Additionally, DeepSeek’s ambiguous privacy policies and lack of robust compliance certifications bring pause. Its privacy policy lacks clarity on data usage, causing unease about data leakage and misuse. There is also no explicit mention of adherence to key international standards like GDPR, making it risky for startups operating in regulated industries.
Moreover, recent attempts by security researchers at Cisco and the University of Pennsylvania to jailbreak Deepseek’s model with adversarial prompts revealed that it failed to block all 50 attempts, indicating that there’s still a security gap in deployment.
Startups handling regulated, sensitive, or proprietary data must, therefore, carefully evaluate DeepSeek’s scalability and adherence to privacy laws, policies, and service-level agreements (SLAs). Compliance with local regulations, like the PDPA for Singapore-based outfits, needs to be ensured before integrating DeepSeek into critical workflows.
Until DeepSeek provides stronger assurances of data privacy and compliance, startups should presently avoid using it with sensitive data and consider alternatives with a privacy-first approach.
Rei Murakami, Venture Partner at Kadan Capital
DeepSeek has delivered a few important AI milestones: In December, version 3 was released with Mixture-of-Experts architecture, faster inference, and longer context windows, followed by the R1 model in January. But let’s be clear: the real story isn’t about one company.
LLMs are on a path to commoditisation, with open source gaining momentum. I believe the recent negative public market reaction is misplaced. In the mid-term, AI adoption is about to dramatically increase. History proves it; back in 1943, IBM’s Thomas Watson infamously predicted a world market for not more than five computers. When costs decreased, demand skyrocketed. The same is now happening with AI.
Also Read: DeepSeeking the future: The ripple effect on tech, crypto, and global markets
Nowhere is this shift more significant than in Southeast Asia. In emerging economies, SaaS solutions struggled with monetisation due to low labour costs. But with token costs set to drop, autonomous agents will become viable even in price-sensitive markets. The AI revolution isn’t slowing—it’s just getting started.
We continue to be excited about application layer AI, particularly in vertical SaaS, where AI can unlock entirely new business models and drive real operational impact.
Dr Sze Tiam Lin, Senior Licensing Advisor, SMU Institute of Innovation & Entrepreneurship
DeepSeek enables startups in Southeast Asia to harness cutting-edge generative AI technologies without the need for massive budgets and push the boundaries of what’s possible, offering a more dynamic and competitive landscape for innovation. It also allows smaller players to adopt efficient inference systems and benefit from significantly reduced training costs, making it easier to develop sophisticated AI solutions.
DeepSeek stands out in its efficient pre-training that accelerates the learning process and shortens the time required to deploy customised and powerful AI models. With performance levels on par with the best versions of GPT-4 and Claude-3.5, startups can harness high-level capabilities at a fraction of the cost.
Additionally, DeepSeek’s unique voting technique offers self-feedback on open-ended questions, enhancing the effectiveness and robustness of the alignment process, a critical factor in refining AI systems.
The pace of adoption across Southeast Asia will depend on the governance frameworks in place regarding the use of open-source models and data privacy in commercial deployments. Different jurisdictions may have varying regulatory environments, and this could affect innovation and AI adoption.
Matt Spriegel, CEO and founder of Atiom, an AI-powered gamified platform
Southeast Asia has been a hub for rapid digital adoption, and DeepSeek is lowering the barrier for startups to leverage AI early in their journey. From customer service automation to deep-learning analytics, its entry democratises access to AI, making advanced technology more accessible to early-stage companies.
Startups will have more access to AI without heavy R&D costs, which will also accelerate innovation. In particular, AI-first startups, especially in fintech, e-commerce and healthtech will likely experience accelerated growth.
Scalable AI solutions will also help startups to compete globally. This will impact the ecosystem, which will experience a surge in demand for more AI engineers, data scientists and machine learning (ML) specialists.
Furthermore, DeepSeek’s launch signals a shift in AI development beyond Western dominance. As more models are built and trained locally, we’ll see AI solutions tailored to Southeast Asia’s unique market needs, driving industry-specific innovation across the region.
Simon Davis, founder and CEO of GOAT Gaming, an AI-powered network of games on Telegram
DeepSeek’s emergence represents a transformative moment for Southeast Asia’s technology landscape, fundamentally altering the economics of AI deployment in the region. By dramatically reducing infrastructure costs for AI implementation, it levels the playing field for smaller companies that previously couldn’t compete due to resource constraints.
