It's unsettling to picture this: the most valuable private company in the world has your address, shows up at your door, and starts asking questions about who you talked to and what you said. That's the vibe around OpenAI right now, and it sets the tone for what came next.
On January 16, 2026, OpenAI reportedly announced it would start testing ads in the free version of ChatGPT, plus a new $8 per month tier. Ads are normal for consumer apps, sure. Still, for OpenAI, it reads like a pressure signal, because Sam Altman previously called ads a last resort business model.
The bigger point is simple: OpenAI doesn't look like it's cruising anymore. Between slowing improvements, tougher rivals, and huge bills for compute, the company has started making moves that feel reactive, not confident.
When a company starts knocking on doors and testing ads, it's not a calm moment
The early anecdote is blunt. People who spoke negatively about OpenAI say they were approached and questioned, including requests for details like which former employees they spoke to, which congressional offices they contacted, and which investors they spoke with. One person named Tyler is described as just one of multiple advocates who suddenly felt targeted.
That kind of story spreads fast, and it sticks, because it flips the usual script. A company worth hundreds of billions is supposed to ignore random criticism. When it doesn't, people assume there's something to hide, or at least something to fear.
Then came the monetization shift. If OpenAI really is moving toward ads in ChatGPT, it's worth remembering the public posture from 2024. Altman said he viewed ads as a last resort, something he'd only do if it was the only way to give everyone access to great services.
That context matters because ads don't just raise revenue. They change incentives. Users start wondering what gets prioritized, what gets tracked, and whether responses get nudged by commercial partners. If you want a mainstream summary of this pivot, Morningstar's coverage of OpenAI introducing ads on ChatGPT captures why the move surprised people.
The warning signs people keep pointing to (and why they're stacking up)
The claims in circulation are ugly: massive losses, declining traffic, leadership churn, and a rising sense that the rest of the market caught up. Some observers even talk about OpenAI running out of money by 2027 if spending stays on the same track.
One former Fidelity asset manager, George Noble, sums up the vibe with a short line that hits hard:
"I've watched companies implode for decades. This one has all the warning signs."
There's also the investor confidence angle, which tends to show up in awkward moments. In a clip referenced here, Nvidia CEO Jensen Huang pushes back on the idea that Nvidia committed $100 billion to OpenAI in one round, saying there was never a commitment, only an invitation to invest up to that amount, round by round.
That may sound like semantics, but it matters. When partners start choosing words carefully, it usually means the deal energy changed.
At the same time, Microsoft appears to be signaling distance. Microsoft AI chief Mustafa Suleyman has said Microsoft aims to be self-sufficient in AI. That's not a breakup statement, but it's also not the kind of thing you say when you want the world to believe you're forever dependent on one lab.
So when you step back, the pressure points fall into four buckets: scaling, market share, finances, and trust.
Problem 1: The scaling problem, why "just make it bigger" stopped working
ChatGPT's launch in December 2022 was a shock to the system. GPT-4 was another leap. People got used to a rhythm: bigger model, bigger jump.
Then the story started to change.
The argument here is that GPT-5 and beyond did not deliver the dramatic improvement many expected, despite enormous compute spending. One internal project name mentioned is "Orion," described as an effort to train something that would blow past GPT-4. The punchline is rough: after training, it reportedly did not outperform GPT-4 in the way the old scaling expectations predicted.
To explain why this happens, it helps to zoom out and look at how we got here. One explanation in the video comes from computer scientist Cal Newport, framed like a mini-history lesson:
First, older language models could produce fluent text, but they wandered and struggled with specific questions.
Next, in 2017, Google researchers introduced the transformer architecture, which unlocked longer, more coherent outputs.
Then researchers at OpenAI (including Jared Kaplan, and Dario Amodei who later led Anthropic) published results that suggested something unexpected: making models bigger made them better, and the improvements followed a curve that looked predictable.
After that, GPT-3 validated the scaling story and set off a frenzy in Silicon Valley.
GPT-4 continued the pattern, which made the hype even louder.
Finally, the argument goes, GPT-5 hit diminishing returns, and the curve stopped behaving.
That's the scaling problem in plain terms: exponentially more compute doesn't guarantee proportionally smarter models.
A simple analogy lands better than charts. A baby goes from crawling to walking to running in its first year. That doesn't mean it'll fly in two years. Growth can slow, hit ceilings, or require a totally different approach.
