Today’s three AI news stories can be summarized in three words:
Fast, Valuable, Expensive.
Do AI companies still want to keep pushing faster, stronger models?
How much productivity can AI truly add to an economy?
And how much money does all the computing power behind those models actually cost?
On the morning of September 20, these three accounts surfaced simultaneously.
① Anthropic Calls for Slowdown Yet Reportedly Considering New Model
Reuters, citing three sources familiar with the matter, reports that Anthropic is evaluating launching a new AI model.
One main purpose is clear:
To respond to competitive pressure from GPT-6 Astra.
This is especially interesting because Anthropic CEO Dario Amodei had only on September 12 publicly urged the AI industry:
To slow down the pace of model capability improvements.
But just one week later, reality inside the company is:
Competitors aren’t standing still.
Reuters notes that after OpenAI released GPT-6 Astra in September, its usage in the enterprise market rapidly increased.
According to corporate spending platform Ramp, Astra accounts for approximately:
13%
of the AI spending they track for businesses.
Claude Fable accounts for roughly:
8%
Additionally, last week there was a rare shift in model spending on OpenRouter:
Users spent more on OpenAI models than on Anthropic’s—for the first time in over two and a half years.
But This Doesn’t Mean Anthropic Has Been Surpassed by OpenAI Yet
The enterprise AI market’s success isn’t determined just by a week’s traffic.
Reuters points out that by the end of July, Anthropic’s annualized revenue run rate had exceeded:
$65 billion.
OpenAI’s for the same period was over:
$40 billion.
Anthropic still has a large existing customer base in the enterprise market.
Moreover, switching AI providers for large companies isn’t as easy as individual users switching from ChatGPT to Claude on a whim.
Behind the scenes there are complexities such as:
- APIs
- Workflows
- Security Reviews
- Data Governance
- Procurement
- Internal Training
So the real question isn’t:
“Should Anthropic launch a new model?”
But rather a contradiction:
If an AI company believes fast capability growth increases risks, but competitors are taking market share, how much can it realistically slow down?
This may be the toughest challenge for the Frontier AI industry going forward.
Anthropic Has Not Officially Confirmed a New Release
It’s important to clarify:
Reuters reports that Anthropic is considering a new model,
not that a new Claude model is confirmed for release.
Anthropic declined to comment on the Reuters report.
Sources indicate the company is still assessing the new model’s safety.
What’s confirmed today is:
Competitive pressure has already entered Anthropic’s release decisions,
not that a new model has been finalized.
② IMF: AI May Boost European Productivity by About 1% Over Five Years
The second story isn’t tech companies touting AI’s importance themselves.
It’s the IMF calculating exactly:
How much AI can add to an entire economy.
The IMF prepared an analysis for the EU finance ministers’ meeting on September 18-19.
It estimates that AI could raise Europe’s productivity over the next five years by approximately:
1%
1% alone may not seem much.
But in a whole economy,
it makes a very big difference.
Because productivity measures:
How much economic output can be produced with the same people, time, and resources.
For AI to truly add economic value,
it all comes down to this.
But About 60% of Workers’ Occupations Are Highly Exposed to AI
The IMF’s analysis also includes another statistic.
In advanced European economies, around:
60%
of workers’ jobs are highly exposed to AI.
But “exposed” should not be interpreted as:
60% of people becoming unemployed.
Some jobs will be done faster with AI.
Some may see parts of routine tasks automated.
The true difference lies in whether:
AI acts as a complement to humans or as a substitute.
For example, an accountant using AI to organize data may be able to serve more clients.
But jobs primarily involving routine data preparation might require less human time.
So the 1% productivity gain doesn’t mean everyone benefits equally.
And a Third Issue: Where Does the Power Come From?
The IMF also points out that European data centers currently use about:
3%
of Europe’s electricity.
With increasing AI usage,
demand will continue to rise.
Large data center clusters in Frankfurt, London, Amsterdam, Paris, Dublin,
are already facing local power grid pressures.
