DeepSeek is the fly in the broligarchs ointment. Swat or pivot?
This plucky Chinese start-up is shaking up the old world order for business tech.
A week is a lifetime in AI.
Tuesday. Jan 21.
America’s “leader” bombastically announces Stargate. This whopping $500 billion investment in AI will create 100,000 jobs to make American AI great again. The collaboration with OpenAI, Oracle and NVidia felt like spin, with no federal investment. But hey - Podiums! Techbros! I like big bucks and I cannot lie!🤑
Monday. Jan 27.
DeepSeek, a plucky upstart AI app made in China at an (alleged) cost of $5M - chump change for OpenAI's $6.6 billion raise in the monolithic age of Oct 2024 - storms to the top of the app download chart.
Tech stocks crash $1 trillion. Sam Altman’s haunting stare morphs into a sad Pierrot clown. Nvidia, the $40K-a-piece AI chip manufacturer, melts.
DeepSeek’s been on the scene some years, but it’s R1 release offered a comparable user experience and results to ChatGPT but made with very cheap chips. Like a No Frills fries versus McCain. Brand counts for little in a blind taste test.
DeepSeek’s free app offers ‘good enough’ for most user needs. And, if all is to be believed would have significantly lower running costs. Its narrow model calls on ‘specialists’ to answer each query rather than the bloated ‘generalist’ of models like ChatGPT. It’s like going out to a motion design rather than peeling the management and strategy layers of a trendy agency. DeepSeek’s smaller, focused models makes it efficient for specialized searches. ChatGPT should, in theory, be a better companion for general search.
Attacks on the challenger Chinese brand are coming in thick and fast, as to be expected. Data privacy concerns resulted in an Italy from the app stores and the American navy for “potential security and ethical concerns”.
OpenAI team hit the caffeine to put overnight crack teams on diagnosistic of DeepSeek’s open source code to quickly claim theft of OpenAI’s models. The irony. After building its $157 billion valuation on the copyright theft of creator work, the tables turned.
In this same cluster-cluck week for big tech, “techno optimistic” billionaire Marc Andreessen gleefully claimed AI will "crash" everyone's wages and then deliver an economic utopia where everything you need or want “costs pennies”. There’s a certain irony that cheap AI from out East may deliver this to the benefit, not of corporate investors, but the common entrepreneur and citizen.
Big tech’s influence on US policy may succeed in stemming the tide. For now. But the genie has left the bottle as stocks went poof.🧞Chinese marketplace Alibaba claims it has a model to rival OpenAI in the wings.
How can we cut through the noise and hype?
First, let’s consider what a costcutter rival means for the wider market. Most people, and businesses, are at the first rung of their ladder trying to work out what AI can do for data-driven decisions in the business, and how on earth best to use GenerativeAI in customer and marketing automation and content creation.
A leaner AI model could be a “good enough” solution. As Prof Ethan Mollick said a month back (Paleolithic AI times):
If AI development stopped this week we would have 5-10 years of absorbing the impact of current models on education, culture, healthcare, and business."
Although let’s not make the mistake of cultural superiority in tech solutions: tests in code and writing about Scottish football show DeepSeek’s output can rival OpenAI’s ChatGPT and Google’s Gemini.
DeepSeek’s lean model is, crucially, going to be more energy-efficient to run. So all those other capital investments the US is banking on for growth, like nuclear, would similarly take a dive.
If we adopt leaner AI models as the standard, we don’t need to wring hands to seek a trade-off between human efficiency and climate change with each AI meeting summary or prawn Jesus image.
DeepSeek’s open source model allows other developers to build atop, which could open up new economies for app creation and new services.
To get GenAI particularly embedded in business, it needs to be as seamless and limitless as the internet. It needs to be ubiquitious and widely used - why would you not at least give it a try? And an as close to zero price point will unlock access. Bedroom startups, college grads and social enterprises in emerging economies can all gain.
But let’s not kid ourselves with the mistakes of digital music’s ‘free like water’ utopia of mass-market AI. Water is rarely free. And everyone eventually pays.
It’s been called a “Sputnik” moment and “wake-up call” as geopolitical dividing lines deepening. TikTok may have been saved for now, but a Chinese app capturing every business query, consumer thought and search query in North America and Europe will have data regulators and protectionist government seeking means to shut it down.
As happened when Huawei 5G was ripped out of infrastructure, despite being cheaper and better than rivals, we could see a geo split where different AI models gain prominence in different friendly nation groups. China and USA wills and firmly on each side of this new Cold War. Europe, and UK, will need to decide their position between access and enterprise, and proprietary and security. But “Global North” AI could become uncompetitive by comparison, hindering adoption and opportunity.
What should you do today?
As always, my advice is to WAIT. You’re Working on AI Transformation. That doesn’t mean turning on a sixpence to grab each new trend. If you’re at the review stage with AI tools, then DeepSeek is too early to call for business users but worth a spin for personal users and tech testers in the business.
AI models will come and go. Many trainers who test all the GenAI tools advise not to sign up for annual contracts. But for organisations trying to affect longer-term change, choosing a model is critical to make adoption consistent and invest in the needed levels of training and security.
OpenAI spat back after the week’s blues with ChatGPT Gov - a secure model for its new pals in government.
The hype cycle spins at full tilt.
Global North AI risks becoming bloated and uncompetitive. If OpenAI insists on being the ‘one model to rule them all,’ it may lose to leaner, more specialised competitors. Microsoft Office does many things, but accountants need Sage ad designers need Photoshop instead of the (surprisingly dexterous) Paint.
In truth, we just don’t know what’s coming next.
So watch and WAIT. If you’re an innovation leader, observe and test. When the wind blows in favour of one path, you’ll be in position to make your move.
I’m exciting about this shake-up. Constraints can be a virtue. This little David could topple Goliath. But the path to tech adoption right now isn’t following a logical path. The road ahead is branching out to a crossroad, with some paths peetering to a goat track.
In the meantime, focus on what you can control: AI governance, literacy, and practical adoption. The EU AI Act isn’t waiting. Feb 1 signifies tighter new regulations like the mandatory AI Literacy requirement to train system users. EVA, my people-first tech consultancy, has programmes and resources to help with that.
Thanks for reading Rethinking The Hype Cycle. I’m sharing each week insights and a practical guide to AI and what’s next in tech for people-first leaders. Subscribe to stay ahead.