DeepSeek, a Chinese AI startup, is disrupting the AI landscape with its R1 open-source model that not only makes advanced AI technology accessible, but also demonstrates a unique approach to AI development, emphasizing performance, cost-effectiveness and transparency.
Performance: DeepSeek claims one of its standout features is its impressive performance metrics. The platform's latest model is said to rival some of the most advanced closed-source models in terms of speed and accuracy. This is a testament to the power of open-source development, where collective contributions can potentially lead to breakthroughs that individual entities might struggle to achieve on their own.
Cost Efficiency: Historically, the first unit of any new technological innovation is always prohibitively expensive. Consider the first ever computer invented compared to what it costs today. However, as the technology evolves and improvements are made, the overall costs decrease at a faster rate. This pattern has been consistently observed across industries, and if history is our guide, it is inevitable to expect the same trend with DeepSeek and AI. We will continue to see more cost-efficient LLMs entering the marketplace.
Accessibility: DeepSeek-R1 is accessible via its app and API. Its open-source model weights can be deployed on local or cloud GPU infrastructure, ensuring full control over security, data and operations. This open-source approach makes its technology freely available worldwide enabling developers, researchers, and enthusiasts to study, reuse and build upon their work, driving further advancements in the field. While this fosters innovation, it brings into question the safety and security of the platform.
Transparency: DeepSeek's architecture and reliance on reinforcement learning provides transparency not often seen in open-source models. Marked by its ability to "think out loud" and provide step-by-step real-time reasoning using test time compute (TTC), this approach lifts the veil of LLM explainability. And once the veil is lifted, the clarity it provides is transformative. You simply can't unsee it.
Competitive Dynamics: The introduction of DeepSeek has significant strategic implications for the global landscape. By offering high-performance AI models at lower costs, DeepSeek is not only challenging the major technology players but also redefining the competitive dynamics between established big tech and startups. This increased competition will drive innovation and expand partner ecosystems, leading to more efficient and affordable AI solutions.
Democratization of AI: DeepSeek’s arrival also underscores the critical role of expertise and innovation along with computational power. This development heralds a democratization of AI, making advanced models more accessible and fostering increased adoption and proliferation of AI. By lowering entry barriers, DeepSeek's emergence could cultivate a more inclusive AI ecosystem, benefiting both established entities and new entrants.
Efficiency: Moreover, a notable impact of DeepSeek's approach is the potential to achieve cutting-edge AI capabilities without the extensive computational resources. Efficient resource utilization, driven by innovative engineering and optimized training techniques, may prove more pivotal than sheer computing power. This paradigm shift will inspire a wave of innovation centered on cost-effective AI development, ultimately enhancing the return on investment (ROI).
At the same time, inferencing and test-time compute (TTC) will play an increasingly critical role in performance and responsiveness, as models like DeepSeek-R1 – and other advanced AI models that will emerge – prioritize deeper reasoning and explainability, driving higher and optimized inference compute requirements to generate thinking tokens.
The real winners will be the organizations that leverage these benefits and master integrating AI into everyday workflows by combining them with enterprise data, institutional knowledge, and trust. Organizations that leverage reasoning models like DeepSeek-R1, and others to come, will shape the future of enterprise AI.
AI efficiency gains, driven by approaches like DeepSeek, are set to transform demand dynamics. While efficiency gains may reduce the cost of individual computations, the Jevons paradox suggests that overall energy and infrastructure demands will likely rise due to increased AI adoption and expanding use cases. This means that any new compute capacity unlocked could be absorbed due to rising consumption, rather than impacting long-term investment trends.
We are at the dawn of a new AI era. While early reasoning models and reinforcement learning are promising, the journey towards advanced training, experiments, and sophisticated AI development demands more compute power. The energy, infrastructure, and technology landscapes in the U.S. will continuously need to supply this demand – through energy expansion or alternative sources, investment in AI infrastructure, technology innovation, or a mixture. No matter the path, expect substantial investments in these sectors.
With new AI entrants and innovations, there is the potential for regulatory reaction – resulting in, at least, short-term a continued/expanded divergence, yet with the recognition for the need for a more coordinated global regulatory approach. Risks and regulatory considerations of DeepSeek (and other AI tools) may vary depending on the use of app or API and whether open-source model weights can be deployed on local or cloud GPU infrastructures, and may include:
Data Privacy Concerns: Currently, data utilized in the development and deployment of AI falls under an array of varying data privacy laws and regulations (e.g., country-specific (e.g., China), EU GDPR, state-specific). Key AI and data privacy and security laws and regulations aim to put safeguards around how data is collected, accessed, used and retained. Many also include protections relative to sensitive consumer data and intellectual property. This complex regulatory environment necessitates strong AI data risk, governance and compliance measures. With rapid innovation, companies must adhere to current laws and regulations while also anticipating the potential for reactionary regulatory actions, including the potential for increases in data localization laws and regulations.
