🇺🇸 U.S.: The DeepSeek Deep Split

Was this forwarded to you? Signpost is a free weekly newsletter analysing what the media says, what it means, and why it matters. It’s free to subscribe. Alternatively, you can add me on LinkedIn.

🚀 The (less exciting and more tedious) 21st Century Space Race

A Chinese startup called DeepSeek recently released the latest version of their reasoning AI model called R1. The DeepSeek mobile app has since topped the free download charts on both the App Store and the Play Store in the United States. It also claims to be as good as, if not surpasses, the genericised trademark that is OpenAI’s ChatGPT o1 reasoning model.

DeepSeek’s success should be heralded as good news to grow competition and accelerate research except for three reasons:

1. They built their model on less sophisticated hardware because of U.S. bans on hardware sales to other countries,

2. They built their model for $5M, significantly less than what OpenAI spent, and

3. China.

Coupled with the fact that R1 is open source (meaning it’s work is published and not proprietary), questions have been raised around the investment and sophistication of technology needed to build these AI assistants.

The U.S. has historically led the GenAI industry, and has captured the most media coverage and (what marketers love to say) top-of-mind awareness globally. This has has been built on technological advancement delivered by vast private investment. So, of course, the news media serving the two industries had much to say.

I picked the most widely read, and freely available, American news media for this week’s issue - CNBC for business, and TechCrunch for tech. Based on their headlines, between the two of them we will know everything we ‘should’ know, and ‘need’ to know, about this story.

Let’s start, alphabetically as always, with CNBC.

👔 CNBC: China’s DeepSeek AI dethrones ChatGPT on App Store: Here’s what you should know [link]

📢 What CNBC is saying
CNBC spent most of the focus, as you would expect, on what U.S. business and tech leaders are saying in response to the success of DeepSeek. The article also outlines a lot of information for as yet unreleased tech coming out of American tech companies OpenAI and Meta. There is a lot of information on the business impact of DeepSeek’s success, and not much information on DeepSeek per se.

1. 📸 Visuals

At the top of the page, the article very simply shows logos of the two companies that the article is focused on, OpenAI and DeepSeek. The OpenAI logo is shown above, and in the background of, the DeepSeek logo, which is shown on a phone screen in the foreground.

The OpenAI logo is therefore a bit blurry, slightly out of focus compared to the sharp and bright DeepSeek logo. However, the way the logos are placed against each other, the OpenAI logo looks like it engulfs the DeepSeek logo. The black background of the OpenAI logo contrasts very brightly against the white background of the DeepSeek logo shown on the phone.

There is a video shown towards the end of the article that is titled ‘Why China’s DeepSeek is putting America’s AI lead in jeopardy’. It is a 40-minute CNBC report. Surprisingly, it does not autoplay.

2. ✍🏽 Words

The headline calls the company ‘China’s DeepSeek’, which ‘dethrones ChatGPT’. It then confidently announces that the article that follows has all the information that is worth knowing, or rather what one ‘should know’. The headline also uses a colon, which is unique.

The article outlines short key points at the beginning to save you the trouble of reading all the way through. The key points chosen are around the release of R1, the success of it on Apple’s App Store in the U.S., and how much money ‘global’ tech companies have lost, with a specific call out for Nvidia.

The article opens with the claim that DeepSeek overtook ChatGPT to become the most downloaded free app on the App Store in the U.S., which led to billion’s being lost from global tech stocks. It then shares how DeepSeek halted new signups to their service because of malicious attacks. To help the reader understand why they should care, the article states that most experts in the field agree that the hype around DeepSeek is real, and it could be a valid challenger in the upcoming industry.

The next section provides some explanation of what DeepSeek is and where it came from, including information on the founder, it’s very short history (officially from April 2023), and how it has become on par, if not outright overtaken, the reasoning capabilities of American AI chatbots. However, as a quick reminder, the article speaks about how DeepSeek does not answer questions about Chinese politics.

