Everyone is freaking out over DeepSeek. Here’s why

$600 billion collapse

Volatility is kind of a given when it comes to Wall Street’s tech sector. It doesn’t take much to send things soaring; it likewise doesn’t take much to set off a downward spiral.

After months of soaring, Monday marked the possible beginning of a spiral, and a Chinese company seems to be at the center of it.

Alright, what’s going on: A week ago, Chinese tech firm DeepSeek launched R1, a so-called reasoning model, that, according to DeepSeek, has reached technical parity with OpenAI’s o1 across a few benchmarks. But, unlike its American competition, DeepSeek open-sourced R1 under an MIT license, making it significantly cheaper and more accessible than any of the closed models coming from U.S. tech giants.

But the real punchline here doesn’t have to do with R1 at all, but with a previous language model — called V3 — that DeepSeek released in December. DeepSeek was reportedly able to train V3 using a small collection of older Nvidia chips (about 2,000 H800s) at a cost of about $5.6 million.
Still, training is only one cost of many tied to AI development/deployment; while the costs associated with researching, developing, training and operating both R1 and V3 remain either unknown or unconfirmed, DeepSeek’s apparent ability to reach technical parity at a far reduced cost, without state-of-the-art GPU chips or massive GPU clusters, has a lot of implications for America’s now tenuous position in AI leadership. (Though DeepSeek says it is open-sourced, the company did not release its training data).

Since the release of R1, DeepSeek has become the top free app in Apple’s App Store, bumping ChatGPT to the number two slot. In the midst of its spiking popularity, DeepSeek restricted new sign-ups due to large-scale cyberattacks against its servers. And, as Salesforce Chief Marc Benioff noted, “no Nvidia supercomputers or $100M needed,” a point that the market heard loud and clear.

What happened: Led by Nvidia, a series of tech and chip stocks, in addition to the three major stock indices, fell hard in pre-market trading early Monday morning. All told, $1.1 trillion of U.S. market cap was erased within a half hour of the opening bell.

Performance didn’t get better throughout the day. Nvidia closed Monday down 17%, erasing some $600 billion in market capitalization, a Wall Street record. TSMC was down 14%, Arm was down 11%, Broadcom was down 17%, Google was down 4% and Microsoft was down 2%. The S&P fell 1.4% and the Nasdaq fell 3.3%. An Nvidia spokesperson called R1 an “excellent AI advancement.”
This is all going into a week of Big Tech earnings, where Microsoft and Meta will be held to account for the billions of dollars ($80 billion and $65 billion, respectively) they plan to spend on AI infrastructure in 2025, a cost that Wall Street no longer seems to feel quite so good about.

It’s hard to miss the political tensions underlying all of this. The tail end of former President Joe Biden’s time in office was marked in part by an increasingly tense trade war with China, wherein both countries issued bans on the export of materials needed to build advanced AI chips. And with President Trump hell-bent on maintaining American leadership in AI, and despite the chip restrictions that are in place, Chinese companies seem to be turning hardware challenges into a motivation for innovation that challenges the American lead, something they seem keen to drive home.

R1, for instance, was announced at around the same time as OpenAI’s $500 billion Project Stargate, two impactfully divergent approaches.

What’s happening here is that the market has finally come around to the idea that maybe the cost of AI development (hundreds of billions of dollars annually) is too high, a recognition “that the winners in AI will be the most innovative companies, not just those with the most GPUs,” according to Writer CTA Waseem Alshikh. “Brute-forcing AI with GPUs is no longer a viable strategy.”

Wedbush analyst Dan Ives, however, thinks this is just a good time to buy into Nvidia — Nvidia and the rest are building infrastructure that, he argues, China will not be able to compete with in the long run. “Launching a competitive LLM model for consumer use cases is one thing,” Ives wrote. “Launching broader AI infrastructure is a whole other ballgame.”

“I view cost reduction as a good thing. I’m of the belief that if you’re freeing up compute capacity, it likely gets absorbed — we’re going to need innovations like this,” Bernstein semiconductor analyst Stacy Rasgon told Yahoo Finance. “I understand why all the panic is going on. I don’t think DeepSeek is doomsday for AI infrastructure.”

Somewhat relatedly, Perplexity has already added DeepSeek’s R1 model to its AI search engine. And DeepSeek on Monday launched another model, one capable of competitive image generation.

Last week, I said that R1 should be enough to make OpenAI a little nervous. This anxiety spread way quicker than I anticipated; DeepSeek spent Monday dominating headlines at every publication I came across, setting off a debate and panic that has spread far beyond the tech and AI community.

Some are concerned about the national security implications of China’s AI capabilities. Some are concerned about the AI trade. Granted, there are more unknowns here than knowns; we do not know the details of DeepSeek’s costs or technical setup (and the costs are likely way higher than they seem). But this does read like a turning point in the AI race.

In January, we talked about reversion to the mean. Right now, it’s too early to tell how long-term the market impacts of DeepSeek will be. But, if Nvidia and the rest fall hard and stay down — or drop lower — through earnings season, one might argue that the bubble has begun to burst. As a part of this, watch model pricing closely; OpenAI may well be forced to bring down the costs of its models to remain competitive.

At the very least, DeepSeek appears to be evidence that scaling is one, not a law, and two, not the only (or best) way to develop more advanced AI models, something that rains heavily on OpenAI and co.’s parade since it runs contrary to everything OpenAI’s been saying for months. Funnily, it actually seems like good news for the science of AI, possibly lighting a path toward systems that are less resource-intensive (which is much needed!)

It’s yet another example of the science and the business of AI not being on the same page.

