Web3投研江南雨

About the AI ​​public chain tau.net (token name AGRS)

Tau
Tau is a universe built from your ideas, suggestions and knowledge. It is the first-ever decentralized blockchain social network capable of gaining consensus from its users and updating its own software in real time, allowing all users to control what the software does for them.

All in all, Tau can aggregate the ideas, suggestions and knowledge of various users, and upgrade its own code in real time, using logical artificial intelligence to bridge the gap between humans and machines.

Notice the difference? If you have recently learned about artificial intelligence and used ChatGPT, you will know that most AI uses machine learning, which relies on huge Internet data sets and makes it difficult to grasp the correct meaning and knowledge of interpersonal communication.

What about logic-based Al? It is more effective than machine learning in:

1. Explainability: Logic-based AI provides clear and transparent explanations of how decisions are made, which can be useful in certain applications, such as legal or medical diagnostics. In these application scenarios, accountability and explainability are important.

2. Predictability: Logic-based systems are deterministic, meaning that given the same input, they always produce the same output.

3. Efficiency: For simple tasks, logic-based AI can be faster than machine learning algorithms because they require no training to make decisions based on a small number of well-defined rules.

4. Adaptability: Logic-based AI can be easily modified and updated to adapt to new knowledge or environmental changes.

These advantages form Tau's moat. Artificial intelligence using logic has uniquely created a model of communication between people with machines as the mediator.

A team of PhDs and professors with expertise in artificial intelligence, logic and mathematics has been working on Tau since 2015. Founder and CEO Ohad Asor is a software developer and machine learning expert with a background in mathematics and experience at GE Healthcare.

$SKHYNIX
Seeing the U.S. market decline, I went short on Hynix and closed the position at the next day’s open.

Nomura Securities
SK hynix currently holds about 25% of global memory revenue…… If these Nomura estimates are correct, and SK hynix maintains its current market share…… then this would mean that SK hynix’s revenue in 2030 will be about $918 billion (US dollars).
From 2026 to 2030…… SK hynix will generate $1.8 trillion in net profit, which is 80% more than its current market value.

$SKHYNIX

Nomura has set an undeniable number for the memory market.

It expects the entire market to be about $1 trillion in 2026. Then it will reach $3.7 trillion by 2030. That means growth of more than 3.5 times in four years.

Data centers will account for 83% of the total by 2030. AI data centers will become the largest single share.

Growth is not smooth.
In 2024 it is about $185 billion.
In 2025 $270 billion.
Then in 2026 it jumps to $1 trillion, after which it keeps rising.

$SKHYNIX

Memory Industry Information Notes

260909
Rumor: Changxin's HBM yield is only 25%. Therefore, Changxin does not pose a threat to other HBM manufacturers.

260908
Rumor: The manufacturer's memory inventory is insufficient for more than 10 days.

260908
It is expected that in 2027, global DRAM production will increase by only a little over 20%.

$SKHYNIX

This morning, go long on $SKHYNIX on a short-term basis. After about 20 minutes from the opening, close all positions.

Last Friday, storage stocks on the U.S. stock market surged. So, I went long at the opening today at $SKHYNIX and made some profit

Nomura released its view on the global memory market and stated the things we all already know

Here are the key points from Nomura’s view on the global memory market

Global memory:
- Manufacturers claim that severe supply shortages will continue
- LTAs will restructure their businesses to make them visible and stable
- Customers are willing to accept unattractive terms to secure supply
- Memory spec degradation is inevitable due to supply shortages
- By 2027, data center demand is expected to be met only 70%, and non-data center demand about 50%
- Nomura believes that, due to shortages, $NVDA cannot broadly adopt high-stack HBM
- Challenges from Chinese players are unlikely to pose a significant threat
- Interest rates and foreign exchange are threats
- The capex train is expected to keep rolling

LTAs:
- There is significant uncertainty because negotiations are still ongoing, forcing institutions to make some assumptions
- LTAs will account for about 50-70% of total $DRAM NAND and SSD sales. Mainly targeted at CSPs (10-20 customers)
- The remaining 30-50% will be sold directly to the market at spot prices
- LTAs mainly run for 5 years, with strong agreements for the first 3 years and flexibility for the next 2 years
- Prepayments account for about 20-30% of revenue. The penalty for customers breaking commitments is equivalent to about 1 year of purchasing value
- In short, LTAs provide tremendous visibility, pricing power, and cash flow for the business

