DeepSeek Disrupts AI Market with Low-Cost Training and Open Source, Yet Many Questions Loom

DeepSeek Disrupts AI Market with Low-Cost Training and Open Source, Yet Many Questions Loom

Analysts: Daniel Newman, Mitch Ashley, Oliver Blanchard, Krista Case, Keith Kirkpatrick, Fernando Montenegro, Nick Patience, and Ron Westfall
Publication Date: January 30th, 2025

What is Covered in this Article:

  • DeepSeek claims breakthroughs in model training efficiency, including some very low-level programming techniques. Questions have arisen about the potential use of competitors’ models to aid the development of R1.
  • The decreasing cost of deploying powerful AI models could democratize their integration into enterprise applications.
  • AI Infrastructure: DeepSeek’s latest AI model debut has significant implications for future investment in AI infrastructure, potentially altering decision-making priorities in the AI ecosystem.
  • Released under the open-source MIT license, DeepSeek is subject to significant industry scrutiny regarding its weights, openness, lower GPU resources, and training claims.
  • DeepSeek is a proof point for improving the efficiency of model training: Models that had to be trained in the cloud a year ago can already be trained closer to the edge, locally (on AI PCs), or with far fewer resources.
  • Contextualizing key cybersecurity issues that the rise of DeepSeek exposed.

The News: A China-based startup, DeepSeek, released its generative AI service using its open-source DeepSeek R1 large language model (LLM). DeepSeek competes with industry leaders like OpenAI ChatGPT, Anthropic Claude 3, Google Gemini, and Meta Llama 3. Purportedly developed using “reinforcement learning” and distillation techniques for $5.6M USD over a two-month period, DeepSeek claims R1 has achieved performance and reasoning similar to OpenAI’s o1 model. DeepSeek’s impact on the AI industry was profound, causing nearly a $1 trillion in stock value loss to the U.S. stock market. Debate across the AI industry continues, including skepticism about DeepSeek’s claims of LLM development cost, shorter development time, potential access to export-controlled GPU infrastructure, the value of open vs. proprietary, and the possible infringed use of competitors’ models for training.

DeepSeek Shocks the World, but Skepticism Should Be High

It is another case of reality being stranger than fiction. On Monday, January 27, 2025, the market erased trillions of dollars from AI equities as DeepSeek’s R1 had allegedly broken all scaling laws and was going to eliminate the need for AI data centers, incremental energy supply, advanced AI chips, increased networking bandwidth, as well as making most of the LLMs in development obsolete. NVIDIA alone lost more than $600 Billion in market cap in one day – the biggest single-day loss for any company in history. Many business leaders, media, pundits, and analysts were quick to laud China and DeepSeek for their ingenuity. DeepSeek claimed its white paper to be 100% accurate and that it could do for $6 million USD what the world’s most resourced and advanced technology companies were currently spending billions to achieve. It was the end of OpenAI, and NVIDIA’s bubble would have burst.

Yet skepticism should remain. While the industry has been aggressively pushing to bring down the cost of training and democratize AI, the claims that a handful of NVIDIA H800s are doing this and the claims of this being genuinely open-source are hard to believe. There is a lack of transparency on what infrastructure was used and whether or not techniques beyond time scale were used.

It now looks like DeepSeek had spent at least $500 million (if not more) on NVIDIA chips, which would violate trade restrictions, and that its researchers didn’t just use open-source distillation techniques but may have stolen IP from OpenAI and others to deliver R1.

This serves as a reminder that the US and China are in a new phase of a multi-decade power struggle and trade war. More than ever, it is happening in the technological sphere. Any technological claims coming out of China should be scrutinized as they are potentially leveraged as political or economic weapons.

Regardless of legitimacy, the past week’s events highlight that techniques can be implemented (such as time-scale) to improve inference and reduce training expenses through hardware and software innovation. Lowering training costs drives efficiency, innovation, model quality, and adoption. The industry is now reevaluating the AI arms race and who is poised to come out as winners.

How much AI infrastructure does the World Need?

