Home
登錄
註冊
學習內容
Loading...
OpenAI Is Bleeding Billions. ChatGPT Is DOOMED - Video học tiếng Anh
聽力練習
聽力練習
/
Video
/
The Infographics Show
/
OpenAI Is Bleeding Billions. ChatGPT Is DOOMED
OpenAI Is Bleeding Billions. ChatGPT Is DOOMED
選擇學習模式:
查看字幕
選詞
重寫單字
Highlight:
3000 Oxford Words
4000 IELTS Words
5000 Oxford Words
3000 Common Words
1000 TOEIC Words
5000 TOEFL Words
字幕 (166)
0:03
OpenAI will run out of money - but not for the reason you think.
0:06
The creator of Chat GPT looks like the king of tech with $20 billion in revenue, but internal
0:12
spreadsheets reveal something startling. Starting in 2026, they face projected losses of $14 billion
0:18
annually. By 2029, cumulative spending could hit $115 billion. The product works, but the bills are
0:24
tied to expensive, real-world constraints. Here’s the thing that most people miss.
0:29
The massive losses lie in a simple fact - AI is not just another app, and it behaves
0:33
unlike any software we have ever built. In the traditional software world,
0:38
if you want to make a better app, you hire better engineers. You write cleaner code. It’s
0:42
a human cost. But AI doesn’t work like that - it works on something called Scaling Laws.
0:47
These are mathematical rules that govern how AI gets smarter,
0:51
and they are incredibly expensive. The rules are simple. If you want a model to be, say,
0:55
twice as good, you can’t just double the effort. You have to ramp up computing power by a lot.
1:01
It’s basically a brute-force equation. Small gains in intelligence mean massive spikes in capital.
1:06
Sounds crazy, right? But wait until you see the numbers.
1:09
Training GPT-4, the model that really kicked off this revolution, cost roughly
1:14
$100 million in computing power. That’s for one full training run, which is the process
1:19
of teaching the model from scratch. For a big tech company, that’s expensive but manageable.
1:24
The next generation - the frontier models arriving in 2026 and 2027 - plays by different
1:29
rules. Each run could cost over $1 billion. We have reached a point where a single
1:34
training session for one AI model costs more than the GDP of some small island nations.
1:40
And it gets worse. But you can’t just train it once and walk
1:43
away. You have to keep doing it. OpenAI is trapped in a cycle where they must spend these billions of
1:50
dollars just to stay slightly ahead of rivals - rivals who are giving similar tech away for free.
1:55
This creates a fundamental gap in their business model.
1:58
Their costs are tied to physical realities - electricity and silicon - which are expensive
2:02
and scarce. But their ability to raise prices is limited because there is so much competition.
2:07
The math is simple, and it is catastrophic. Explosive costs are outpacing revenue,
2:12
and the money is running out. And the financial bleed gets even worse.
2:16
To do the heavy lifting, OpenAI needs high-end AI chips, like Nvidia’s Blackwell B200s.
2:21
These aren’t your typical CPUs or GPUs - each one runs $30,000 to $40,000.
2:28
And you can’t just buy one. To train a frontier model,
2:31
you need a cluster. That means tens of thousands of these chips, all wired together
2:35
with high-speed links and liquid cooling systems. This is where the costs really start to pile up.
2:41
But the problem isn’t just buying the chips. The problem is that these chips have a limited shelf
2:45
life. Unlike a machine in a factory or a delivery truck, which might run for 20 years, AI hardware
2:51
doesn’t last. It becomes outdated the moment the next generation of chips hits the market.
2:56
And then companies are playing catch-up. OpenAIThey have to replace their entire
3:00
system of chips roughly every 18 months to 3 years just to stay competitive with Google and Meta.
3:06
Imagine a trucking company having to buy a brand-new fleet every 18 months
3:10
because the old trucks suddenly can’t deliver packages fast enough. That’s
3:14
the economic reality of AI hardware. This means the billions of dollars
3:17
OpenAI spends on hardware isn’t a long-term investment. It’s an expense that disappears.
3:22
The value of that hardware drops fast. If the cost of the chips wasn’t enough,
3:27
there is another bill that is starting to look even scarier.
3:30
The electric bill. This is best illustrated by Project Stargate.
3:34
It’s described as just a big new supercomputer, but it’s actually a $500 billion gamble.
3:40
$500 billion. Yes, that’s right. To put that in perspective,
3:44
10 gigawatts could power millions of homes - it’s the equivalent of multiple full-scale
3:48
nuclear reactors just for this one project. Why does this matter? Because the costs aren’t
3:54
going away… and the grid can’t keep up. The scaling costs aren’t going away.
3:58
They’re fixed. You can’t build the next generation of AI without this level of power.