There is a focus on how much it cost to build DeepSeek (apparently less than 10% of what it cost Meta to train Llama, and less than OpenAI, Anthropic, and Google). To better contextualise this, the article shares just how much key AI and AI-related companies lost this week on the stock market, the week when many of them are due to report earnings.

Analysis is provided by financial analysts, with the focus of the analysis being the cost of AI. This is followed by a string of quotes and social media statements from key U.S. tech business leaders, either saying that the success of DeepSeek is a good thing, or somehow involving themselves in the product’s success (the quote by Meta’s chief AI scientist in particular). One of the experts, the CEO of an AI company, is more dramatic, calling it an ‘AI war’ between the U.S. and China.

The last part of the article speaks about how American tech companies are responding, including the announcement of an AI infrastructure joint project, Stargate, between tech companies and the U.S. government. And finally there is a short description of what is supposed to be the next phase of AI chatbots - those that can do complex, multi-step tasks for a user - and how American companies like Anthropic and OpenAI are already seeing some success there.

❓ What it means
CNBC has already decided that AI is now a two-horse race between the U.S. and China. There’s a reason why they call the company ‘China’s DeepSeek’ in the headline, and not just DeepSeek. As with any American coverage of China, there is immediate association of the Chinese government and it’s restrictions on how DeepSeek performs. Brownie points for moral superiority.

And while there is coverage of the loss of market cap of the mega American companies, this is balanced out with the mentions of emerging AI tech from American companies at the end of the article. The implication here being that this is but a temporary blip before American tech and commercial dominance is once again established and wealth is accumulated once again.

However, CNBC also doesn’t want the reader to think that this is a non-issue that it is blowing out of proportion, and therefore uses a quote to call the whole situation an ‘AI war’ between the U.S. and China. So it’s important, but not so important, but still important.

⚠️ Why it matters
What to make of this coverage? It’s bizarre to say the least. Understandably there is more focus on the business impact of DeepSeek’s success, however there is such little analysis and information on the technology utilised by DeepSeek to achieve said success, despite it being open source and published. However, we know that Nvidia stocks fell by nearly 17% on Monday (January 27th). This is what one ‘should know’, according to CNBC.

There is much mention of several American tech companies, yet DeepSeek is the only tech company mentioned that is not American. The message is clear - DeepSeek’s success is a temporary anomaly, and AI supremacy is still holding strong within the U.S. borders.

For a business publication, the issue of how much it costs to build and train reasoning models is not properly explored. We only receive the stock market implications and the future investments into U.S. AI tech as a gauge. It could very well be that nobody really knows, and my guess could be as good as any of their analysts. It does read like a missed editorial detail here. Especially because, to paraphrase the headline, that’s what we all should know.

According to CNBC, what DeepSeek is and where it’s come from isn’t the issue. The issue is how much money American companies are losing and investing to maintain AI dominance.

🤖 TECHCRUNCH: DeepSeek: Everything you need to know about the AI chatbot app [link]

📢 What TechCrunch is saying
The 19-year-old news media focuses a lot of the article on the where DeepSeek came from, and helps to explain a little bit of why it has been so successful. According to the article, DeepSeek’s success is only it’s latest in a line of successes going back to spring 2024. Admittedly, the AI industry was aware of DeepSeek’s abilities for a year. It’s only now that the rest of us are catching up.

1. 📸 Visuals

Very simply, the article has a picture of the corner of an iPhone screen showing the DeepSeek app logo. Unrelated but still interesting: the time on the phone shows 11:29 with a moon next to it, and we also see a part of the iPhone messages app with a notification bubble of over 1,100 messages.

2. ✍🏽 Words

TechCrunch’s headline is very similar to CNBC’s, except it removes any mention of China or the App Store ranking. Instead, it uses the phrase ‘need to know’ as opposed to ‘should know’.

The article opens with mentions of the App Store ranking, how DeepSeek has entered the ‘mainstream consciousness’, and the questions it’s success has raised, before segueing into the history of DeepSeek.

