This tweet explains a bit about what CPU, TPU, and GPU are 
https://x.com/semivision_tw/status/1994376766719418437
6 Likes
IBM CEO Arvind Krishna estimates that the enormous costs of data centers cannot generate profit; for example, according to his “napkin math,” 100 gigawatts of capacity would cost approximately 8 trillion dollars and require an annual profit of 800 billion.
He also doubts that current technology would reach AGI (Artificial General Intelligence), estimating the probability at only 0 – 1 percent.
- IBM’s CEO walked through some napkin math on data centers— and said that there’s “no way” to turn a profit at current costs.
- “$8 trillion of CapEx means you need roughly $800 billion of profit just to pay for the interest,” Arvind Krishna told “Decoder.”
- Krishna was skeptical of that current tech would reach AGI, putting the likelihood between 0-1%.
4 Likes
GitLab reported strong growth and slightly exceeded expectations. Profitability clearly improved, and cash flow strengthened.
The company anticipates stable growth for the coming year and slightly better development than the market. AI-based DevSecOps solutions are also reportedly performing well, and the focus is shifting from individual sales to broader adoption of the complete platform.
https://x.com/InvestingVisual/status/1995964359487750204
Company’s own materials
2 Likes
According to former Intel CEO Pat Gelsinger, the AI boom might not be as genuine as it seems.
Major players are now funding their own growth with a somewhat “circular financing,” where investments and cloud service agreements circulate among the same companies. According to him, this can inflate the perception of demand, even though real customer money is missing.
Nevertheless, Gelsinger believes in the future of the industry, but according to him, energy consumption sets a clear upper limit.
4 Likes
C3 AI’s revenue grew steadily, with the majority of revenue coming from subscriptions and services.
The result was a loss, but “federal business” was a clear bright spot and more large contracts were secured. According to the company, customers want to accelerate AI adoption, and C3 AI sees itself as well-positioned for this.
https://x.com/earnings_guy/status/1996324998735163640
Company’s Own Materials
4 Likes
A rather interesting point regarding companies in this sector
According to the tweet, the AI boom relies on data centers, which are being built slower than planned, e.g., the power grid and capacity are years behind, but AI stocks are priced as if there were no problems, even though funding, turbines, and transmission lines often run out, etc.
https://x.com/ekwufinance/status/1997033540807004577
6 Likes
Goldman Sachs compares the current AI craze to the pre-tech bubble era, specifically 1997; investments and debt of giant companies are growing, additionally the Fed is cutting interest rates and credit risk premiums are widening, even though earnings are still strong.
The tweeter believes the imbalance is rapidly escalating because cloud giants have heavily indebted themselves, OpenAI has a large funding gap, ChatGPT’s growth is slowing compared to competitors, and the actual adoption of AI in companies is generally progressing slowly.
4 Likes
Here is Pekka’s tweet regarding AGI, or (Artificial General Intelligence.
https://x.com/vontuchman/status/1997749147437592873
5 Likes
Contrary to popular belief and what the business community in Europe claims, US AI and digital regulation is massive. For example, there are literally a 4-digit number of legislative proposals concerning these sectors under consideration at the federal and state levels in the US. The situation is downright chaotic.
Nvidia and Alphabet bigwigs have recently gone to kneel before the Great Orange One, so that he would issue an edict on federal-level regulation. But even that has been disputed, and we’ll see what comes of it. In any case, it’s a very essential question for the industry and companies.
3 Likes
Synopsys performed strongly and achieved a record result, although the full-year net profit slightly decreased compared to the previous year. The last quarter was clearly better than the previous year in terms of both official and adjusted figures.
The company emphasizes the strengthening of its position as a leader in system-level engineering solutions. For the coming year, a new record is expected in terms of revenue, as well as the integration of Ansys into operations and improved efficiency.
https://x.com/earnings_guy/status/1998861770333573367
Company’s own materials
3 Likes
This table from the tweet shows how many paid subscriptions companies have for AI services.
Undeniably, Anthropic’s rise has been tremendous this year.
https://x.com/EugeneNg/status/1999627239558906063
4 Likes
The tweet below highlights that the AI boom has driven the memory chip market into a crisis.
SK Hynix, the world’s largest memory chip supplier, expects the shortage to persist until late 2027.
https://x.com/KobeissiLetter/status/2000321021145202696
I’ll add this one here too.
https://x.com/SanCompounding/status/1995671777901772958
3 Likes
The article behind the link suggests that Microsoft and Nvidia stocks have fallen because investors fear a slowdown in data center investments and are now taking their profits.
Therefore, money may rotate into companies that have remained “cheaper,” where AI also has its own significant impact. For example, Salesforce, Adobe, Intuit, and Accenture are seen as attractive because valuations are lower and sales of AI products are only just accelerating.
Analysts expect steady growth, a good risk-reward ratio, and an upside over the coming years as the AI market matures rapidly.