Going forward, one of the biggest questions surrounds the demand for AI infrastructure, which has been the driving force behind the AI industry. The growing Capex spending across the hyperscalers (as reflected by Nvidia results over the past two years) and the announcement of Stargate are all examples of a heated arms race to build out this infrastructure. Even though many of DeepSeek’s claims are being scrutinized by the industry, there will undoubtedly be widespread reassessments of AI investment strategies related particularly to infrastructure needs. Investors are already scrutinizing those Capex investments made to date, and most hyperscalers plan to continue with their spending.

Conversely, ecosystem-wide GPU demands might shift downward. DeepSeek has questioned the notion of an on-premise AI industry that runs on (relatively) lower infrastructure needs. After all, proponents of this model highlight that this is where all the valuable enterprise data resides. In this scenario, high-performance storage, low-latency networking, robust data management and capabilities are still needed. This could benefit companies like IBM, Dell, HPE, Lenovo, Pure Storage, and other infrastructure providers. Dell has already highlighted that its customers can run DeepSeek AI on-premise.

If, however, the infrastructure claims that DeepSeek is making turn out to be completely false, then further questions will arise on the flow of advanced GPUs to the market. There are already claims by some in the industry that many AI labs in China, including DeepSeek, have thousands of advanced GPUs, including Nvidia H100s, violating export controls that are in place. This may lead to even stricter measures being put in place or further expanding restrictions as is already being reported.

Will DeepSeek Change The AI Game in Software?

DeepSeek’s claimed breakthroughs in model training efficiency, driven in part by constraints placed on China in the form of export controls for the most powerful chips, will concentrate the minds of AI researchers worldwide. But the potential that DeepSeek is operating with less compute power than some of its rivals doesn’t mean that more compute wouldn’t make models more powerful.

The ability to deploy and run powerful AI models at a far lower cost could change the calculus for deploying AI within enterprise applications. Although we have not seen this outside the freemium model, vendors may be able to embed generative AI into their platform without charging additional fees, thereby helping to spur usage and with less pressure to generate revenue specifically attributed to AI-enabled services.

A reduced cost structure will likely level the playing field between full-stack enterprise application vendors that own and operate compute resources versus those that need to acquire that compute from third-party vendors. As the cost to develop, train, and run models declines, the pace at which AI technology becomes an embedded and transparent capability increases.

Positive News for the AI Device Industry

Another view to bear in mind is that DeepSeek may not actually change the game but may instead merely shine a light on a larger transformative trend in the AI space—specifically, that of a confluence of improving efficiencies both in the models themselves and in AI-relevant processors. One way to understand this shift is that models that required significant data center/cloud resources to be trained a year ago have already become so much more efficient that they can now be run far more economically and on much smaller machines, sometimes even AI PCs. Meanwhile, semiconductor vendors, such as NVIDIA with its Blackwell releases, have improved hardware efficiency, further accelerating this trend.

Broadly, simultaneous improvements in efficiency in both hardware and software have already been driving the AI ecosystem towards a more federated “hybrid” model, where both training and inference workloads can be handled in the cloud, at the edge, or on devices, depending on each workloads compute, security, and velocity needs, as well as cost-to-outcome considerations. This shift from a predominantly Cloud-based AI ecosystem to a more federated (or “hybrid”) AI ecosystem slightly reframes the hardware, software, and services calculus for the industry. For instance, pre-2025 projections about data center spending relating to the hardware requirements needed for large-scale AI training and inference may have been over-indexed IF they failed to take into account the speed with which hardware and model efficiency are, together, already taking cost and resource utilization out of the overall ROI equation. Conversely, the importance of Edge AI and on-device AI segments of the market (i.e., mobile, PC, automotive processors, and small, budget-friendly on-premise hardware) may have been underindexed.

While DeepSeek’s disruption doesn’t appear to draw from hardware efficiency improvements, it shows how efficiency gains in model training can disrupt assumptions about future resource and cost requirements of LLM and LMM training.

DeepSeek may not be a resilient, long-term disruptor but rather merely the first company to shine a light on how this AI efficiency paradigm has changed the investment equation for companies looking to compete against incumbent AI vendors or develop alternatives to their products. If so, it may have opened the door to both a new wave of competition from startups looking to replicate DeepSeek’s success (albeit with more advanced chips) and a move by enterprises to train and develop their custom models, either for internal use or as service offerings for customers and partners.

