Photo by Solen Feyissa on Unsplash.
DeepSeek’s R1 chatbot has burst onto the scene with equal parts fascination and fear, with claims that it not only uses less data than its competitors but is cheaper to run sending the shares of technology companies worldwide into a nosedive as a result.
Supposedly trained in two months on less than $US6 million - a fraction of the computational power required to train comparable models like Open-AI’s GPT-4 (which cost $US100 million) – the app piqued the interest of consumers with its free to use and open-source model.
In response, OpenAI and Microsoft began investigating whether the Chinese rival used OpenAI’s API to integrate the company’s AI models into DeepSeek’s own, while closer to home, the NSW Government blocked the chatbot from its devices and systems over privacy concerns, as did fellow federal corporations NBN Co, the ABC, and Australia Post.
The Australian advertising and marketing industry has been watching developments closely, with some media agencies praising the large language model (LLM) as a gateway for higher-quality AI and faster campaign rollouts, but others caution that questions around data and security remain opaque.
The potential to ‘revolutionise local advertising and tech industries’
Orange Line’s SEO director Emil Mequita said she foresees fewer people relying on traditional search engines like Google as a result of DeepSeek’s introduction.
“If DeepSeek can keep its promise of cost efficiency it could change the scope of what's possible with AI and LLMs - eventually I fully expect to see local advertisers penetrating the LLM space,” she said.
Jaywing’s strategy director, Ed Raine, said his agency can already see people choosing to move to DeepSeek as it’s free and providing good, if not in some cases better, results.
“It gives more people access to high quality AI with lower costs associated which has benefits to all types of Australians,” he said.“This in turn should mean that tools and platforms that leverage APIs into AI models should be able to charge less, which will mean a great take up of their services, creating the potential for more innovation and leveling the playing field for smaller companies in particular.”
Raine did caution that there are still a lot of question marks over data privacy when it comes to Deepseek, however.
“Open AI only recently came out with an enterprise solution that appeases the concerns of larger organisations so it will take time for DeepSeek to do the same, and even when they do, will Australian businesses be ready to hand over sensitive information to a Chinese owned business?” he said.
CX Lavender’s managing partner and head of technology, Clint Bauer, said DeepSeek brings powerful AI capabilities that could revolutionise local advertising and tech industries – think hyper-personalised ads, automated content creation, and smarter audience targeting.
“For marketers, this could mean faster campaign rollouts and pinpoint accuracy, saving time and boosting efficiency,” he said.
“For the broader tech sector, especially AI-driven applications, DeepSeek offers an alternative to the usual players, but unlike Western models that follow strict data protection rules, DeepSeek’s privacy policies hint at data being stored in China. That’s a red flag for businesses operating in regions with tight data sovereignty laws.”
Bauer said that accessibility is also a question, as DeepSeek’s power comes from massive datasets and cloud infrastructure – resources that smaller businesses might not have the budget for.
“So, while large corporations could jump in, startups and mid-sized companies might struggle to keep up, and in industries like finance and healthcare, where data security is non-negotiable, adoption will likely be slow until the regulatory fog clears,” he said.
DeepSeek’s R1 model, released in January, was built on DeepSeek’s V3 model, released a month earlier.
Kinesso’s national head of AI and analytics, Kellyn Coetzee, said that any breakthrough that slashes processing time and costs is a major shift, and for Australian developers, the V3 version of DeepSeek’s chatbot could reduce the headache of code generation, documentation, and testing, effectively bringing speed and simplicity to their daily grind.
“For local advertisers, especially the smaller players, V3 will usher in an era of AI adoption,” she said.
“When a cutting-edge AI tool becomes more affordable and more accessible, it levels the playing field and we will start to see smaller organisations punching above their weight on the global stage. Automated, high-quality copywriting gets easier, letting creatives and strategists stay in their lane and focus on the big ideas.”
Coetzee said that she doesn’t believe it’s ideal for commercial use due to a number of reasons, though.
“It’s heavily biased and doesn’t work well in application, it’s highly sensitive to how it’s prompted, while it excels in English and Chinese, its lacking with other languages and has glaring gaps in its knowledge base,” she said.
“One would need to use it carefully, sparingly and wisely, if at all.”
OpenAI seeing audiences increase
IAB Australia, the country’s trade association for online advertising, monitors digital consumer media habits closely via the Ipsos iris ratings currency and while there have been significant increases in the amount of people and time spent on AI services for a range of activities including discovery, it is currently additive to “traditional” search time.
For the search engine category, the average monthly audience for calendar year 2024 versus 2023 was up 2% and time per person was up 9%.
By comparison, OpenAI has seen audiences increase 36% in the same time period and time per person increase 90%.
IAB CEO Gai Le Roy said three things have been top of mind for her among the attention placed on DeepSeek.