4:03
The bottleneck isn’t just the cost of electricity. It’s the national grid.
4:07
Getting enough high-voltage transformers and grid capacity is a huge hurdle. The old
4:11
utility system can’t grow fast enough to keep up. So, OpenAI is now in the position of negotiating
4:16
for direct access to nuclear power and massive solar farms. These utility costs create a high
4:22
floor for their operating expenses. Every free ChatGPT user is
4:26
literally costing billions. And there’s no way around it.
4:29
It makes it nearly impossible to maintain healthy profits when you are trying to offer a free tier
4:34
to hundreds of millions of users. Every time someone uses ChatGPT for free, OpenAI has to pay
4:40
for the electricity and the silicon wear-and-tear. So, if OpenAI is losing billions of dollars on
4:45
chips and electricity, how are they still open? How do they pay their employees?
4:49
And that leads us to one of the most misunderstood pieces of the OpenAI story…
4:54
Its deal with Microsoft. We often hear that Microsoft has
4:57
invested billions into OpenAI. And on paper, it looks like billions came in. In reality,
5:02
it’s more like a financial merry-go-round that hides how tight the startup’s cash really is.
5:07
When Microsoft invests billions, a lot of that money doesn’t actually
5:11
leave Microsoft. They give OpenAI cloud credits instead - sort of like a gift card.
5:16
You might think that counts as real cash… it doesn’t.
5:19
OpenAI can record it as capital raised, so it looks like cash. But the credits
5:24
have to be spent on Azure, Microsoft’s cloud service, to run their models.
5:28
This effectively recycles the investment back into Microsoft’s revenue stream. It boosts Microsoft’s
5:34
cloud earnings and stock price. But here is the dangerous part.
5:38
You cannot pay your employees with cloud credits. When OpenAI hires a top researcher for $2 million
5:44
a year, they need hard cash. When they have to pay for office space or legal fees, they need money..
5:50
This creates a financial optical illusion. Microsoft invests $10 billion, but that money
5:55
doesn’t actually land in OpenAI’s account. It’s basically digital coupons that can only
5:59
be spent on Microsoft’s servers. The result is massive pressure. Every fiscal quarter,
6:04
OpenAI has to raise hard cash from other investors just to pay payroll and
6:08
cover the bills the Microsoft credits can’t touch. If the flow of new outside investment slows down,
6:13
OpenAI faces a cash flow crisis. They might have plenty of computer time,
6:17
but not enough hard currency to keep their team from leaving for rival companies.
6:21
Despite all these costs, investors kept pouring money in. In March 2025,
6:27
OpenAI managed to raise $40 billion - the largest private funding round in history, even bigger than
6:32
the IPO of the oil giant. Saudi Aramco. But here’s what’s really odd about it.
6:37
Saudi Aramco has hundreds of billions in revenue - and more importantly, it has real,
6:42
tangible assets. Oil reserves you can measure and sell. OpenAI is a startup with no profits,
6:48
burning cash at a rate of billions a year. Its value is mostly intellectual
6:52
property… which anyone can try to copy. So, what does this mean for the long-term
6:56
survival of OpenAI? The answer will surprise you. Investors are pouring money in based on the
7:02
promise of a market that doesn’t fully exist yet. For OpenAI to be worth a trillion dollars,
7:08
it can’t just be impressive - it has to replace dozens of cheaper tools companies
7:12
already use. Right now, most businesses spread their AI budgets across multiple
7:16
smaller providers, not one giant system. OpenAI is building something massive and
7:21
expensive, betting that eventually everyone will need it. But right now,
7:26
there’s no guarantee of that demand. And that leads us to the risky business model.
7:30
In software, companies survive by making it hard for customers to leave. Salesforce
7:34
does this because moving all your data is a huge pain. Netflix does it because they own
7:39
shows you can’t watch anywhere else. OpenAI is discovering a hard lesson.
7:44
Users are mercenary. If Google’s Gemini or Meta’s Llama offers a similar answer
7:49
for cheaper, they leave instantly. About 75% of Open AI’s revenue comes
7:53
from consumer subscriptions, but the number of cancellations is rising.
7:58
Once the novelty fades, most users won’t pay. Big business is even more skeptical. Only
8:03
about 20% to 30% are sticking with OpenAI’s API long-term. Many are choosing open-source models
8:09
like Llama to keep data private and costs down. With nothing keeping them tied to OpenAI - no
8:14
built-in network, no way their data is stuck - they can jump to another provider overnight.
8:20
And the competition is just as deadly as OpenAI’s own cash burn.
8:24
Meta’s decision to release the Llama models for free was not an
8:27
act of charity.It was a tactical strike. When Mark Zuckerberg gives everyone access
8:31
to their top-of-the-line AI for free, he effectively sets a ceiling on what OpenAI
8:36
can charge. Meta can burn cash on open-source models because they’re using the tech to
8:40
improve ads on Instagram and Facebook. Their business isn’t selling AI - it’s selling ads.