Open Source Communities will Play a Significant Role

DeepSeek R1 is distributed as open-source (MIT license), causing the industry to reassess and reexamine open-source vs. closed or proprietary approaches. Although the weights for DeepSeek-R1 were open, the code and the training data were not, highlighting the overarching skepticism. Hugging Face (a prominent AI open-source community) has set itself the task of replicating R1 with Open R-1, while Meta engineers are also working on unpacking how R1 works. Meanwhile, Hugging Face notes that more than 600 forks of R1 have been created over the past few days.

Over the coming days and weeks, the industry will learn more about what is truly open source and transparent about DeepSeek and what is not. Ultimately, fully or partially open source may not hold the same weight placed on DeepSeek’s early claims. Regardless, R1 forks and investigations into DeepSeek have inspired AI tech titans and startups to consider the merits, competitive advantages, and threats of more open AI models, testing, and data strategies.

Security and Privacy, Once Again, Take Center Stage

Even before DeepSeek, the growing usage of AI apps and the underlying components, such as large language models (LLMs), has created many cybersecurity and data privacy concerns. AI governance is one of the key topics in many organizations.

DeepSeek’s sudden surge in interest and popularity has simultaneously made it a target of malicious actors – news items mentioned that DeepSeek was subjected to a large-scale cyberattack on Monday, January 27, 2025 (believed to be a distributed denial of service (DDOS) attack targeting DeepSeek’s API and web chat platform).

DeepSeek’s ties to China, broadly recognized as a perpetrator of state-sponsored cyber-attacks, raise additional concerns. These include misuse of sensitive data that may have been provided to DeepSeek either through prompting or additional data from RAG (retrieval-augmented generation), as well as uncertainty about how ideological and/or geopolitical considerations may influence the output from DeepSeek.

More broadly, the DeepSeek release touches on many areas of concern as cybersecurity teams ramp up their roles in AI governance. DeepSeek’s availability in multiple formats—mobile app, remote service, local model (via frameworks like Ollama), or cloud service (like AWS Bedrock)—raises the risk of unauthorized application use, insecure data sharing with SaaS providers, and the proliferation of shadow AI.

If DeepSeek did use training techniques based on existing OpenAI models, it could highlight abuse of terms of service and potential unauthorized exfiltration of data, which are other common topics of concern for those deploying AI services.

What to Watch:

  • During the current earnings season, attention will be on what Microsoft, Google, Amazon, Meta, and NVIDIA say about growth prospects and capital expenditure plans.
  • Irrespective of DeepSeek’s actual capabilities, SaaS vendors will continue to focus on how advanced AI functionalities can be integrated or embedded into their products at lower costs, enabling them to demonstrate greater value to customers.
  • Should the open-source DeepSeek model’s economics and energy efficiencies prove replicable, enterprises may accelerate the training of tailored models on-premises, accelerating hybrid AI implementations that increasingly distribute AI training and inference across the edge, including devices.
  • DeepSeek’s lower development costs and release under the MIT license could open the market to many more entrants and open-source go-to-market models.
  • More AI startups are following in the footsteps of DeepSeek and entering the market with equally disruptive solutions – albeit with access to more advanced and high-performance hardware.
  • Enterprises are working to replicate DeepSeek’s model, leveraging efficiency improvements in model training and processor capabilities to develop their in-house custom LLMs and LMMs.
  • We expect cybersecurity providers to highlight the concerns around DeepSeek as a key example of the kind of challenges that organizations must be ready to address. More tactically, those organizations deploying AI security posture management (AISPM) tooling or similar AI access control offerings will likely include more stringent controls around DeepSeek.

Advisory Notice: Multiple aspects surrounding DeepSeek are highly dynamic and can change quickly. This report reflects information available at the time of its writing. Futurum will continue to monitor developments and respond with our analysis as conditions warrant. Check back at https://futurumgroup.com/news-insights/ for the latest information on this and other important topics.

