AI is Eating the World

OpenAI wants to build WHAT!?? Liberation Day & DeepSeeking Missile: Open Source AI. OpenAI and the Wrong side of History. 📜⌛🏛️🏺

April 2nd, 2025. —

• Welcome to Liberation day, where tariffs super-charge the world, trade, macro economics and geopolitics with grave uncertainty!

• However on the AI News spectrum, there’s also a certain measure of uncertainty with regards to all of this AI capex spending on AI chips, infrastructure and pesky closed source models. Is it all worth it in the end?

Does closed-source even win against open-source in the end? Open-source software often benefits from a broader range of innovations due to community collaboration and transparency and those that win on cost, democratization and go-to-market virality may win and end up disrupting incumbents.

Wrong Side of History

“Shortly after DeepSeek’s model was released in January, Altman said that OpenAI was “on the wrong side of history” regarding open models, signaling a likely shift in direction.” - Wired

Outside of scale, generational funding and tons of features in ChatGPT, what is OpenAI’s moat exactly?

---

Today we have a guest post on DeepSeek’s Open-Source impact by esteemed

Kurt Cagle

of the Newsletters The Ontologist, Generation AI, and the Cagle Report, among others. I’ve been reading and talking to Kurt for years.

Whether it was an April’s fools joke or not, the Studio Ghibli stunt was bad enough but yesterday OpenAI says it too is building an “open-source” model. DeepSeek’s downloads really surged in the last quarter. But OpenAI’s, well ChatGPT’s revenue, has surged 30% in just three months.

With the usual vague bravado of a high sales entity and the freakish charm of manipulative Sam Altman, OpenAI says that it intends to release its first “open” language model since GPT‑2 “in the coming months.”

The DeepSeek Threat 🐲

DeepSeek gets more mobile and web traffic than most of OpenAI’s other other rivals, so OpenAI is going after DeepSeek with yet another PR stunt: just don’t call anything OpenAI does, real open-source innovation.

According to the Top 100 Gen AI report by a16z and indeed traffic websites, the radically open-source DeepSeek is the biggest competitor now of OpenAI.

Let's take a closer look at a quick snapshot of web traffic through March, 2025: According to data from analytics firms Similarweb and Sensor Tower let’s take a look at some of the traffic trends around Gen AI chatbots and apps:

• DeepSeek’s chatbot eclipsed 16.5 million visits in March, 2025.

• Gemini’s web traffic grew to 10.9 million average daily visits worldwide in March, up 7.4% month-over-month.

• Anthropic’s Claude reached 3.3 million average daily visits in March.

• Microsoft’s lame Copilot increased to 2.4 million — up 2.1% from February.

Is “AI Eating the World” ?

• In March 2025, openai.com saw 499 million visits, a significant increase of 251 million visits compared to the previous month. In early 2025, AI isn’t eating the world, but OpenAI is.

When such a deceptive company as OpenAI can raise $40 Billion with funds Softbank needs to take out a gigantic loan for, we have a problem on the internet. An AI Infrastructure filled world funded by third party debt financing and Middle East oil money that Softbank is known for isn’t exactly healthy or sustainable.

• Hard to know if OpenAI has capitulated or if this is just another crazy PR stunt.

• What does 500 million weekly active users get OpenAI besides perhaps $12 Billion in revenue in 2025?

• With a $300 Billion valuation, what if competitors in China are building models, apps and utility at a fraction of the price? OpenAI sound a bit confused about how to respond to DeepSeek:

• OpenAI yesterday also posted a webpage inviting developers to apply for early access to the forthcoming model. Altman said in his post that the company would host events for developers with early prototypes of the new model in the coming weeks. OpenAI needs to change their company name when they go for-profit and much of their latest funding round is contingent on their new business structure.

• I believe OpenAI’s ChatGPT is likely to increase its Plus plans prices and introduce Ads soon into ChatGPT.

• If “AI is eating the world”, what OpenAI becomes has relevance to us all and civilization at large.

“With DeepSeek, everyone’s realizing the power of open weights.” - Clem of Hugging Face.

Top Guest Works

Kurt Cagle is an extremely prolific writer, author, editor and history of tech nerd. He also doesn’t get the credit he deserves imho. Kurt’s a true historian of data science, tech and AI.

• The Taxonomist, the Ontologist and the Linguist

• All You Need is Context

• Why AI Agents Aren't Agents

• What to Study in 2025 If You Want A Job in 2030

And also what might be DeepSeek’s long-term impact on the Open-source vs. Closed-source debate? DeepSeek’s impact on China AI vs. U.S. BigTech and AI? Read our other articles on DeepSeek for more context.

For less than $2 a week get all access.

DeepSeeking Missile: Open Source AI

Written in February, 2025 by Kurt Cagle

Kurt Cagle is the editor of The Cagle Report, a newsletter focusing on knowledge systems, AI, cognitive computing, and how it impacts with the rest of the world.

As I write this, the US Tech Market (especially the Nasdaq) is in freefall, as the release over the last week of both the Chinese DeepSeek LLM and a new multimodal version called Janus Pro has sent shock waves through both the AI tech and investment communities. It’s about time.

DeepSeek has a lot of people in a snit right now. It cost $6 million dollars to train the large language model, and both its code base and its weights have been released as open-source. It is available for free. It can run on older nVidia h100 GPUs. To put this into perspective, it costs OpenAI roughly $100 million per model to train a new model, to get to the latest models is now costing users $200 a month, and, neither its source code nor the training weights (which determine the optimizations made to the model based upon its training sets) are available to anyone outside the organization.

Or, looked at another way, OpenAI spends in one week what cost the DeepSeek team a year in training, and it requires far fewer GPUs to run once it is compiled. Oh, and it reasons better than anything that OpenAI has out now or potentially will have in the next six months.

It does have a few limitations. The service, though not the raw model, will not let you ask questions about Tiananmen Square. It is very likely that the service model prompt feeds are regularly reviewed by the Chinese Communist party. OpenAI won’t talk about David Mayer, Alexander Hanff, Jonathan Turley, Brian Hood, Jonathan Zittrain, David Faber and Guido Scorza. Who are these people? No clue, but they have ceased to exist in the ChatGPT world (though David Mayer, at least, may have been jailbroken). OpenAI prompts may be monitored by specific US agencies. No one will confirm or deny this, of course. In other words, if you want to get your knickers twisted about security or bias, don’t use LLMs in the first place.

The Fallout