DeepSeek: The Game-Changer in AI Technology and Its Impact on IT Services & MSPs

[00:00:00] Richtig spannend, richtig vielfältig. Das ist deine Karriere bei Kaufland. Ob Trainee-Programm, Direkteinstieg, Studentenjob oder Praktikum, finde bei uns den Einstieg, der zu dir passt. Profitiere von einer attraktiven Vergütung, spannenden Aufgaben und individuellen Entwicklungsmöglichkeiten. Werde jetzt Teil unseres Teams. Bewirb dich jetzt unter www.kaufland.de. slash Studenten. Kaufland. Hier bin ich richtig.

[00:00:55] Ich bin, die Frage zu den Daten. Ich bin, die Frage zu den Daten. Ich sage, warum wir in dem? So, ich starte mit der Geschichte. Ich bin, die Begeber zu sagen. Big shout-out to Ben Thompson. DeepSeek ist ein Chinese AI-Company. Die News starten dieses Wochenende. Die key events date back zu letzten Januar, wenn die Kompany released der V2-Modell. Two key breakthroughs there. DeepSeek MOE, oder Mixture of Experts.

[00:01:24] Unlike models that activate their entire network, like GPT-3.5, this system only activates relevant expert components. For comparison, GPT-4 used 16 experts with 110 billion parameters each. V2 improved this by adding specialized and shared experts plus better load balancing during training. And DeepSeek MLA, or Multi-Head Latent Attention.

[00:01:49] This breakthrough significantly reduced memory usage during inference by compressing the storage needed for context windows, which typically requires substantial memory for token keys and values. The key result? It's vastly cheaper to train. Orders of magnitude cheaper. The company then released a V3 of their model over the holidays, revealing those lower costs then.

[00:02:15] The V3 model also had capabilities in line with OpenAI's 4.0 model and Anthropics Sonnet 3.5. Finally, they announced the R1 model, which is comparable to OpenAI's O1 reasoning model. That's notable, as is their app, which was released on January 20th. The app is part of how this firestorm of attention came about, much like the ChatGPT release.

[00:02:40] Plus, they did all of this on NVIDIA's cheaper H800 chips because the company could not access the better H100 chips due to the export ban. This forced them to be incredibly creative within constraints. In fact, going below NVIDIA's programming language to program more on the chip itself. The end result? Similar results to OpenAI at 3-5% of the cost.

[00:03:05] The pricing for R1 starts at just $0.14 for 1 million tokens, compared to OpenAI's charge of $7.50 for the same amount. Now, I'd be remiss if I didn't note that researchers at UC Berkeley recently developed an open-source model comparable to O1 in just 19 hours at a cost of around $450. And how has everyone reacted? Well, hair on fire.

[00:03:31] The spotlight is on the market's adjustment to a cheaper and more accessible future for artificial intelligence. Following a significant market downturn, experts suggested that companies can now achieve greater profitability with reduced spending on AI resources. While NVIDIA experienced a staggering loss of $600 billion in market value in a single day, Apple gained over $100 billion.

[00:03:54] As companies collectively plan to spend over $300 billion on capital expenditures for AI this year, the implications of DeepSeq's advancements could reshape their financial strategies. There's upheaval on the geopolitical stage, too, as assumptions about China's relative position to the U.S. appear also to have been flawed. A lot of VC money is tied up in AI, which may prove to be a gamble that will not pay off.

[00:04:21] Despite bipartisan efforts from both President Donald Trump and former President Joe Biden to limit AI advancements in China, DeepSeq's model raises questions about the necessary investment for AI. Startup Perplexity has launched a new uncensored AI search tool based on DeepSeq, aiming to revolutionize how users access information. The development comes just before Google's significant $2.7 billion deal with the startup, highlighting the rising competition.

[00:04:50] And there is privacy backlash. Chinese companies are required to disclose to the Chinese government data that is requested. Quote, The personal information we collect from you may be stored on a server located outside of the country where you live. Quote, DeepSeq's privacy policy states. Quote again, We store the information we collect in secure servers located in the People's Republic of China. End quote. It continues, Quote,

[00:05:16] Where we transfer any personal information out of the country where you live, including for one or more of the purposes as set out in this policy, we will do so in accordance with the requirements of applicable data protection laws. End quote. Microsoft CEO Satya Nadella highlighted the phenomenon known as Jevons Paradox, suggesting that as artificial intelligence becomes more efficient and accessible, its use is expected to skyrocket, turning it into a widely used commodity.

[00:05:46] Why do we care? So we care about the freak out because a number of technology companies bet that the value was in AI infrastructure. So let's assume that the research published holds up and the fundamentals of DeepSeq are sound. That being true, then the assumption that AI will need dramatically ever increasing compute, data and energy is a flawed one. And those companies who were betting on that are wrong.

