The AI landscape has been shaken by the emergence of DeepSeek, a Chinese AI startup that has rapidly risen to prominence. Its latest large language model (LLM), the DeepSeek R1, is not only performing on par with, or even surpassing, models from industry giants like OpenAI, but it's doing so on a fraction of the budget, using H800 GPUs instead of the latest H200 GPUs from Nvidia. This has caused a stir in the market, with global technology stocks taking a hit as investors digest the implications. DeepSeek's success highlights a critical shift in the AI industry: the real value isn't just in the UI or the model, but in the data and metadata that power them.
Here's a breakdown of how DeepSeek's emergence could impact Agentforce and Salesforce, with a "Deepgold" twist, and how it relates to Marc Benioff's vision:
The "Deepgold" Thesis: Data and Metadata as the Real Treasure
- DeepSeek's breakthrough underscores that the LLM itself is becoming a commodity. The ability to produce a comparable LLM with significantly less capital investment and using older technology has turned the focus to the data that these models are being trained on, and how companies leverage their data.
- This is where the "Deepgold" concept comes in: the real treasure is in the data and metadata that fuels AI. It's not just about having access to the latest models, but about having high-quality, relevant data to drive business outcomes.
- For Salesforce, this is a validation of their long-standing strategy: focusing on the integration of AI within their Customer 360 platform, using a customer's own data. Unlike generic AI offerings, Agentforce is built for CRM – delivering trusted AI grounded in a customer's data.
How DeepSeek's Success Might Impact Agentforce
- Validation of Agentforce's Architecture: DeepSeek's success validates Salesforce’s approach with Agentforce. Agentforce is built as a platform that uses customer data to tailor AI agents for specific use cases, not just a generic AI model. The low-hallucination rates of Agentforce are thanks to it being tailored for specific capabilities using the customer's own data, instead of relying on general LLMs.
- Focus on Data Quality and Governance: With data becoming the new "Deepgold," the importance of data quality and metadata increases. Salesforce emphasizes the use of Data Cloud to ensure that its customers have access to their own data in a scalable, secure way, to make their agents smart without sacrificing security. This could be an opportunity for Salesforce to further emphasize their data classification metadata and their PII masking feature in their Generative AI Gateway when grounding a prompt. Customers will need to focus on the data they are using to ground the prompts, and the fields on their "hero objects" like Case, Opportunity, and Account, to make sure their data is high quality.
- Model Agnosticism as a Strength: Salesforce's model-agnostic approach could also be an advantage as it enables customers to use the best foundation model for their needs, with no model lock-in. This aligns with the rapidly evolving landscape of LLMs, where DeepSeek’s arrival is just one example of fast innovation. Salesforce can position itself as a gateway that is secure and allows customers to focus on their customer value, rather than dealing with the integration and security concerns.
- Pricing Considerations: DeepSeek's model is charging $0.55 per million tokens compared to OpenAI charging $15 per million tokens. This lower cost model will put pressure on the major cloud providers to lower their costs. Salesforce may need to evaluate their pricing strategies for Einstein Requests, as customers look for cost-effective AI solutions. Salesforce has already reduced barriers to adoption by making Agentforce tools free for Salesforce customers and by offering free AI training to customers on Trailhead through the end of 2025. This should help encourage experimentation with the product.
Benioff's Vision: A Validation and a Challenge
- Marc Benioff has described Agentforce as "what AI was always meant to be," . This statement aligns with the concept that the real value of AI is not in the model itself, but in its ability to be tailored for real business needs. Benioff's push for every Salesforce employee to focus on agents by the end of 2025 aligns with this idea, positioning Agentforce as a core offering and a key differentiator for Salesforce.
- "Don't DIY Your AI" is the mantra that Salesforce was using to echo their message from their cloud computing days. As such, the rise of DeepSeek reinforces that customers shouldn't be building AI from the ground up or trying to integrate multiple products to amalgamate their data, but that they should use platforms like Agentforce to leverage trusted data for practical business outcomes.
- Benioff's criticism of Microsoft Copilot, saying that it's not grounded in his data and spills data, highlights the importance of data security and accuracy—core tenets of Agentforce. The Einstein Trust Layer ensures that no customer data is viewed or retained by 3rd-party model providers.
- DeepSeek's emergence should push Salesforce to keep innovating and stay ahead of the curve to remain competitive.
The "Deepgold" Opportunity for Salesforce and its Partners:
- DeepSeek’s rise creates an opportunity for Salesforce and its partners to emphasize the importance of data strategy and governance when implementing AI. By positioning themselves as experts in helping customers leverage their own data to build custom AI agents, partners can capture and expand market share.
- Salesforce's emphasis on an open ecosystem allows customers to choose the best model for their needs. This flexibility is important in a rapidly evolving AI market.
- By providing pre-built agents, as well as tools like Agent Builder, Prompt Builder, and Model Builder, Salesforce is giving customers the ability to create intelligent AI solutions quickly, without needing to develop complex AI capabilities from scratch.
DeepSeek's ascent is a reminder that the AI landscape is constantly changing. While the emergence of a cost-effective, high-performing LLM presents challenges, it also reinforces Salesforce's strategy of leveraging its customer's data to create tailored AI agents. The "Deepgold" is in the data, and Salesforce, with its Agentforce platform, is well-positioned to help businesses unlock the value of that "Deepgold" for maximum business impact. The focus shifts to the importance of a solid data foundation to fuel the future of AI.