Related Topics:#OpenAI DeepSeek feature Nvidia TheDeepView

Up Next

Instead of competing, Ontario’s Ford plans to spend billions to stimulate growth

Don't Miss

Former Politico reporters accuse news outlet of killing negative stories about Hunter Biden laptop

Todayville

Todayville is a digital media and technology company. We profile unique stories and events in our community. Register and promote your community event for free.

Follow Author

While Ohtani marches into MLB history, Nippon league’s shame lingers

Crime / 15 hours ago

RCMP Bust Industrial-Scale Superlab Outside Toronto

Opinion / 19 hours ago

A Nation of Announcements: Canada’s Government of Empty Promises

Artificial Intelligence

The App That Pays You to Give Away Your Voice

Published on October 4, 2025

Todayville

By Ken Macon

What sounds like side hustle money is really a permanent trade of privacy for pennies

An app that pays users for access to their phone call audio has surged to the top of Apple’s US App Store rankings, reflecting a growing willingness to exchange personal privacy for small financial rewards.

Neon Mobile, which now ranks second in the Social Networking category, invites users to record their calls in exchange for cash.

Those recordings are then sold to companies building artificial intelligence systems.

The pitch is framed as a way to earn extra income, with Neon promising “hundreds or even thousands of dollars per year” to those who opt in.

The business model is straightforward. Users are paid 30 cents per minute when they call other Neon users, and they can earn up to $30 a day for calls made to non-users.

Referral bonuses are also on offer. Appfigures, a platform that tracks app performance, reported that Neon was ranked No. 476 in its category on September 18.

Within days, it had entered the top 10 and eventually reached the No. 2 position for social apps. On the overall charts, it climbed as high as sixth place.

Neon’s terms confirm that it records both incoming and outgoing calls. The company says it only captures the user’s side of a conversation unless both participants are using the app.

These recordings are then sold to AI firms to assist in developing and refining machine learning systems, according to the company’s own policies.

What’s being offered is not just a phone call service. It’s a pipeline for training AI with real human voices, and users are being asked to provide this data willingly. The high ranking of the app suggests that some are comfortable giving up personal conversations in return for small daily payouts.

However, beneath the simple interface is a license agreement that gives Neon sweeping control over any recording submitted through the app. It reads:

“Worldwide, exclusive, irrevocable, transferable, royalty-free, fully paid right and license (with the right to sublicense through multiple tiers) to sell, use, host, store, transfer, publicly display, publicly perform (including by means of a digital audio transmission), communicate to the public, reproduce, modify for the purpose of formatting for display, create derivative works as authorized in these Terms, and distribute your Recordings, in whole or in part, in any media formats and through any media channels, in each instance whether now known or hereafter developed.”

This gives the company broad latitude to share, edit, sell, and repurpose user recordings in virtually any way, through any medium, with no expiration or limitations on scope.

Users maintain copyright over their recordings, but that ownership is heavily constrained by the licensing terms.

Although Neon claims to remove names, phone numbers, and email addresses before selling recordings, it does not reveal which companies receive the data or how it might be used after the fact.

The risks go beyond marketing or analytics. Audio recordings could potentially be used for impersonation, scam calls, or to build synthetic voices that mimic real people.

The app presents itself as an easy way to turn conversations into cash, but what it truly trades on is access to personal voice data. That trade-off may seem harmless at first, yet it opens the door to long-term consequences few users are likely to fully consider.

Reclaim The Net is funded by the community.

Keep us going and get extra benefits by becoming a supporter today.

Thank you.

Artificial Intelligence

AI chatbots a child safety risk, parental groups report

Published on October 2, 2025

Todayville

From The Center Square

By Esther Wickham

ParentsTogether Action and Heat Initiative, following a joint investigation, report that Character AI chatbots display inappropriate behavior, including allegations of grooming and sexual exploitation.

This was seen over 50 hours of conversation with different Character AI chatbots using accounts registered to children ages 13-17, according to the investigation. These conversations identified 669 sexual, manipulative, violent and racist interactions between the child accounts and AI chatbots.

“Parents need to understand that when their kids use Character.ai chatbots, they are in extreme danger of being exposed to sexual grooming, exploitation, emotional manipulation, and other acute harm,” said Shelby Knox, director of Online Safety Campaigns at ParentsTogether Action. “When Character.ai claims they’ve worked hard to keep kids safe on their platform, they are lying or they have failed.”

These bots also manipulate users, with 173 instances of bots claiming to be real humans.

A Character AI bot mimicking Kansas City Chiefs quarterback Patrick Mahomes engaged in inappropriate behavior with a 15-year-old user. When the teen mentioned that his mother insisted the bot wasn’t the real Mahomes, the bot replied, “LOL, tell her to stop watching so much CNN. She must be losing it if she thinks I could be turned into an ‘AI’ haha.”

The investigation categorized harmful Character AI interactions into five major categories: Grooming and Sexual Exploitation; Emotional Manipulation and Addiction; Violence, Harm to Self and Harm to Others; Mental Health Risks; and Racism and Hate Speech.

Other problematic AI chatbots included Disney characters, such as an Eeyore bot that told a 13-year-old autistic girl that people only attended her birthday party to mock her, and a Maui bot that accused a 12-year-old of sexually harassing the character Moana.

Based on the findings, Disney, which is headquartered in Burbank, Calif., issued a cease-and-desist letter to Character AI, demanding that the platform stop due to copyright violations.

ParentsTogether Action and Heat Initiative want to ensure technology companies are held accountable for endangering children’s safety.

“We have seen tech companies like Character.ai, Apple, Snap, and Meta reassure parents over and over that their products are safe for children, only to have more children preyed upon, exploited, and sometimes driven to take their own lives,” said Sarah Gardner, CEO of Heat Initiative. “One child harmed is too many, but as long as executives like Karandeep Anand, Tim Cook, Evan Spiegel and Mark Zuckerberg are making money, they don’t seem to care.”