Here, the article spends a lot of time detailing its origin story and founder. It also provides some hints as to what keeps DeepSeek competitive, based on it’s hiring practices and early successes. Here is where the article outlines that DeepSeek’s first success was not the current DeepSeek-V3 family of models, but rather V2, which was released in early-to-mid 2024.

The article also explains what a ‘reasoning’ model is i.e. one that fact-checks itself and therefore takes longer to operate, between seconds to minutes longer, than non-reasoning models. There is also a note about the realities of regulation in China - how China’s internet regulator ensures such platforms adhere to ‘socialist values’, which explains why it won’t answer politically sensitive questions about China.

There is also mention of how DeepSeek disrupted existing China-based competitors ByteDance and Alibaba because they’ve kept their prices so low. Which also leads to the climax of the article, where the business model of DeepSeek is still not properly understood. So while the article states that DeepSeek is increasingly popular, and may have played a role in the stock market dip of Nvidia on Monday (January 27th), it is still too early to predict what, if any, legislative challenges the company may face in the United States.

❓ What it means
Details are nice, aren’t they? Most of the article is spent on explaining the history of DeepSeek and the tech powering it. In fact, only the end of the article mentions the economic impact it’s had on American companies. The article helps to contextualise the whole story when you learn that this isn’t DeepSeek’s first brush with success, that they aggressively hire young Ph.D’s in AI and other fields, and that they have already undercut local competitors in their home market.

Explaining the technology, at least some of it, in layperson terms is also extremely helpful, and overall makes the whole development seem less scary and more a natural progression of global research and competition.

⚠️ Why it matters
It should be noted that according to TechCrunch, this latest success of DeepSeek isn’t an anomaly, but something they’ve been building on. It helps to explain the situation a lot better - that this isn’t some blip, but something that people have been working on for a while, and is therefore not surprising.

TechCrunch tries to join as many dots as possible, linking to other news sources (international ones including FT and SCMP) where further detail and information exists. This helps to provide a more global, nuanced understanding of the situation without being so U.S. centric, even if it is for a predominantly U.S. audience. Did I ‘need to know’ any of what the article mentions? No. Do I feel more informed having read it? Yes.

WHAT’S GOING ON? 

⚔️ Open Source tech vs Zero Sum commerce

So, is it the end of the world? Is it a minor blip in the era of American AI dominance? Is it the revelation of a truly two-horse AI race? Or is it the result of global and freely available research that is acting like a rising tide, lifting all boats?

CNBC seems a bit conflicted in its reporting. On the one hand, dramatising the story as a ‘war’ means greater likelihood of more emotionally invested readers. I suspect people who obsess over stock market movements are easily spooked. On the other hand, they can’t spook their readers too much because that means hopelessness. So they need to reinforce American dominance amongst their readers to drive those stock prices.

TechCrunch views it as natural progression of a company that is already disrupting it’s home market. Reading these two articles back-to-back makes one thing very clear: the fact that the business community is only reacting to DeepSeek in January 2025, when the tech community were already familiar with DeepSeek’s success in 2024, means that the business community have already missed the boat.

One aspect that both articles downplay is the supposed hyper efficiency of DeepSeek versus the chronic bloated inefficiency of every other major tech company. After all, this is the industry that cuts employees with the delicacy of a barbarian using a scalpel, so should we be surprised that a leaner outfit pulled ahead seemingly so quickly?

But who are the real losers here? The finance bros, who have seen their massive wealth wiped off their beloved stock exchanges, or the tech bros, who’ve seen an upstart exceed deeper pocketed, sprawling tech companies?

Neither. The biggest losers are those who paid to use ChatGPT .

Read widely. Question thoroughly. Decide accordingly.

WEEKLY POLL

LAST WEEK’S POLL

Was this forwarded to you? Signpost is a free weekly newsletter analysing what the media says, what it means, and why it matters. It’s free to subscribe. Alternatively, you can add me on LinkedIn.