Disclosure: The Futurum Group is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.

Other Insights from Futurum:

2025 State of the Market – CEO AI Insights *subscribers

2H 2024 AI Software & Tools Market Sizing & Five-Year Forecast *subscribers

US Government Unveils $500B Stargate Alliance to Lead US AI Infrastructure Push

Do AI Foundation Model Specialists Have a Long-Term Future?

Will Google’s New Pricing Model Spur Usage of Gemini Leading to AI Revenue?

* Inquire About Futurum Intelligence Subscriptions here.

Author Information

Daniel is the CEO of The Futurum Group. Living his life at the intersection of people and technology, Daniel works with the world’s largest technology brands exploring Digital Transformation and how it is influencing the enterprise.

From the leading edge of AI to global technology policy, Daniel makes the connections between business, people and tech that are required for companies to benefit most from their technology investments. Daniel is a top 5 globally ranked industry analyst and his ideas are regularly cited or shared in television appearances by CNBC, Bloomberg, Wall Street Journal and hundreds of other sites around the world.

A 7x Best-Selling Author including his most recent book “Human/Machine.” Daniel is also a Forbes and MarketWatch (Dow Jones) contributor.

An MBA and Former Graduate Adjunct Faculty, Daniel is an Austin Texas transplant after 40 years in Chicago. His speaking takes him around the world each year as he shares his vision of the role technology will play in our future.

Mitch Ashley is VP and Practice Lead of DevOps and Application Development for The Futurum Group. Mitch has over 30+ years of experience as an entrepreneur, industry analyst, product development, and IT leader, with expertise in software engineering, cybersecurity, DevOps, DevSecOps, cloud, and AI. As an entrepreneur, CTO, CIO, and head of engineering, Mitch led the creation of award-winning cybersecurity products utilized in the private and public sectors, including the U.S. Department of Defense and all military branches. Mitch also led managed PKI services for broadband, Wi-Fi, IoT, energy management and 5G industries, product certification test labs, an online SaaS (93m transactions annually), and the development of video-on-demand and Internet cable services, and a national broadband network.

Mitch shares his experiences as an analyst, keynote and conference speaker, panelist, host, moderator, and expert interviewer discussing CIO/CTO leadership, product and software development, DevOps, DevSecOps, containerization, container orchestration, AI/ML/GenAI, platform engineering, SRE, and cybersecurity. He publishes his research on FuturumGroup.com and TechstrongResearch.com/resources. He hosts multiple award-winning video and podcast series, including DevOps Unbound, CISO Talk, and Techstrong Gang.

Research Director Olivier Blanchard covers edge semiconductors and intelligent AI-capable devices for Futurum. In addition to having co-authored several books about digital transformation and AI with Futurum Group CEO Daniel Newman, Blanchard brings considerable experience demystifying new and emerging technologies, advising clients on how best to future-proof their organizations, and helping maximize the positive impacts of technology disruption while mitigating their potentially negative effects. Follow his extended analysis on X and LinkedIn.

With a focus on data security, protection, and management, Krista has a particular focus on how these strategies play out in multi-cloud environments. She brings approximately 15 years of experience providing research and advisory services and creating thought leadership content. Her vantage point spans technology and vendor portfolio developments; customer buying behavior trends; and vendor ecosystems, go-to-market positioning, and business models. Her work has appeared in major publications including eWeek, TechTarget and The Register.

Prior to joining The Futurum Group, Krista led the data protection practice for Evaluator Group and the data center practice of analyst firm Technology Business Research. She also created articles, product analyses, and blogs on all things storage and data protection and management for analyst firm Storage Switzerland and led market intelligence initiatives for media company TechTarget.

Keith has over 25 years of experience in research, marketing, and consulting-based fields.

He has authored in-depth reports and market forecast studies covering artificial intelligence, biometrics, data analytics, robotics, high performance computing, and quantum computing, with a specific focus on the use of these technologies within large enterprise organizations and SMBs. He has also established strong working relationships with the international technology vendor community and is a frequent speaker at industry conferences and events.