This adds a cost to “AI productivity” that hasn’t been discussed much before:
AI can save office time,
but the models require:
GPUs,
Servers,
Cooling,
Networking,
and large amounts of electricity.
So it’s not enough to count only:
How much time AI saves employees.
We must also consider:
How much infrastructure is needed to support those AI capabilities.
IMF Also Warns About Europe’s Dependency on Imported Models
Currently, global frontier AI models
are still mainly led by US and Chinese companies.
The IMF warns that if Europe widely adopts AI,
but does not develop its own models, computing power, and AI industry,
it could create a new:
Strategic dependency.
Europe used to worry about energy dependence.
Now AI is entering the same type of problem:
If your productivity increasingly depends on AI, who provides that AI?
③ Nscale: Revenue Surges 1252% But Posts $1.02 Billion Loss in Half-Year
The third story best illustrates how costly the AI boom is.
UK AI cloud provider Nscale is preparing for a US IPO.
Its filing reveals:
First half 2026 revenue:
$140.6 million.
Compared with just:
$10.4 million
The same period last year.
In other words,
revenue grew by an astonishing:
1252% year-over-year.
Such growth is rare in any industry.
But the same report also reveals another figure.
Net loss for the first half of 2026:
$1.02 billion.
Compared with a loss of:
$368.9 million
The year before.
So while revenue skyrockets, losses are expanding as well.
Why Can AI Cloud Grow So Fast Yet Burn So Much Cash?
Because Nscale isn’t just a software company.
It provides:
- Compute power
- Electricity
- Data centers
- AI infrastructure
- Model training and inference infrastructure
Before customers can use AI,
Nscale must first build out massive, costly infrastructure.
The company operates in:
14 regions.
Power capacity exceeds:
10GW.
Total contracted value is over:
$103 billion.
This is very different from traditional SaaS businesses.
SaaS gaining one new customer usually doesn’t require building another data center.
For AI cloud, adding a batch of large clients means increasing:
GPUs, land,
substations, cooling, fiber,
and power contracts
all at once.
One Anthropic Contract Alone Could Be Worth $45 Billion
Nscale’s scale is evident from a single contract.
Last month, Anthropic signed an agreement with Nscale to lease:
$45 billion
of AI computing capacity at Nscale’s West Virginia data center campus.
Numbers like this show that competing in frontier AI isn’t just about hiring hundreds of researchers.
As model capabilities advance,
the physical assets behind them grow immensely.
But Nscale Faces a Risk: Customer Concentration
The IPO filing also reveals that in the first half of 2026,
Nscale’s largest single customer accounted for:
52%
of the company's revenue.
This creates another issue.
If a few major frontier AI companies:
Reduce investment,
Build their own data centers,
Switch providers,
Or slow down model development,
then Nscale’s revenue growth could also change accordingly.
So the AI infrastructure companies clearly have characteristics of:
Huge demand.
Massive contracts.
Huge capital investments.
But also highly concentrated risks.
The Real Takeaway Today: AI Faces Three Intertwined Realities
Putting these three stories together,
shows that the AI boom is entering a stage far more complex than just “who ranks first in benchmark scores.”
First force: Competition
Anthropic can talk about safety.
But if OpenAI’s new model is grabbing enterprise spending,
the company still has to consider when to launch its next model.
Second force: Economic benefits
Governments and businesses will ultimately ask:
Is AI truly increasing productivity?
The IMF is already quantifying this across Europe’s economy.
Third force: Cost
With every step forward in model capability,
there must be matching investments in power, GPUs, data centers, and financing.
Nscale’s:
1252% revenue growth
and
$1.02 billion net loss
illustrate this paradox on one financial statement.
So moving forward, watching AI doesn’t mean just asking:
“How strong is the next model?”
We also need to ask three more questions:
How fast is the market pushing it?
How much additional economic output does it truly generate?
How much money is ultimately spent to achieve that capability?
These three questions may ultimately decide how far the AI boom can go.
Today, grow alongside AI.
Learn one AI skill every day.
Save a little time each day.
Improve a little every day.
SasaDaily, growing with you.
