A deep dive into DeepSeek: How the channel can thrive

DeepSeek’s R1 launch rocked the AI world last month, with channel partners and analysts alike keeping a close eye on what comes next to tap into more opportunities in the AI space.

While the Chinese company released its R1 model in January, DeepSeek has been around since 2023 and offers a range of AI models, including a chatbot.

However, R1 is different, as DeepSeek refers to it as a reasoning model, which can provide quantitative and logic reasoning. According to Gartner distinguished vice president and analyst Arun Chandrasekaran, this stands it out from the crowded market of large language model providers.

He claimed that very few companies that have released reasoning models, with only really two viable competitors – OpenAI with its o1 and o3 models and Google with Gemini.

Adding a bit more juice is the fact that DeepSeek made its model open source, allowing users to tinker around and potentially create their own iteration, requiring minimal hardware investment, like a single laptop.

Systems integrator Synapsewerx has been analysing a wide variety of AI models, and DeepSeek is no exception. The integrator’s CEO, Malcolm Ingram, spoke to ARN, claiming that while he or his team haven’t personally dived deep into the code of how it works, the fact that option is there is an attractive prospect.

“This is really unusual,” he said. “If you look down the ranking lists on any table and they’re all proprietary, but this one is open source.

“There’s a couple of other open source ones out there, but nothing in that top leadership board. What does that mean? If you want, you can run this model locally behind your own firewalls, in your own secure environment — not even connected to the internet — and you can interact with that model. You can do that under an open-source licence. There are no subscription fees, there are no API consumption fees; this starts to set it apart a little bit in the marketplace.”

Both the nature of DeepSeek being open source and functioning on smaller devices lead Ingram to imagine potential use cases for the model, such as scenarios that don’t require sensitive information and data, like vision-based scenarios or the orchestration of other AI agents.

“Using a reasoning model that works and engages through thought process with the right AI agent to use the right tool — I think that’s a really powerful thing,” Ingram said.

“Being able to do that cost effectively and securely, I personally find that exciting. I can understand why with a number of use cases — maybe manufacturing, maybe in farming, or even automotive where doing that on edge compute is really, really interesting.”

Ingram added that he has a customer that’s been talking to Synapsewerx about a scenario that involves in-car technology, and while he hasn’t put DeepSeek forward yet, the notion of having a small form factor AI model in that context is intriguing to him.

“Being able to do that in a remote environment in Australia and help provide advice and guidance to the driver about weather conditions or safety, for example, and having that in car on a smaller computer device and being responsive is really compelling,” he said.

Brad Clarke, founder of partner community platform Channel Guru, said he constantly uses AI, which includes DeepSeek. For him, one of its key strengths is its flexibility, similarly to products from OpenAI.

One of its differences compared to other industry stalwarts though is that he believes it isn’t as thorough in some of its responses.

“It seems to lack a little bit of personality to me. Gives you a wider vanilla flavour response, which is good, but it doesn’t seem to have the richness that you would get from OpenAI,” he said.

While some partners out there may be inclined to jump onto the bleeding edge of brand-new technology as quickly as possible, it’s important not to react to market movements too quickly.

In Chandrasekaran’s view, the US stock market’s immediate reaction to send shares plummeting was quite a knee-jerk move, but there are still ramifications to DeepSeek’s introduction to the world – that AI in general doesn’t have to be expensive.

“In the US, some of the large technology companies have an AI capex of anywhere from $30 billion to $35 billion, or even slightly in excess of that,” he said.

“[Venture capital firms] have invested tons and tons of money on startups where it’s almost like a conventional wisdom now, where this is all about capex; ‘This is a very expensive game to get into. So, you need hundreds, millions and billions of dollars to get into it.’”

Chandrasekaran added that the global spotlight on DeepSeek has forced a rethink on that particular strategy moving forwarded – that businesses don’t need to necessarily brute force compute and data with millions upon millions of dollars with supercomputers.

Ingram echoed Chandrasekaran’s sentiment, calling DeepSeek a “wake-up” call for companies spending a lot of money on developing and iterating models.

Keeping secure

Regardless of whichever AI model a business uses, Chandrasekaran said it’s important to consider how secure any given model is and to base that decision after proper analysis.

“I think the country of origin is certainly an inhibitor to more widespread adoption for DeepSeek, particularly in best aligned markets, broadly speaking,” he said.

Carlos Iturria, Synapsewerx CTO, held a similar belief, saying that there has been a lot of attention around DeepSeek from a political point of view.

