What the DeepSeek AI frenzy means for marketers

DeepSeek, a new large language model known for its low-budget computing potential, could reshape how marketers develop AI by enabling more affordable and customizable applications for their brands—an alternative to relying on AI from Big Tech companies, experts told Ad Age. But marketers are split on whether DeepSeek itself, not just its research, could be a viable part of their AI toolkits, on account of the platform’s ties to China and opacity around data collection. This week, DeepSeek’s surge in visibility after releasing a consumer-facing app, gave a new roadmap to ad agencies that could lead to more AI products, developed more efficiently than marketers thought was possible, according to ad leaders. Since late 2022, as large language models (LLMs) became mainstream, the ad industry has come to expect that its AI applications would rely entirely on a few Big Tech companies, such as Google, Microsoft and Meta, said Martin Pagh Ludvigsen, director of AI and creative tech at Goodby Silverstein & Partners. These giants and well-funded startups such as OpenAI were the ones with the money to develop the computational resources to train advanced AI systems. “This [thinking] has all been thrown into the air” with the release of DeepSeek’s next-generation R1 large language model, Ludvigsen said. Marketers have been trying to understand the implications of DeepSeek’s advances. The AI proved to be potentially as sophisticated as OpenAI, the maker of ChatGPT. It could, however, require far less computing power and money to develop, and its open-source software makes it readily available to all. Open-source refers to underlying code that is transparent and free to use, as opposed to the black-box of closed-source code, with which most Big Tech companies build their AI systems. Meta’s Llama model is partially open-source, but the tech giant still keeps important aspects of its AI, such as how it is trained, in the dark. Also read: AI marketing expert tests DeepSeek GS&P and other agencies have mostly disregarded the idea of building bespoke models through open-source code because they lack customer support, which is readily available through partnerships with enterprise vendors, Ludvigsen said. But the sheer cost-saving opportunity demonstrated by DeepSeek may force a reconsideration from agency holding companies. Last month, DeepSeek researchers published a paper that showed they could develop an LLM that is nearly as powerful as other top AI systems, at a fraction of the cost. DeepSeek released its consumer app this week, which shot to No. 1 on Apple’s app store and set off an AI frenzy in the U.S. over fears about what it meant for tech companies. AI chip giant Nvidia’s share price dropped 17% on Monday, before rebounding 9% on Tuesday. Meta reportedly put together war rooms to study DeepSeek’s budget-friendly AI. DeepSeek used just over 2,000 chips for training purposes, costing roughly $6 million, according to the platform’s researchers. That is a far cry from the hundreds of thousands of chips—worth billions of dollars—that Big Tech companies are using to train their own models. This cost differential is largely why renewed fears of a bubble have crept into the AI market. NextGen Marketing Summit Join us for a crash course on Gen Z, DTC marketing and the creator economy Register today Advertising companies have also invested heavily in AI. Last year, WPP, Publicis and Havas each announced investments in the technology of more than $300 million, while other holding companies expanded budgets of their own. To be sure, Ludvigsen does not see these deals as any less valuable in light of DeepSeek. Much of the money invested is targeted towards integrating AI across workflows as opposed to training models directly. But the cost-efficiency proved by DeepSeek could encourage ad holding companies to more seriously consider training their own models, Ludvigsen said. Not only would these models be cheaper to use compared to Big Tech’s AI systems, they could also be developed from the start to handle a parent company’s specific network of agencies and clients, Ludvigsen said. Nicole Penn, chief executive of agency EGC Group, expects the DeepSeek moment to birth a new crop of enterprise marketing platforms. Custom-built AI agents could be developed with fewer resources, lending to innovations in customer service, e-commerce and other marketing-related services, Penn said. “This is going to make AI integration more accessible and 50 to 60% more cost-efficient,” Penn said. Marketers split on DeepSeek While many marketers seemed to agree that DeepSeek heralds a new moment in the AI space, they are less in harmony about the viability of DeepSeek itself. Penn, who has experimented with DeepSeek, is already introducing the platform to curious clients, although EGC hasn’t yet built any applications on top of it. The R1 model, the latest generation of the LLM, is especially valuable for its transparent reasoning, which allows users to better understand the system by learning how it arrives at a response, Penn said. This quality is not as prevalent in OpenAI’s models. The cost to use DeepSeek is also an advantage for marketers, said Aaron Andalman, chief science officer for programmatic ad platform Cognitiv. For every million output tokens, or units of text that the AI model produces, DeepSeek’s R1 model costs roughly $2.16, according to Andalman. In contrast, OpenAI’s o1 model costs $60 per million output tokens, and its cost-efficient version, o1-mini, is priced at roughly $12 per million. This week, another important endorsement came from Perplexity, the AI search startup, which began serving results powered by DeepSeek’s application programming interface (API). Perplexity has a handful of brands testing ads on its platform, which may now surface alongside DeepSeek’s content by virtue of appearing in its search results. Perplexity did not respond to a request for comment. Meanwhile, some marketers advised against using DeepSeek because of its opacity. For instance, the Asman Group, a marketing consultancy, will not use DeepSeek because of concerns about how the platform collects data, founder and managing principal Greg Asman wrote on LinkedIn. Jon Hackett, VP of technology at agency Huge, also recommended that brands not yet migrate to the platform and away from Big Tech providers because of uncertainties surrounding DeepSeek’s training data, Hackett told Ad Age. DeepSeek is owned by a Chinese hedge fund and, according to some ad leaders, is already showing instances of political bias. In experiments with DeepSeek, GS&P’s Ludvigsen found that the platform would not provide historical information on Tiananmen Square from 1989, when the Chinese government killed protesters of a student-led demonstration. DeepSeek also spouted anti-Taiwan independence rhetoric in accordance with the Chinese government’s position, Ludvigsen said. As noted by Asman in his LinkedIn post, DeepSeek’s political ties are reminiscent of the ongoing drama with TikTok, which is facing a ban in the U.S. on account of its relationship to China. TikTok shut itself down earlier this month before the ban was set to take into effect, but was brought back online through a 75-day extension to explore a sale. Many advertisers have resumed their spend on the platform in the meantime. The concerns around DeepSeek are reason enough for GS&P not to consider the platform a viable partner, Ludvigsen said. But the insights underlying DeepSeek—succinctly published by the company in its paper last month—were still valuable. “The breakthrough is not the platform; it’s the research,” Ludvigsen said.

What the DeepSeek AI frenzy means for marketers

Featured Stories