8:46
OpenAI doesn’t have that luxury. Their only product is the AI itself. They’re
8:50
fighting to establish themselves while their competitors aggressively undercut the market
8:54
to keep them from gaining ground. And the clock is ticking.
8:58
OpenAI is squeezed from all sides. On top, giants like Microsoft and Google
9:02
with practically unlimited cash. On the bottom, lean competitors like Anthropic and Mistral.
9:07
Anthropic runs a much more efficient operation, focusing on safety and enterprise reliability
9:12
with a much lower burn rate. Meanwhile, Google DeepMind keeps stealing talent, forcing OpenAI to
9:18
offer massive stock-based pay packages. Those only work if the company’s valuation keeps climbing. If
9:23
it stalls, the researchers - the company’s only real asset - could walk out the door.
9:27
As if burning billions, fighting competitors, and losing talent weren’t enough, regulators
9:32
in Washington and Brussels are circling. In early 2026, the FTC and European Union
9:37
intensified their antitrust probes into the Microsoft-OpenAI partnership. Regulators
9:42
are checking whether Microsoft’s investment is actually a de facto
9:45
acquisition designed to skirt merger laws. If they decide to limit the power Microsoft
9:50
has over OpenAI, or force a split, it would cut the startup’s financial lifeline.
9:55
And then there’s the mounting geopolitical friction.
9:58
Export controls on AI chips are shrinking the global market,
10:01
while new AI safety regulations are creating a massive compliance burden. OpenAI now needs armies
10:07
of lawyers and safety researchers - roles that are costly and generate zero revenue.
10:12
The danger becomes clear when you look at history. Uber lost billions before its initial public
10:17
offering, or IPO. But it was building a physical network in thousands of cities.
10:22
Tesla struggled for years, but it was building factories and a global charging
10:26
network - something real that competitors couldn’t copy overnight. OpenAI? It’s burning billions with
10:32
no real network or physical assets to lean on. OpenAI’s production is all about raw computing
10:37
power - the expensive chips that mostly come from Nvidia. Unlike Tesla or Uber,
10:42
OpenAI’s product loses money every time someone asks it a complex question.
10:46
And there’s nothing stopping users from leaving tomorrow.
10:49
The company is now effectively betting everything on a single, desperate timeline.
10:54
They’re racing to build Artificial General Intelligence (AGI) AI that can think and learn
10:59
like a human - before the bank account runs out. This isn’t a standard software business strategy
11:04
anymore. If OpenAI can build a model smart enough to do the work of a human expert in any field,
11:09
their current cash burn wouldn’t matter. Revenue could, in theory, skyrocket. They’re picturing a
11:14
world where their AI doesn’t just summarize emails - it replaces entire departments,
11:18
handling corporate taxes, writing complex code, and planning strategic business
11:22
moves at super-human speed. Reach that milestone, and they could charge a premium
11:27
that covers any debt, no matter how massive. If OpenAI is losing $14 to $17 billion a year,
11:33
every month of delay costs over a billion dollars. If the breakthrough to AGI takes 5 years instead
11:39
of 2, they’d face a funding gap of nearly $100 billion just to keep the lights on.
11:44
And no investor can fix that overnight. So, what happens when the money runs
11:48
out? You might expect a dramatic crash… but the reality is different.
11:52
The most likely outcome is not a dramatic crash or a bankruptcy filing, but a quiet absorption.
11:56
By mid-2027, based on current projections, the cash reserves raised in the 2025 rounds
12:02
will be nearly empty. At that point, OpenAI will face a choice:
12:05
raise another massive round at a lower valuation - crushing their employees' stock options - or sell.
12:11
Microsoft is the natural - and maybe the only - buyer. They already host OpenAI’s systems on Azure
12:16
and have deep integration with the software. More importantly, Microsoft has over $80 billion in
12:21
cash reserves, making them one of the few entities on Earth that could sustain OpenAI’s burn rate.
12:26
For Microsoft, this is the crown jewel - the engine of the next computing
12:31
era. For investors, it’s a fire sale, but one that buys survival.
12:35
This is the end of the startup frontier. OpenAI proved scaling works - but only if
12:39
you have a nation-state-sized budget. The AI revolution has gone industrial, where
12:44
success is measured in acres of data centers, not lines of code. OpenAI started the trend,
12:49
but it doesn’t have the resources to compete alone. And now, the independent pioneer is
12:54
likely to be absorbed by a larger corporation. Now go check out ‘Real Reason Humanity Is NOT
12:59
Ready for AI Superintelligence’. Or click on this video instead.