“It’s not a US company anymore that is doing it; it’s a Chinese company,” he said. “It probably wouldn’t mean anything if it was a Canadian company, right? But it happens to be a company from China, so it creates a lot of attention.”

Ingram said the business has thought long and hard about whether it even openly talks about DeepSeek, but came to the conclusion that “we shouldn’t be afraid” to speak about it in general.

Partners may feel some level fear following the Secretary of the Department of Home Affairs issuing a mandatory direction under the Protective Security Policy Framework on 4 February, preventing Australian government entities to prevent the use or installation of DeepSeek products, applications and web services.

If a government is banning internal use of a product or services, that could cause concern among some businesses for their own usage. Similarly, business owners could be concerned that aligning themselves with a banned product may cause them to look less desirable when it comes to applying for government contracts.

However, turning away from the investigation of a product when customers have questions, or even have demand for its usage, can hurt in the long run.

Trevor Clarke, co-founder and Asia Pacific research director at Tech Research Asia, said every partner should be able to discuss the impact of DeepSeek on the market, just like with any other major AI development.

“This doesn’t mean they need to necessarily embed anything from DeepSeek – or indeed from any vendor – into their offering as that decision will really depend on the type of service they are trying to provide and a host of other factors,” he said.

“But, they should be able to provide clients with evidence-based advice on what DeepSeek’s work actually presents.”

In fact, in his view, anyone that is familiar with AI, cyber security and government technology usage would not have been surprised by the announcement of the ban.

“It really is standard operating procedure to take this step as it is in most enterprises across the world with new technology or vendors,” he said.

Ingram raised a similar viewpoint — that Synapsewerx has “a duty to understand with an open mind what the capabilities are of different technologies that come to market”.

“It’s DeepSeek today and there’s a lot of noise around it,” he said.

“Assessing these technologies is going to be important so we can provide the right advice to our customers, so that we can continue to guide them. We haven’t got one yet, but I have no doubt in the near future a customer will ask us for our opinion on DeepSeek’s open source model and I’d like to be able to give them an informed opinion.”

Determining whether new technologies are a right fit for clients is something that Brad Clarke also believes in, which means having a full understanding of what DeepSeek actually has on offer.

“If you look at DeepSeek in the context of its competitors, obviously it’s open source, it’s Chinese-backed, and that creates some potential concern. But there is some cost of energy [concerns], so it is cheap in some regards. Some partners may be looking for the greatest spend control. DeepSeek and OpenAI are probably the two best platforms for customisation and they’re looking to provide real time and real value solutions for customers,” he said.

“One of the things I see a lot in the channel, and something that Channel Guru is really focused on, is a lot of partners are struggling to be really proactive in guiding their customers,” he said.

“As new technology comes on, like DeepSeek, it’s a really great opportunity, not just to understand what it is, but use that as an opportunity to go have a proactive point with new customers. Educate not just your key contacts inside an organisation, but all of their employees. I think partners do that more regularly with new technologies like DeepSeek and other innovations in the market.

“All of that helps to justify the ongoing engagement of managed services offerings that many partners are seeking.”

With that being said, there are still concerns to be had around DeepSeek’s security, as there would be for any new AI model.

In the context of DeepSeek, the company has its web- and app-based consumer-facing chatbot and its other models, including R1, and it’s important to not confuse one with the other.

However, this is no different than most popular competitors in the market, with Ingram pointing to DeepSeek’s openness in its terms and conditions for its web-based and mobile app-based products.

“You can just go check on their terms and conditions and they openly say that they will collect your data, they will use that data, they will store that data and they talk about the sovereignty of data,” he said.

“In general, from an Australian market point of view, we would always advise customers to be very considered by any model that they use and employ within their organisation. In fact, we go through quite a rigorous selection process with customers around models to help them navigate this field, but it’s all the usual stuff.

“You’ve got to think about security, you’ve got to think about your governance framework. You’ve got to think about sovereignty, especially so when you’re talking about personal information, when you’re talking about sensitive information or sovereignty. These are really important factors for many Australian businesses.

“That’s true of DeepSeek just as much, in my opinion, as OpenAI or Meta or whoever.”

However, the R1 model can be contained entirely within an Australian firewall, which prevents prompts leaving a given network.

Chandrasekaran noted that OpenAI and Google’s consumer products also face similar concerns.

“With Google and OpenAI … the consumer products have a very similar concern in terms of my input to the model could be used to train the model, because it’s a free service,” he said.