The ability to run these models locally is particularly significant for Southeast Asian startups. Not only does it slash operational costs, but it also addresses crucial data sovereignty concerns that have historically complicated AI adoption in the region. Companies can now process sensitive data on their own servers, ensuring compliance with local privacy regulations without compromising AI capabilities.
This cost-effective, local-first approach opens up exciting possibilities for market-specific AI solutions. In Indonesia, for instance, companies can now realistically develop specialised models tailored to the unique cultural nuances of the market. This localisation potential extends across Southeast Asia’s diverse markets, where one-size-fits-all solutions often fall short.
The implications are profound. We might see a surge of innovative AI applications emerging from previously underserved markets. This democratisation of AI technology could catalyse a new wave of regional innovation, powered by companies that understand their local markets intimately.
Nitin Vyas, Sr. VP (Product & Data) at RedDoorz, a budget hotel network
As a hospitality technology company, we view the emergence of DeepSeek as part of the industry’s ongoing evolution and also making AI a level-playing field for everyone.
DeepSeek, as an entity focused on AI and advanced technology solutions, represents an opportunity to enhance operational efficiency, personalise services, and elevate customer experiences in the hospitality sector. A healthy competition among the AI giants will be beneficial for this industry as it gives a more balanced global view of how AI will shape the course of our civilisation.
Also Read: AI, personalisation, and 5 marketing activities you should be doing
This competition may enable corporations to implement AI more cost-effectively. At the same time, we could see an even faster pace of innovation and more advanced capabilities in the large language model (LLM).
Remi Choong, Elev8.vc, a VC firm
Cost-effective AI infrastructure: Increasingly cost-effective models, such as DeepSeek’s, will lower API costs. This enables AI startups to deliver cheaper and better solutions, similar to the disruptions we’ve seen in the solar panel industry.
Stronger demand for AI: With more affordable options, global AI adoption and spending will likely surge. Southeast Asian startups must quickly adapt to agile go-to-market strategies to capture the growing demand, particularly among SMEs.
Competition: Given lower AI costs, it is increasingly important for companies to adopt AI to maintain a competitive advantage. We can also expect to see an uptick in adoption across a broad spectrum of industries.
Balancing AI and hard tech: Investors need to ensure a balance between funding cutting-edge AI ventures and the infrastructure that supports them, especially in countries like Singapore with strong hard tech capabilities.
Jeff Pan, Belli.ai, which builds air cargo software for airlines
Speaking broadly, it’s unlikely that you’ll see the same velocity of change in Southeast Asia as you do in the EU/US, primarily due to the types of problems that we are solving.
Founders in the EU/US typically deal with high-capability customers who already have high internal capabilities and large IT budgets. In contrast, founders in Southeast Asia typically deal with early digitisation problems (helping low-capability customers with small IT budgets transition from paper and spreadsheets to basic CRUD applications), which will largely be unaffected by the cutting-edge progress you see being made by DeepSeek.
Much of the impact you will see (which is already happening) is that teams can generate 10x more impact with smaller teams, which you will see play out in (a) less hiring demands from startups, who don’t need as much headcount and (b) VC funds skewing away more towards top 10 per cent founders who can seize on these advantages rather than a broader base of portfolio companies.
Nofi Bayu Darmawan, founder and CEO of Komerce, an e-commerce enabler
The promise of AI in simplifying tasks and creating value, especially in the context of cost-effectiveness is huge.
When it comes to e-commerce, the impact of generative AI on customer engagement, especially its role in social media interaction, e-commerce, and automating customer service, is massive. DeepSeek can consolidate brand knowledge and communication across various platforms, including Google reviews and online chat on websites, enhancing the commerce ecosystem.
Warren Leow, CEO of Designs.ai
Deepseek has sparked a lot of interest because its progress has lowered service costs. This spurs more innovation and adoption among users as barriers to adoption are reduced.
In addition, having more choices across different vendors and technology providers, especially from open source, will only spur further improvements in the quality of applications being produced.
The post The DeepSeek debate: Opportunity or overhype for startups in ASEAN? appeared first on e27.