[00:06:12] A number of startups in the AI space just were reminded what most small business owners who don't have outside funding know every day. Lack of resources causes innovative thinking. A number of tech companies here were swept up in the idea of get more money and solve the problem rather than actually focusing on building an efficient, profitable company from day one. This suggests that the AI market is transitioning from bigger and costlier is better to optimized and efficient is better.

[00:06:40] While this doesn't diminish the role of high-end chips or advanced hardware entirely, it puts a premium on engineering ingenuity rather than brute force. But that's not services companies. In fact, AI expertise may be a deciding factor here. A piece in GeekWire focuses on that. The best approach may be to apply, quote, Deep domain expertise to create highly optimized specialized models at a fraction of the usual cost, end quote.

[00:07:08] And now we get to solution providers and more. Why do we care? That thinking with one word change. Apply deep domain expertise to create highly optimized specialized solutions at a fraction of the usual cost. The focus is moving from what you can build with AI versus how much compute you can assemble. This is great news for customers and for service providers.

[00:07:33] Instead of allocating massive budgets toward compute infrastructure, MSPs and IT consultants could redirect funds to value-added services. AI is no longer a luxury offering. It becomes an everyday tool. Developers haven't come up with a lot of good ideas yet. Lots of stuff is half-baked. Cheaper AI is great for accessibility, but it also means that AI's perceived value could decline.

[00:07:57] If everyone can deploy high-functioning AI solutions at low cost, differentiation shifts entirely to execution. MSPs will need to double down on delivering business outcomes, not just deploying models and products. But here's the good news. The work around making data useful for AI is still just as valuable as it was before. And now that the cost of infrastructure should plummet, those limited technology resources can be focused on customer needs, not infrastructure.

[00:08:26] The value chain is shifting away from owning or managing compute resources to data preparation, consensual understanding, and solving customer-specific problems. For providers, this means offering expertise in data wrangling, integration, and model tuning, becoming the primary source of differentiation. Solving problems still retains its high value. And now you don't have to waste money on expensive infrastructure.

[00:08:50] DeepSeq's breakthroughs may not be a silver bullet, but they're a powerful reminder that the future of AI is about doing more with less. And that's a philosophy service providers are well-equipped to embrace. Today's episode is supported by Huntress. You want to focus on your clients and are always looking for ways to get more time. Use Huntress' fully managed cybersecurity platform to fight off cyber threats.

[00:09:18] Huntress is more than cybersecurity software for endpoints and identities. It's a 24 by 7 security operations center. It's security awareness training, community engagement, and dedicated partner support with an average CSAT score of 99.3%. Technology can only get you so far. Human expertise is what's needed to truly elevate and protect small businesses. And you get that with Huntress.

[00:09:45] Secure your clients and help them thrive with the number one rated EDR for S&Bs on G2. Visit Huntress.com slash MSP radio to find out more. Thanks for listening. Today is National Daisy Day. I tried something new here. I love your feedback. Let me know in the comments or shoot a note on LinkedIn. The Business of Tech is written and produced by me, Dave Sobel, under ethics guidelines posted at businessof.tech.

[00:10:13] If you've enjoyed the show, make sure you've subscribed or followed on your favorite platform. It's free and helps directly. Give us a review too. If you want to support the show, visit patreon.com slash MSP radio and you'll get access to content early. Or buy our why do we care merch at businessof.tech. Have a question you want answered?

[00:10:36] We take listener questions, send them in, ideally as a voice memo or video to question at MSP radio.com. I answer listener questions live on our Wednesday live show on YouTube and LinkedIn. If you've got a comment or a thought on a story, put it in the comments if you're on YouTube or reach out on LinkedIn if you're listening to the podcast. And if you want to advertise on the show, visit MSP radio.com slash engage. Once again, thanks for listening and I will talk to you again on our next episode.

[00:11:09] Part of the MSP radio network. Richtig spannend, richtig vielfältig. Das ist deine Karriere bei Kaufland.

[00:11:37] Ob Trainee-Programm, Direkteinstieg, Studentenjob oder Praktikum. Finde bei uns den Einstieg, der zu dir passt. Profitiere von einer attraktiven Vergütung, spannenden Aufgaben und individuellen Entwicklungsmöglichkeiten. Werde jetzt Teil unseres Teams. Bewirb dich jetzt unter kaufland.de slash Studenten. Kaufland – hier bin ich richtig. Und jetzt hinter dich. Serve unseres Teams. Wir müssen uns sehen. Möglich, wenn wir in die Revolution sind.