“Firstly, innovation and disruption is to be expected in this space,” she said.
“Secondly it reminds us of the need to educate consumers on privacy alongside fitting privacy regulation.
“And finally, the most interesting element for me is the development of a model that needs less computing power. It is most encouraging that new models are being developed that may reduce the impact that AI is having on the climate crisis.”
No immediate impact for Australia
Not everyone is sold on the potential effects for the local advertising industry, though.
A number of media agencies AdNews spoke to said that they don’t see the shakeup from DeepSeek as major as perhaps initially projected, with existing technologies still preferred – both due to greater understanding of how they work and less questions about security or safety practices.
That privacy factor, raised even by those who were bullish on DeepSeek’s entry into the market, is a possible impediment to mass takeup, as questions around how to ethically use data and how secure said data is within DeepSeek’s models are yet to be solved.
Prophet’s CEO Jordan Taylor- Bartels said DeepSeek's R1 model, particularly its open-source version, is not anticipated to have a direct and immediate impact on local advertising or technology industries, but the impact will be more nuanced and indirect through a chain reaction.
“The open-source cycle promotes continuous improvement and other models such as Llama and potential new entrants will learn from DeepSeek's open-source release to further advance their capabilities,” he told AdNews.
“Local advertising and technology industries will indirectly benefit as ‘Western’ friendly models (such as Llama, ChatGPT, or even models from Anthropic) that build off of DeepSeek's innovations get better, ultimately providing more sophisticated and accessible AI tools for businesses.
“A reducing cost trend in deploying AI models like DeepSeek's R1 makes the technology more accessible, not only lowering barriers for existing companies but potentially enabling new startups to enter the market, pushing further advancements.”
The Pistol’s head of strategy, Emilia Chambers, agreed that it will be a while before the industry sees any significant impact due to the fact that there is often a concern around safety when it comes to Chinese-developed technology due to cybersecurity challenges raised over the last few years.
“Given this, I'm expecting that brands and agencies will still explore the DeepSeek capabilities and how they could improve processes and performance, but in a more cautious way than what we have seen with other AI technology,” she said.
OMG’s chief product officer, Alex Pacey, said DeepSeek represents another (and potentially even cheaper and therefore more widely available) way to get to a rapid answer quickly, but it’s a rapid answer in the same way a microwave meal is the answer to an anniversary dinner.
“It’ll get you to an answer very quickly, but it’ll taste the same as the rest - humans are the difference makers, and we need to remember that to protect the quality of the product we put out there,” he said.
Yango’s GM of product, Robert Nagy, said using various AI tools allows for the ability to analyse a lot of data to create more personalised campaigns and optimise at scale.
“It can speed things up by automating content creation and getting messages out there faster, while also improving campaign measurement and optimisation for increased effectiveness,” he said.
“Will DeepSeek specifically drive this change? No. Each tool in market has its unique strengths and weaknesses which we believe will continue to promote competition in this area.”
Clemenger BBDO’s strategy partner and head of product, Maya Mausli, said the idea that DeepSeek can tap into sub-conscious drivers is fascinating, but it won't lead to better storytelling.
“We could easily see the market inundated with personalised but transactional creative - advertising creativity should not fall at the altar of personalisation,” she said.
Half Dome’s head of digital and growth, Joe Frazer, said he also wouldn't anticipate a significant direct impact on the local market, but if DeepSeek truly manages to reduce the cost of AI development and training to the level initially claimed, that could be good news for smaller or less-funded AI players.
“Historically, these companies have struggled to compete with larger rivals who can invest heavily in data centres,” he told AdNews.
“It’s worth noting though that most advertising industry use cases don’t demand so much compute that costs become prohibitive—even at relatively higher rates. The real shift will be if DeepSeek’s approach proves genuinely scalable and cost-effective, which could encourage broader AI adoption and spur competition.
DeepSeek failed to provide accurate information
An audit from NewsGuard, a rating system for news and information websites, found that DeepSeek failed to provide accurate information about news and information topics over 80% of the time.
It also debunked provably false claims only 17% of the time.
NewsGuard applied the same prompts it used in its December 2024 AI Monthly Misinformation audit to the Chinese chatbot to assess how DeepSeek performed against its competitors on prompts users might make relating to topics in the news.
NewsGuard found that with news-related prompts, DeepSeek repeated false claims 30% of the time and provided non-answers 53% of the time, resulting in an 83% fail rate.
By comparison, NewsGuard’s audit on 10 other chatbots (OpenAI’s ChatGPT-4o, You.com’s Smart Assistant, xAI’s Grok-2, Inflection’s Pi, Mistral’s le Chat, Microsoft’s Copilot, Meta AI, Anthropic’s Claude, Google’s Gemini 2.0, and Perplexity’s answer engine) found that they had an average fail rate of 62%.