In his career as a financial and technology journalist he has written for national and trade publications, including BusinessWeek, CNBC.com, Investment Dealers’ Digest, The Red Herring, The Communications of the ACM, and Mobile Computing & Communications, among others.

He is a member of the Association of Independent Information Professionals (AIIP).

Keith holds dual Bachelor of Arts degrees in Magazine Journalism and Sociology from Syracuse University.

Fernando Montenegro serves as the Vice President & Practice Lead for Cybersecurity at The Futurum Group. In this role, he leads the development and execution of the Cybersecurity research agenda, working closely with the team to drive the practice's growth. His research focuses on addressing critical topics in modern cybersecurity. These include the multifaceted role of AI in cybersecurity, strategies for managing an ever-expanding attack surface, and the evolution of cybersecurity architectures toward more platform-oriented solutions.

Before joining The Futurum Group, Fernando held senior industry analyst roles at Omdia, S&P Global, and 451 Research. His career also includes diverse roles in customer support, security, IT operations, professional services, and sales engineering. He has worked with pioneering Internet Service Providers, established security vendors, and startups across North and South America.

Fernando holds a Bachelor’s degree in Computer Science from Universidade Federal do Rio Grande do Sul in Brazil and various industry certifications. Although he is originally from Brazil, he has been based in Toronto, Canada, for many years.

Nick is VP and Practice Lead for AI at The Futurum Group. Nick is a thought leader on the development, deployment and adoption of AI - an area he has been researching for 25 years. Prior to Futurum, Nick was a Managing Analyst with S&P Global Market Intelligence, with responsibility for 451 Research’s coverage of Data, AI, Analytics, Information Security and Risk. Nick became part of S&P Global through its 2019 acquisition of 451 Research, a pioneering analyst firm Nick co-founded in 1999. He is a sought-after speaker and advisor, known for his expertise in the drivers of AI adoption, industry use cases, and the infrastructure behind its development and deployment. Nick also spent three years as a product marketing lead at Recommind (now part of OpenText), a machine learning-driven eDiscovery software company. Nick is based in London.

Ron is an experienced, customer-focused research expert and analyst, with over 20 years of experience in the digital and IT transformation markets, working with businesses to drive consistent revenue and sales growth.

He is a recognized authority at tracking the evolution of and identifying the key disruptive trends within the service enablement ecosystem, including a wide range of topics across software and services, infrastructure, 5G communications, Internet of Things (IoT), Artificial Intelligence (AI), analytics, security, cloud computing, revenue management, and regulatory issues.

Prior to his work with The Futurum Group, Ron worked with GlobalData Technology creating syndicated and custom research across a wide variety of technical fields. His work with Current Analysis focused on the broadband and service provider infrastructure markets.

Ron holds a Master of Arts in Public Policy from University of Nevada — Las Vegas and a Bachelor of Arts in political science/government from William and Mary.

SHARE:

Latest Insights:

Novin Kaihani from Intel joins Six Five hosts to discuss the transformative impact of Intel vPro on IT strategies, backed by real-world examples and comprehensive research from Forrester Consulting.
Messaging Growth and Cost Discipline Drive Twilio’s Q4 FY 2024 Profitability Gains
Keith Kirkpatrick highlights Twilio’s Q4 FY 2024 performance driven by messaging growth, AI innovation, and strong profitability gains.
Strong Demand From Webscale and Enterprise Segments Positions Cisco for Continued AI-Driven Growth
Ron Westfall, Research Director at The Futurum Group, shares insights on Cisco’s Q2 FY 2025 results, focusing on AI infrastructure growth, Splunk’s impact on security, and innovations like AI PODs and HyperFabric driving future opportunities.
Major Partnership Sees Databricks Offered as a First-Party Data Service; Aims to Modernize SAP Data Access and Accelerate AI Adoption Through Business Data Cloud
Nick Patience, AI Practice Lead at The Futurum Group, examines the strategic partnership between SAP and Databricks that combines SAP's enterprise data assets with Databricks' data platform capabilities through SAP Business Data Cloud, marking a significant shift in enterprise data accessibility and AI innovation.

Thank you, we received your request, a member of our team will be in contact with you.