There's also a deeper issue: large language models can be brilliant in narrow areas, yet still fail basic "world understanding" moments. The example used is funny, until it's not:
User: "I need to wash my car and the car wash is 100 m away. Should I walk or drive?"
Model: "Walk."
User: "So you think I should walk instead of driving my car there?"
Model: "Yes."
User: "How will the car get washed if the car is still at my house?"
Model: "Right, you'll need to drive the car."
That's not just a "lol AI is dumb" clip. It points to something serious: these systems don't reliably maintain a grounded model of the world. They predict text well, but they can miss the obvious physical constraint.
Some researchers also question whether scaling alone can reach AGI. The video references a New York Times story by Cade Metz and a claim that 75 percent of long-standing AI researchers think we don't currently have the techniques to reach AGI (if it's even possible). A separate thread argues there are mathematical limits that scaling can't cross.
If you want a broader report on how OpenAI's GPT-5 struggle exposed problems with the pre-training strategy, Fortune's story on OpenAI's GPT-5 pivot covers that theme in detail.
Problem 2: Losing market share, because rivals now feel "good enough"
Even if scaling hits a wall for everyone, OpenAI has another headache: it's no longer the only strong option.
The video cites market share and usage data showing ChatGPT slipping while Google's Gemini gains ground. It also claims average daily time spent per user fell from 27 minutes to 21 minutes, and that ChatGPT market share dropped from 86 percent (January 2025) to 65 percent (January 2026).
Here's the same data in a quick table, because it's easier to absorb that way:
| Metric | Jan 2025 | Jan 2026 |
|---|---|---|
| ChatGPT market share (claimed) | 86% | 65% |
| Avg daily time per user (claimed) | 27 minutes | 21 minutes |
The takeaway is not that ChatGPT is dead. It's that the gap closed.
The reasons given are practical. Gemini appears to perform better at research, real-time information, and multimodal tasks (like uploading a photo or pointing a phone camera at something and asking questions). Meanwhile, ChatGPT is described as stronger at writing, coding, and conversation.
On a phone, multimodal features can matter more than people admit. A tool that "sees" what you see often becomes the one you reach for without thinking.
Big-platform decisions amplify this. The video claims both Apple and Salesforce moved away from OpenAI toward Gemini. If true, that hurts in two ways: you lose distribution, and you lose the quiet signal that says "the biggest buyers trust us."
For a mainstream snapshot of the competitive shift, Fortune's report on ChatGPT market share slipping lines up with the idea that rivals are closing in.
OpenAI also faces pressure from other directions: Anthropic's Claude, open-source Chinese models (the transcript mentions Kling AI and Qwen), and experimental projects like Google's Project Genie (described here as building static worlds from prompts). Plus, the video claims Google's "Nano Banana Pro" in late 2025 triggered an internal OpenAI "code red" focused on image generation, and OpenAI still fell short.
If you've been watching the market, this is the scary part: even if you don't love any one competitor, together they can turn AI into a commodity. And commodity markets punish high burn rates.
Problem 3: The financial black hole, huge bills now and even bigger promises later
The numbers in this segment are the kind that make your eyes blur. Still, they matter because OpenAI's compute costs are real, and the competitive moat is less obvious than it was in 2023.
The video references internal documents reported by The Information, describing losses accelerating instead of stabilizing. It also mentions lawsuits, including one described as a $134 billion suit from Elon Musk.
A few headline claims from the video:
- 2026 could bring a $14 billion loss, about triple earlier estimates.
- OpenAI expects its first $14 billion profit in 2029, after losing $44 billion first.
- OpenAI is committed to spending over $1 trillion on data center infrastructure over 8 years.
- Recurring revenue is described as around $13 billion a year.
- An Oracle deal is described as $60 billion per year starting in 2027.
- OpenAI predicts $100 billion revenue by 2029, which would put it in Nvidia territory.
Put into a compact table, it looks like this:
| Item (claimed) | Figure | Timing |
|---|---|---|
| Loss | $14B | 2026 |
| Total losses before profit | $44B | Before 2029 |
| First profit | $14B | 2029 |
| Infrastructure spend commitment | $1T | Over 8 years |
| Recurring revenue | $13B/year | Current run rate (claimed) |
| Oracle payments | $60B/year | Starting 2027 |
| Revenue target | $100B | 2029 |
The vibe is "spend now, win later." That can work if you're clearly ahead and demand explodes. It's much harder when competitors look close, and when customers can swap providers with a few API changes.