In the case of three of the 10 false narratives tested in the audit, DeepSeek relayed the Chinese government’s position without being asked anything relating to China, including the government’s position on the topic.
Cost of training sparking interest
One of the most tantalising possibilities of DeepSeek is how much cheaper it is to train than other models, such as ChatGPT.
CX Lavender’s Bauer said that AI requires serious computing power, vast amounts of data, and solid infrastructure, but DeepSeek benefits from China’s lower cloud costs and government-backed AI initiatives, making it a potentially more affordable option than OpenAI’s GPT models.
“But affordability doesn’t always equal reliability - western AI providers focus on privacy-first solutions like federated learning and stringent compliance measures, while DeepSeek collects a broad range of user data, which helps fine-tune its AI but raises concerns in regions with strict privacy laws (Australia, the UK, and the EU, to name a few),” he said.
“One way around this? Businesses could look at hybrid AI models – using DeepSeek for less sensitive tasks while keeping critical data in more secure environments. Privacy-enhancing technologies and close collaboration with regulators can also help navigate compliance challenges. The real question is whether the cost savings outweigh the potential regulatory headaches.”
Bauer said for advertisers, the appeal is obvious – AI-driven personalisation at scale - but consumers are getting savvier about their data, and brands can’t afford to erode trust.
“A smart approach? Use DeepSeek where it makes sense – like content generation for non-sensitive applications – while keeping a tight grip on data governance for high-risk areas,” he said
“As AI regulations continue to evolve, companies that stay ahead of compliance will have the edge.”
Lexlab’s director and founder, Alfie Lagos, said DeepSeek’s arrival is less about groundbreaking innovation and more about the ongoing correction in AI’s commoditisation.
“A prime example of this is the recent US fund (up to $US500 billion) created to support AI development, which may have contributed to an overestimation of the true cost of innovation in this space,” he said.
“The hype around it is, in part, a reflection of the media’s tendency to latch onto the tall poppy syndrome - an eagerness to build something up, only to tear it down. At best, what DeepSeek will do is help price AI model development closer to its true value.”
Lagos said there’s an ongoing debate about who’s allowed to copy from whom, given that models like ChatGPT were initially trained on publicly available internet data, so even if DeepSeek was built on top of existing LLMs, none of these companies created the original content they now train on.
“The real innovation in AI will come from advancements in reasoning, practical virtual assistants, and the seamless integration of AI with real client data and processes,” he told AdNews.
“However, the most critical issue for advertisers isn’t which AI model is newest or cheapest; it’s data security. Businesses need to be extremely cautious about what client data they feed into these models. Without proper safeguards, sensitive information could end up contributing to broader AI training, putting both brands and agencies at risk.
“DeepSeek has yet to provide a clear way to opt out of training beyond its Perplexity Pro tier, which is how we’ve been trialling it. If it wants to be taken seriously as a viable alternative, it must offer explicit guarantees on data protection. Perplexity, for instance, has addressed similar concerns by promising that data remains US-hosted and does not get sent to China.”
Nick Hayes, The Media Store’s head of digital, said if DeepSeek’s claims that their models are 95% more efficient than OpenAI’s are true and big tech has grossly over-invested in AI capital expenditure, then high powered chips, large data centres and high volume energy consumption won’t be needed to support its growth in the way they’re currently forecasted to.
“All of that capital expenditure investment is predicated on a return and if investors can’t make their profit, this could have major ramifications for the evaluation of these giant tech companies (and related industries),” he said.
“While this might impact the energy and real estate sectors in Australia, the low cost of entry for AI startups through platforms like DeepSeek opens a world of opportunity for the Australian media industry.”
Hayes said that with the ability to process complex data sets 25x cheaper and faster than the current leading LLMs, the industry is likely to see more near-real time data signals influencing everything that happens in media.
“Examples could include websites with high CO2 emissions being identified in real time and removed from bidding with each auction; MMM models rerunning daily to become part of BAU media optimisations and pacing; DSPs that offer forecasting and planning tools could become scarily accurate; customer lifetime value modelling and audience journey mapping would be less encumbered by walled gardens; real-time audience signals could inform more auctions, custom bidding algorithms, dynamic creative optimisation and personalisation; privacy tools could capture and prevent data breaches/fraud as they’re happening; and programmatic inventory management and bidding could become much more efficient, decreasing costs in media,” he said.
Hayes said that this change would likely impact agency fees as well.
“As each step in the planning, buying and delivery process becomes more refined and efficient, clients will further consider AI’s impact on retainers and media fees,” he said.
“To evolve with these changes, media agencies will need to ensure that all aspects of their business are growing with the pace of AI, including their scope of services.