“Of course, when it’s a free service, in some sense, you’re becoming the product.”

With DeepSeek, Chandrasekaran said the issue is that its commercial offering is not very well known in the market, so there hasn’t been as much time placed into investigate the extent of its risk postures.

“The entire world woke up to this company just [a month] ago, so they haven’t had that time to really fathom and investigate all of these risk postures adequately,” he said.

“With Open AI and Google, we’ve had a long time, at least two years to really wait and understand their SLAs.”

For now, DeepSeek is in its relative infancy on the global stage, and whether R1 or any of its other models are of use to partners comes down to whether there is customer demand for its usage.

If there is, there are ways of keeping data within country, Chandrasekaran added.

“You could certainly offer the outward model as an inference in API, but keep the data like within the Australian boundaries,” he said. “A lot of American companies already have started to do that. They’re offering the DeepSeek model within their platform, but they’re keeping that data in country. So, I don’t see any reasons why Australian partners cannot do it if the Americans can.”

Will DeepSeek last? Ask again later

Another factor to consider, which is impossible for anyone without the power to see into the future, is whether DeepSeek has lasting power in the market.

“It’s one thing to acknowledge the fact that DeepSeek has been a very innovative company and really raised a lot of attention to their innovation, but I also keep thinking do they have the money? Do they have the intent? Do they have the know-how and the talent to innovate on a continual basis?” Chandrasekaran said.

“It’s one thing to innovate once and raise a great model, but it’s an entirely different story to be on this high speed treadmill and continue to innovate. As a partner, you’re not looking for one trick ponies. You’re looking for companies with sustained innovation and ability to run this race for a long period. It’s too early to determine whether DeepSeek has that stamina to run the race for long.

“They definitely surprised the world, and they’ve definitely shown themselves to be an innovative company, but we have to wait and watch, because I’ve seen a lot of companies that create something amazing, and then after that, they go quiet for a year … and then they pan out.”

In Brad Clarke’s view, DeepSeek’s long-term viability is unclear, referring to it as the “great unknown”.

“We don’t really know what the long term road map is, or is there going to be any regulation from local regulators or US regulators,” he said.

Meanwhile, Trevor Clarke said that while DeepSeek’s output is impressive, he cautioned against partners aligning themselves with any one vendor.

“While I have already had discussions with partners that are playing with the DeepSeek open source models in a secure sand box – and most of them are impressed with how good it is compared to previous open source models – it is very likely that we will see even more advanced and capable models released in the near future,” he said.

“So, I don’t think the question should really be about adopting DeepSeek or not. The real challenge for partners is going to be their adaptability and velocity in keeping their AI-related services evergreen with the latest capabilities from wherever they originate. We can’t really predict the next DeepSeek moment, but we can certainly expect it to happen pretty soon.”

Ingram said Synapsewerx is looking and evaluating DeepSeek at the moment, but isn’t proactively proposing it to customers.

“Certainly, customers mention it in conversation, but we have a pretty standard default response at the moment just around AI models in general and it’s a structured approach; it’s about going through a process to select the right model,” he said.

“Would we include it in a list of things for a customer to consider? Most likely in the right scenario.”

Regardless of whether DeepSeek keeps its momentum going or fades into obscurity, Ingram believes its existence will stimulate innovation in the market, adding that he would love to see some of that effort coming from a home-grown source.

“I would love to see Australian universities and startups getting the investment that they need to be building Australian models of this level and capability,” he said.

“Will DeepSeek still be on the leaderboard in six months’ time? I’m not sure. Will there be other models coming out of other more traditional stables? Quite possibly. It’s exciting to watch the market and see where we are in that innovation life cycle of intelligence.”

Additionally, the Synapsewerx CEO added that he would also like to see OpenAI take a page from DeepSeek’s playbook and open up a bit itself.

“It’s been, what, two years since they had an open source model? They’ve really gone into this commercial proprietary model and I get the impression they transitioned from being a not-for-profit to profit.

“We understand, people have invested heavily [but] having open-source models that are extremely capable — I think we’re going to see more of them.”

Whether DeepSeek holds onto its crown as the top innovator in the AI space or another company takes it, there’s one thing Ingram is certain about – that the future is unclear.

“I can’t even predict where we’re going to be in three years’ time. The pace at which [AI is] going is pretty phenomenal,” he said.

Iturria however was able to provide his own forecast: “the only thing that I can predict is that 2025 is going to be a wild ride [and] January did not disappoint.”