Since its launch, DeepSeek has captured the attention of businesses worldwide, promising a revolutionary shift in AI development with its cost-effective model for training and deployment. As it gains traction, enterprises are integrating DeepSeek’s technology, positioning it as a notable player alongside OpenAI. But what do industry experts think about its real-world potential?
To assess its strengths, risks, and long-term implications, we gathered insights from industry experts: Ivan Sysoev, Chief AI Officer at Quadcode, Sergey Sergyenko, CEO at Cybergizer, Pavel Podkorytov, co-founder at elDinero and AI Future Hub, and Anton Karbanovich, CEO & Founder at Efficien.cy, Flaik.ai, and INDEX.cy, to assess its strengths, risks, and future implications. Their perspectives offer a nuanced look at DeepSeek’s impact on the AI landscape.
Follow THE FUTURE on LinkedIn, Facebook, Instagram, X and Telegram
The Advantages: Why Businesses Are Interested
Cost-Effectiveness and Scalability
DeepSeek’s affordability is one of its most attractive features. Ivan Sysoev, highlights the cost savings of using quantized models, which allow for rapid testing without requiring expensive infrastructure.
“DeepSeek’s model suite is cheap, scalable, and useful—when you know how to use it properly. It’s a great tool for companies with limited budgets that need AI solutions, especially if they have the technical capacity to manage it. But without a strong understanding of the models and necessary infrastructure, inefficiencies can arise.”
Sysoev also notes that his team opted out of using DeepSeek’s online version due to security concerns, stating that for many companies, the existence of security risks outweighs the need to explore the exact nature of those risks. Instead, they focused on local deployment using quantized models, which provided fast and cost-effective testing without requiring high-end infrastructure.
Pavel Podkorytov underscores the financial benefits, pointing out that DeepSeek’s training costs are significantly lower than competitors—just $5.57 million compared to over $60 million for others—making it a viable option for cost-conscious businesses.
Performance and Innovation
DeepSeek excels in areas like code generation and logical reasoning. Sergey Sergyenko sees its emergence as a signal that the AI industry is evolving beyond OpenAI’s dominance:
“DeepSeek represents an exciting disruption in the AI space, but it’s not yet a complete alternative to platforms like OpenAI. It proves that the AI industry isn’t a monopoly and that new, adaptable solutions are emerging.”
Podkorytov praises its efficiency, noting the model’s 93% efficiency compared to the usual 30-50% in similar models, while Sysoev confirms that even its smaller quantized versions perform well for specific tasks.
Podkorytov also pointed out that multimodal versions, like Janus-Pro, surpass DALLE-3 in image generation, offering more flexible and powerful capabilities for creating visual content.
However, Sysoev cautions that DeepSeek’s models are not yet suitable for consumer-facing products without significant pre-processing and post-processing procedures. He warns that DeepSeek tends to generate answers very freely, which, if integrated “as is,” could pose reputational risks for businesses.
Open-Source Flexibility And Customization
DeepSeek’s open-source model offers businesses greater adaptability. Sergyenko highlights that companies can modify the model to suit their needs better than proprietary alternatives.
Podkorytov adds:
“DeepSeek’s MIT open-source license allows for commercial use without royalty fees, reducing vendor lock-in and fostering greater innovation.”
Sergyenko further emphasizes the strategic importance of avoiding dependency on a single AI provider, encouraging businesses to build an “AI-on-demand” model to future-proof operations.
Anton Karbanovich also points out the significance of DeepSeek’s open-source approach, stating that it enables smaller companies to compete by giving them access to advanced AI capabilities without the financial burden of proprietary models.
The Challenges: Risks And Limitations
Security And Data Privacy Concerns
While DeepSeek offers cost-effective AI solutions, it raises security concerns. Sysoev warns that businesses should be cautious about using DeepSeek online due to potential data privacy risks:
“For most of us, the existence of security concerns is more important than the investigation of the exact concern.”