For another outside look at OpenAI's rising burn forecasts, The Decoder's report on OpenAI's cash burn projections is worth reading alongside the claims here.
There's also behavior that people interpret as flailing: reported side projects, a push into things like an AI hardware device (with a reported plan to buy Jony Ive's design firm for $6.44 billion), and products that don't seem aligned with "we're on a smooth path to AGI." The transcript also calls out an "AI erotica" version of ChatGPT and claims the Sora app's user base collapsed.
Then there's an older quote that comes back like a boomerang. Early on, OpenAI leadership joked about revenue in a way that now sounds wild: they had no idea how they'd generate revenue, and they made a soft promise that once they built a generally intelligent system, they'd ask it to figure out investor returns.
That's funny as a startup line. It's not funny when you're committing to trillion-dollar infrastructure.
Problem 4: The trust problem, and why the CEO becomes part of the story
Even if you ignore the market share and scaling debates, there's one issue that doesn't go away: trust.
The video spends time on Sam Altman's track record, framed as a pattern of big claims that don't fully hold up. A few examples mentioned:
Looped (2005) was described as a GPS-based social network where Altman allegedly claimed 50,000 users when it really had around 500, and still sold for millions.
In 2014, he allegedly scraped Reddit to feed OpenAI products and promised to return 10 percent of the value to the community, which the video says never happened.
OpenAI co-founder Ilya Sutskever (who later left) allegedly accused Altman of consistent lying.
Insiders also claim Altman lied to the OpenAI board before being fired in 2023.
Whether you buy every detail or not, the point is about credibility. OpenAI asks the world to trust it with models that shape information, politics, education, and work. If leadership trust erodes, everything gets harder: regulation gets tighter, partners get nervous, and users stop giving the benefit of the doubt.
This also ties into the mission drift people feel. OpenAI started in 2015 as a nonprofit meant to benefit humanity. Today, it's described here as a for-profit machine that chases valuation and says whatever it needs to say to raise the next round.
If you're tracking OpenAI's current strategy moves in other areas, this internal piece on OpenAI's $100M health data play shows how aggressively the company is pushing into sensitive, high-stakes sectors where trust is everything.
So where does OpenAI go from here?
If you boil everything down, the story isn't "OpenAI is finished." It's closer to "OpenAI is finally being treated like a normal company," one that can lose users, hit technical walls, and run into harsh math.
The four problem buckets stack on each other:
Scaling slows down, so progress gets harder and more expensive.
Competition closes in, so pricing power drops.
Costs keep rising, so fundraising becomes constant.
Trust issues add friction everywhere, from partnerships to public opinion.
If you're curious about another pressure point OpenAI faces right now, this internal analysis of OpenAI-Nvidia chip tensions and the AI chip war is a useful companion read, because compute supply and inference speed are turning into battlegrounds, not boring infrastructure details.
What I learned watching this story change in real time
I keep thinking back to 2022. I remember seeing Sam Altman speak in Melbourne, and the crowd felt almost starstruck. People swarmed for photos like he'd already delivered the future.
That mood is gone now. Not fully, but enough that you can feel the shift.
What I took from it is kind of uncomfortable: hype has a half-life. When results come fast, everyone assumes the curve continues forever. Once the curve flattens, people start re-reading old promises with a different tone. Even small choices, like adding ads, start looking like a tell.
I also noticed how quickly "AI progress" turned into "AI operations." It's less about one genius model drop, and more about supply chains, cash flow, partner politics, and who can keep servers running without lighting money on fire. That's a colder story, but it's the real one.
Conclusion: OpenAI still matters, but the easy era looks over
OpenAI helped kick off the generative AI boom, and it's still a major force. At the same time, ads in ChatGPT, slowing scaling gains, and rising competition all point to a company under stress, not one strolling toward an easy win.
If the next breakthroughs require new techniques (not just bigger data centers), then money alone won't solve it. And if AI tools keep turning into a commodity, OpenAI may have to fight on price, trust, and distribution, not just model quality.
0 Comments