Kinesso’s Coetzee said that training models like DeepSeek’s often involves repeated experiments, data wrangling, hefty salaries for research teams, and all the miscellaneous bits and bobs (infrastructure, overheads, failed prototypes) that don’t always make the press release.
“DeepSeek claims to have used 2048 H800 GPUs for training, which alone would cost between $50-100 million,” she said.
“In DeepSeek’s case, a generous GPU stockpile from its parent investment fund points to big-time infrastructure backing that isn’t fully reflected in the upfront figure.
“If DeepSeek can genuinely reduce the financial barrier to entry for advanced models, that’s a welcome shake-up for local agencies and tech companies alike. After all, cheaper AI training means smaller outfits can jump into the deep end and innovate in ways once reserved for big-budget juggernauts.”
The Pistol’s Chambers said that the cost of training DeepSeek's models is cheaper when compared to the more widely used LLMs, such as Anthropic’s Claude.
“The reason for this comes down to the hardware, with many Chinese LLMs relying on less advanced hardware and often starting off by focusing on more specific applications that require less computational power,” she said.
“This helps them to keep the cost down compared with many of the bigger American LLMs. We’ve seen this across other LLMs out of China, including Alibaba’s Qwen plus and ByteDance’s Doubao 1.5-pro.
“It is worth noting though that despite the cheaper cost, initial performance comparisons, specifically accuracy, have been positive with DeepSeek on par with the AI major players, including ChatGPT.”
Yango’s Nagy said while there are still a lot of questions surrounding what the actual cost was to deliver the R1 model, what’s clear is that DeepSeek is able to offer this model to customers at a fraction of the price of other established AI models, making it more accessible.
“DeepSeek was able to use less advanced technology and provide comparable performance in many areas,” he said.
“However, it is not as advanced in creative writing and providing detailed explanations as some of its competitors, but it excels in more technical areas such as coding.
“The release of more accessible models challenges the notion that AI requires ever-increasing amounts of funding to develop more advanced technology, and larger amounts of energy to run.
“There are improvements that can be made in terms of efficiency which allows these tools to be available to more people and businesses and help with a wider array of tasks that will help support the growth of the media and advertising industry as a whole.”
OMG’s Pacey said that the real game-changer DeepSeek appears to represent is a leap in learning efficiency and thus a reduction in cost versus its competitors.
“This could signal a future where smaller, distilled models can punch well above their weight,” said Carey.
“Why does that matter? Because while supersized models like R1 will keep pushing AI’s outer limits, it’s the leaner, meaner versions that will actually put AI agents to work in the real world.”
“As AI agents evolve, specialised systems are built to tackle complex tasks autonomously, the question isn’t just can they work? It’s can they work at scale? The answer hinges on models getting smaller, faster, and cheaper to run. And right now, all signs point to yes.”
Prophet’s Taylor-Bartels said while the DeepSeek training costs remain a substantial point of investment, the decreasing deployment costs highlight a trend towards democratising AI technology, providing powerful tools at lower costs which can be used broadly across diverse domains.
“The significant reduction in the cost to train (not necessarily build) models like DeepSeek’s is an important trend, signaling a crucial transition in the AI field, but the more intriguing factor currently is the cost to deploy and use these models via API, which is becoming significantly cheaper,” he said.
“The lower deployment costs mean that advanced AI technologies are now accessible to virtually anyone with a relatively capable device, regardless of their socioeconomic status. This level of accessibility is unprecedented, making sophisticated technology available to a broader audience at faster speeds - essentially democratising it.
“By ensuring that compute-light large language models (LLMs) are becoming more efficient, the technology is evolving dramatically, akin to the transition observed from 3G to 4G in telecommunications.”
Taylor-Bartels said that privacy concerns and the practical use of consumer applications by enterprises may be overstated.
“The real story lies within the innovation, cyclic progression of open-source models, and the cost dynamics reshaping accessibility and practical deployment of advanced AI technologies,” he said.
Half Dome’s Frazer said while DeepSeek’s claim of lower cost is encouraging, the full picture doesn't seem entirely clear.
“Reports suggest the company has a comparatively modest level of investment and limited access to the leading technology, but conflicting information from industry insiders seems to be clouding the accuracy of initial claims - it is a wait and see for me in this space,” he said.
“Regardless, what’s certain is that competition in the AI space is intensifying. As more players enter the market with innovative approaches, we can expect faster progress overall—and ultimately, both consumers and business users of AI models stand to benefit from this.”
Have something to say on this? Share your views in the comments section below. Or if you have a news story or tip-off, drop us a line at adnews@yaffa.com.au
Sign up to the AdNews newsletter, like us on Facebook or follow us on Twitter for breaking stories and campaigns throughout the day.