Cybersecurity and Resiliency: Quick expansion of AI competition and capabilities will increase the likelihood of cyberattacks, as well as uncover vulnerabilities in terms of resiliency and data security protocols. Similarly, the adoption of AI capabilities by an increasing number and variety of AI providers may also expose interconnected risks and vulnerabilities via third/nth parties. Events in this space will necessitate corporate actions (e.g., shut-downs, restrictions to access/registrations) and may also pressure lawmakers and regulators, particularly in areas deemed to be of critical security.
National Security Implications: DeepSeek's rapid ascent in the AI sector will expand the focus on national security threats (e.g., misuse by state actors, spread of malicious misinformation, frequency of cyberattacks). Companies should anticipate the potential for policy and regulatory shifts in terms of the export/import control restrictions of AI technology (e.g., chips) and the potential for more stringent actions against specific countries deemed to be of high(er) national security and/or competitive risk.
DeepSeek’s R1 model has opened the world to LLM explainability by providing both answers and the reasoning behind them. Future models will need to demonstrate their "thinking" process, showcasing how they arrive at conclusions, and engage in a form of meta-cognition, which involves self-reflection and awareness of their own reasoning steps.
The DeepSeek moment is a wake-up call for those who questioned AI’s long-term potential. It's clear that AI is not just a future promise – it's a present reality and this should drive all organizations to establish an enterprise AI strategy.
With advancements in AI technology and increased accessibility, it's critical for organizations to assess how their AI strategies align with their business objectives. Now organizations can more easily build their own models, and build-versus-buy along with the partner ecosystem strategy become essential.
Consider open-source as a trend that is here to stay – embrace it. Work with security and governance teams to establish a safe experimentation environment, preferably air-gapped to mitigate risks. Additionally, ensure that legal, risk, security and data privacy teams evaluate potential risks associated with open-source models and licensing terms & agreements for compliance.
You can download DeepSeek-R1 model weights and deploy them on GPU-enabled compute, whether a cloud hyperscaler, private GPU appliance, or locally (Note: While the R1 model weights are open-source, the training data used to create the model is not publicly available).
For experimentation, consider DeepSeek R1 Distill Llama 8Bor DeepSeek R1 Distill Llama 70B. Before deployment, conduct thorough security and integrity checks, including:
Establish, measure and maintain ROI goals and KPIs for your organization’s AI program. Continuously evaluate opportunities for refinement.
Efficient AI models directly support sustainability initiatives by lessening the environmental impact of AI through reduced energy consumption and decreased reliance on resource-intensive hardware. This aligns with organizational sustainability goals and enhances their environmental credentials.
Ensure your AI governance framework evaluates key components, including intended use, data reliability, privacy, security, and ethical risks. Perform red teaming tests for adversarial threats, vulnerabilities and data leakage. Increased accessibility and lower-cost solutions could lead to more use of publicly available AI tools by employees, including those that are restricted in companies. This makes robust oversight and compliance essential.
The statements contained within this article are based on publicly available information at the time of publication. They are subject to change as new information becomes available. The author(s) and the organization do not assume any responsibility for the accuracy or completeness of the information presented, and readers are encouraged to conduct their own research and verify any data or statements independently.
With extensive experience, advanced AI technologies and accelerators, and a powerful network of strategic alliances and investments, we can help you unlock the power of AI to shape and continuously evolve your future. Get in touch.
For media inquiries, contact Olivia Weiss (oweiss@kpmg.com) and Melanie Batley (mbatley@kpmg.com).
KPMG AI Quarterly Pulse Survey
As macroeconomic pressures, investor demands, and efficiency imperatives mount, business leaders are actively working to align their GenAI ambitions with practical execution, according to the latest KPMG AI Quarterly Pulse Survey.
2024 Tax Reimagined: Perspectives from the C-Suite
9 in 10 C-suite leaders recognize their tax department as a pivotal contributor to enhancing trust among diverse stakeholders, according to the latest KPMG report.
KPMG LLP Announces Strategic Investment in Agentic AI Startup Ema
KPMG announced a minority equity investment in Ema, an agentic AI startup building universal AI employees.