Karbanovich echoes this sentiment, highlighting that open-source models, while transparent, also pose risks of misuse and potential vulnerabilities if not properly secured.
Infrastructure, Integration Challenges, And Hidden Costs
DeepSeek’s capabilities come with high infrastructure demands. Podkorytov points out that companies will need powerful GPUs, like the H200s, to achieve meaningful results. Additionally, FP8-optimized training requires the latest NVIDIA Hopper GPUs, further increasing hardware costs. The MLA architecture also demands custom CUDA cores, with fine-tuning costs reaching approximately $110K per domain.
Inconsistencies In Performance
Podkorytov notes that some DeepSeek models, such as R1-Zero, exhibit more hallucinations compared to GPT-4, with a 22% increase in incorrect responses in medical tests. Additionally, translation quality discrepancies, such as a 17% difference in Chinese-to-English translations, highlight areas needing improvement. Furthermore, the limited tool integration in DeepSeek models contrasts with the robust ecosystem offered by OpenAI, emphasizing the challenges in adapting to a more interconnected environment.
Open-Source Misuse Risks
Anton Karbanovich raises concerns about the ease with which bad actors could modify DeepSeek for unethical purposes, emphasizing the need for thorough security checks before deployment.
“Moving fast could mean skipping safety checks, and open models might be misused or hacked. Leaders should test it carefully, focus on security, and balance speed with safety.”
Market Uncertainty And Lack of Corporate Support
Sergyenko underscores the significant uncertainty surrounding DeepSeek’s long-term sustainability, emphasizing the potential geopolitical risks that loom large. These include the possibility of regional bans and growing scrutiny over technologies developed in China, both of which add multiple layers of complexity to the company’s prospects. He warns that while DeepSeek is a powerful disruptor, businesses should assess its long-term reliability before full-scale adoption.
Podkorytov highlighted the absence of corporate SLAs, emphasizing the risks for businesses reliant on consistent service standards. Meanwhile, he observed that competitors like Alibaba are beginning to adopt similar strategies, making the competitive landscape even more unpredictable.
MLOps And Scalability Challenges
Podkorytov highlights that DeepSeek’s MLOps tools lag behind industry leaders by 12-18 months, requiring additional engineering effort for large-scale deployments. However, compressed models like DeepSeek-R1-Distill can run on a single NVIDIA A100 GPU, offering an option for edge computing.
Final Takeaways: Industry Experts Weigh In
Ivan Sysoev:
“DeepSeek is a great budget-friendly AI tool, but companies must have the right technical expertise to avoid inefficiencies. It’s safe and fast when used locally with quantized models, but not private or secure enough as a direct OpenAI API replacement.”
Sergey Sergyenko:
“DeepSeek is an exciting disruptor, demonstrating that AI is not monopolized. However, he emphasizes that businesses must approach this innovation cautiously, considering its long-term sustainability and avoiding over-dependence on a single vendor. While DeepSeek may not replace OpenAI’s comprehensive suite of services today, its existence signals a future where owning your own AI — running models on-premise with your hardware and data — can offer significant advantages in terms of cost, control, and security. Adopting a hybrid strategy, combining cloud-based services with on-premise AI, could be the key to maintaining a competitive edge in this rapidly evolving industry.”
Pavel Podkorytov:
“For tech-savvy companies, DeepSeek delivers a major leap in efficiency. However, early adoption comes with the need for a budget to support integration and keep up with the fast-evolving open-source LLM landscape.”
Anton Karbanovich:
“Its open-source nature is both an asset and a risk. Businesses should focus on security and quality assurance before widespread adoption. DeepSeek proves that open-source AI can work at scale, but companies must approach it with caution.”
Conclusion
DeepSeek represents a compelling alternative in the AI landscape, offering cost-effective and customizable solutions. However, security, infrastructure demands, and market uncertainties remain key concerns. For businesses exploring AI adoption, a balanced approach—leveraging both proprietary and open-source solutions while maintaining flexibility—may be the best path forward.
