DEEPSEEK PROMISES TO REDRAW THE PLAYBOOK: WHY REAL OPTIONS MATTER MORE THAN EVER

With Praveen Tanguturi, Ph.D. and Shalabh Kumar Singh

DeepSeek is challenging the current dominant industry paradigm of AI’s massive compute power requirements and high entry barriers. In what seems to be an inflection point in AI model development, DeepSeek has leveraged ingenious engineering to build upon existing open-source approaches, demonstrating that foundation model development can be significantly more compute-efficient.  

By rethinking model training from the ground up and making it available as open source like Linux operating system with estimated cost per token more than 90 percent lower than closed models, DeepSeek could potentially democratize knowledge, experience and insights, making AI more accessible across industries and, thereby, amplifying productivity, sparking innovation and enabling unprecedented scaling.  

DeepSeek’s impact extends beyond technical innovation. When viewed through the lens of real-options thinking—derived from financial stock options—DeepSeek’s advent enables investments in AI to be treated as flexible opportunities rather than fixed commitments, enabling decision-makers to adapt dynamically to changing conditions. This could be a game changer for enterprises. Rather than committing vast resources to a single AI model or architecture, organizations can now 

• Buy or license “ready-to-go” solutions for quick, cost-effective deployment, though with limited customization. 

• Boost or fine-tune existing LLMs with enterprise data to strike a balance between customization and efficiency. 

• Build proprietary generative AI models from the ground up for maximum control and specificity at affordable cost 

Technology selection in terms of real options can be viewed as a compound option (option on option), which considers the threat of new models superseding others, the variety of players and models in the market, and the experience of the end users. For example, the option to discover— which is to conduct feasibility study before adopting a specific model—can enable the option to expand with new investment in new markets or uses. Alternatively, companies can defer further investment (option to defer) at different stages if the results are not suitable and it works better to gather further information. If a pilot project turns out to be successful, companies can scale their investment and use case (option to scale). Finally, there is always the option to abandon when no other options are viable.  

Enterprises that do not have in-house talent and capabilities also have the option to contract but will need to manage the entire AI ecosystem. They will need to contract with service vendors who specialize in deploying and maintaining the entire AI portfolio and associated infrastructure.  

Here are some ways in which this could work in various industries: 

1. Technology and AI Development: AI development will become more inclusive, allowing companies of all sizes to innovate, accelerating applications across industries and domains like healthcare, life sciences and sustainability. DeepSeek’s efficient approach can enable new business and revenue models, including cheaper API access for AI-powered services. Growing interest in alternative AI architectures like Mixture-of-Experts (MoE) could drive more efficient compute. From a real-options perspective, organizations can now choose between models, start small, experiment and scale as they learn from their initial investments. 
2. Financial Services: AI-driven algorithmic trading strategies could improve, leading to smarter, faster financial models. Investors may shift capital away from high-cost AI infrastructure toward more cost-efficient AI development approaches. Financial firms can structure AI investments as “strategic bets” adjusting their exposure based on evolving market trends.  
3. Healthcare and Life Sciences: DeepSeek’s shift from broad, general-purpose AI to task-specific models could lead to specialized healthcare AI solutions, with further specialization tailored to solve specific problems in tasks across diagnostics, medical imaging and drug discovery. From a real-options perspective, organizations can test and validate solutions before committing to large-scale investments. 
4. Retail and e-Commerce: Retailers and e-commerce players could use DeepSeek’s AI for product recommendations and demand forecasting at a fraction of current costs. Small and mid-size companies can now afford AI-driven services, leveling the playing field. Large enterprises could integrate DeepSeek’s models into their platforms, making AI-powered services more cost-effective and widely accessible. Retailers can start with targeted investments and expand depending on ROI and success of adoption. 
5. Telecommunications and Internet Services: Telcos and ISPs may use DeepSeek’s AI to optimize 5G networks, manage traffic and automate network operations, reducing operational AI costs. Telcos can expand, defer or wait and watch for technological advancements that will allow for more suitable solutions. 

The benefits that we can expect from a new wave of AI approaches that Deepseek may spawn will also likely come with some market disruptions. Industrial automation and robotics may need to prepare for a world where new innovations can enable faster deployment of autonomous systems. Lower-cost AI-driven advertising and content creation could disrupt traditional media and ad-tech players. Open-source AI models could introduce new cybersecurity and risk management challenges, requiring stronger AI security frameworks. 

In this uncertain environment with various possibilities, framing it in options perspective allows businesses to remain agile and responsive to market conditions while mitigating risks. Leaders no longer have to gamble on AI—they can invest more strategically, exercising specific options or the option of options to build a portfolio of innovations and scale their investments as opportunities evolve.   

#AI #AIEthics #AIResearch #AIInnovation#AIInvestment #DeepLearning #AIApplications #MachineLearning #ArtificialIntelligence #GENAI #GenarativeAI #DeepSeek

(Thanks to Ramani Moses for editing this article)

(The views expressed in this article are solely those of the authors and do not necessarily reflect the opinions or positions of any affiliated organizations.)