Bearishness from Chanos
In this interview, Chanos breaks down why hosting GPUs is a commodity business with low returns and why the depreciation of AI chips (like Nvidia’s) creates a massive financial risk for companies like CoreWeave and Oracle. He also discusses the dangers of private credit, the accounting tricks at Live Nation, and why the “unprofitable” nature of today’s AI customers makes this cycle riskier than the Dotcom era. Recorded on December 11, 2025.
1 Like
Jukka’s AI picks from the latest TC. According to him, he aimed to pick mainly profitable, reasonably priced stocks.
And podcast recommendations.
3 Likes
I’ve been learning how to use Gemini in coding, and I noticed that they start billing quite quickly when the application generates text.
According to my applications’ usage, the price estimates for Gemini for about 400,000 words per month with different models are as follows:
Gemini 2.0 Flash $0.42 / month
Gemini 2.5 Flash $1.65 / month
Gemini 3.0 Flash $2.40 / month
I then asked for a price estimate for the same number of words using OpenAI’s language models, and these were the figures:
GPT-4o mini $0.74 / month
GPT-4o $12.38 / month
o1-preview $74.25 / month
So, using OpenAI is quite expensive compared to Gemini, and I doubt if there’s any technical advantage and thus pricing power there. Of course, it depends on the use case. My applications only summarize texts, so the cheapest one is sufficient for them.
Are there any more enlightened enthusiasts in the field here who could better evaluate those figures?
3 Likes
THE GOSPEL OF THE AI BUBBLE
The catechism of the AI boom is built upon the story of OpenAI. This messiah is followed by a group of disciples, where the role of Judas has been predicted for Oracle, for example.
Oracle has taken a very aggressive stance, and for example, its net debt is currently $112B according to TIKR, which is approximately 4 times the last 12 months’ EBITDA. This is, as I understand it, often considered a kind of pain threshold for indebtedness, and the company’s own target seems to be around 3.5x.
In addition, the company’s latest 10-Q report contains the following section:
This means that Oracle is ready to increase its aggressive stance and grow its debt load to an amount that, based on 2028 forecasts, is 5x relative to EBITDA.
Naturally, one of the capacity users is thought to be none other than OpenAI.
In the text above, it is also worth paying attention to the duration of the liability, which is 15-19 years. It is a bold aggressive move, as customer relationship durations are typically 3-5 years, as in that OpenAI deal.
Although Oracle is currently growing at a frantic pace, the latest quarterly report had a small revenue miss, which might raise questions among non-believers about how capable Oracle is of forecasting its future revenues.
In light of the figures presented in my message above, the question inevitably arises: are OpenAI’s products attractive enough for AI application developers when cheaper solutions exist on the market?
Let’s hope, however, that peace remains on earth and goodwill to all, and that OpenAI manages to gather a paying congregation around it with open hands for the collection, so that the world can benefit from the resurrection of AI. Or will it happen that the same old thing happens to AI fanatics and they are left waiting for the second coming of the messiah for thousands of years?
8 Likes
Status Update
First, ancient history
Then, the present day
https://x.com/ShanuMathew93/status/2001771811940503608?s=20
The importance of optionality in DC construction is emphasized in the MS memo. Great. Sounds familiar.
Multiple regions. Great. I wonder if it might slip into the neighboring country too?
In any case, a wonderful change in less than two months. That’s what strategic partnership brings about 
8 Likes
Language model usage is billed according to input and output tokens; for example, the Gemini 3.0 Flash you mentioned costs $0.50 per 1 million input tokens and $3.00 per 1 million output tokens. Output tokens are also consumed by reasoning tokens, i.e., the language model’s internal thinking.
A token is roughly one word. Long and inflected words consist of multiple tokens, and special characters as well as spaces are also included in the token count.
Price lists can be found on the manufacturers’ websites, and token counters are available on various sites. AI can likely code a counter for you if you ask.
Then we get to the interesting part: which model is cheap and which is expensive? How much is it worth paying for an intelligent model? Or which manufacturer offers intelligence most cost-effectively?
A somewhat accurate answer can be found in various benchmarks that also include cost data. Here is one: https://artificialanalysis.ai/ . Lots of graphs, but the most interesting one in our case is called “Cost Efficiency”. It features two different graphs:
- Total cost when running the “Artificial Analysis Intelligence Index”
- Total cost-Intelligence matrix
The matrix above provides the beginnings of an answer to your reflections on costs.
Gemini 3 and GPT-5.2 are pretty much neck and neck, but Gemini has a slight cost advantage. Gemini 3 Flash, on the other hand, is a hair less intelligent than the former but clearly more affordable. I would say Google is somewhat in the lead, especially when you start looking into other factors like response speed.
11 Likes
https://x.com/moninvestor/status/2003137454350692769?s=20
A completely unknown outfit, renewable energy, storage and colo - apparently, at least in the planning stages. Chasing capacity.
Alphabet to acquire data center and energy infrastructure company Intersect
5 Likes