Investment perspective
- Nomura believes memory stocks can maintain strong growth
- It believes a rebound and growth are only a matter of time
- Total DRAM
- $DRAM wafer capacity utilization is about 100%
- The share of data centers in memory revenue is expected to rise from 40% in 2025 to 83% in 2030
- Memory revenue is expected to reach $3 trillion by 2029
- Samsung Electronics at 5.2x FY26E P/E and 2.2x P/B. At 2.9x FY27E P/E and 1.4x P/B
- $SKHY at 4.8x FY26E P/E and 2.8x P/B. At 3.1x FY27E P/E and 1.7x P/B
- Both companies are rated "Buy" based on improving margins, free cash flow, and the likelihood of higher shareholder returns

Samsung’s target price is set at 670K (167.5% upside), and SK Hynix’s target price is set at 4.7M (191.4% upside)

#美股 #TradFi
$SKHYNIX

Micron’s own HBM architect, on site at the Hot Chips conference:

“Computing power grows by about 3x every two years. Although HBM has solved many bandwidth issues and improved bandwidth, it still lags behind demand, growing only about 2x every two years. Therefore, the memory wall problem still exists, and in fact, it may be getting worse.”

— Raghu Sreeramaneni, Micron Fellow, HBM Design Architecture

This is a structural gap that keeps widening with each generation of products. The market has not yet realized this difference.

#美股 $SKHYNIX $MU

Memory trader NexSTI President Yamada told the Nikkei that by the end of 2026, memory prices could rise by as much as 50%.

The driving force is AI. First the cloud, then more edge devices. Usage is still at an early stage, so he believes demand will continue.

His intuition about the tightness: demand is 100, supply is 40 to 60.

#美股 $SKHYNIX

SK Hynix (000660) is carrying out one of the largest company share repurchase programs in history, amounting to 3.3% of the shares outstanding.

They still have 26 days left, buying back 650,000 shares per day.

Volume is drying up, which means sellers are running out.

SK Hynix may surge wildly. It’s time for Koreans to HODL so they can come out of the sauna and go back home.

#美股 #TradFi

Goldman Sachs released an interesting report.

To put it simply: once humanoid robots become widespread, companies that produce “memory semiconductors” will make money.

In the semiconductor content of a single robot, by price, up to 40% can be storage, and in terms of money it amounts to more than $600–$800 per unit. Also, some of it needs high-performance “HBM,” and in practice only three companies can produce it: Micron, Samsung, and SK hynix.

In short, the more humanoid robots there are, the easier it is for these three companies to earn money, right?

#USStocks

Leopold Aschenbrenner’s Complete Updated Holdings

The maximum position is SNDK MU

Top tier:
• SanDisk $SNDK [Long, 28.1%]
• Micron $MU [Long, 27.6%]

Second tier:
• Bloom Energy $BE [Long, 9.4%]
• Taiwan Semiconductor $TSM [Long, 6.3%]
• Nebius $NBIS [Long, 6.1%]

#US stocks

As the memory market expands, by 2027 it will exceed $1 trillion; in 2028, it is expected to hit a “trough”】

On August 24, 2026, the U.S. research firm Gartner released its latest forecast for global semiconductor sales. Global semiconductor sales in 2026 are expected to grow by 92% year-on-year, reaching $1.6 trillion (about ¥254 trillion).

In particular, growth in memory sales is especially notable. Memory sales are expected to exceed $1 trillion in 2027 (about ¥159 trillion).

Source: NIKKEI TECH Foresight
#US stock

This morning right after the South Korean stock market opened, things looked promising. I went long on SK Hynix, and within ten-odd minutes I made over 500 U—pretty decent $#$SKHYNIX

Introducing a hardcore tech project.
Formerly known as Maidsafe, it's now rebranded as Autonomi.

The project name is Autonomi, and the token is ANT.
The ANT token is issued on the Ethereum Layer2 Arbitrum One.

Even though the mainnet is live, the product isn't fully polished yet.
Currently, it's only listed on MEXC and a few tier-three exchanges. Marketing efforts are nearly non-existent.
The team claims that once the product is refined, they'll ramp up marketing and get listed on major exchanges.
A standout feature is that you can mine using a home PC.
They raised tens of thousands of BTC back in the day, so they're well-funded.
Right now, they are working on expanding functionalities into the Agent space.
This includes messaging protocols between Agents, AI data storage, a Trust Layer, and collaboration among multiple Agents.

Bitcoin is likely to go down.
The Nasdaq surged three days ago, with small increases two days ago and yesterday. It should decline in the coming days.
Moreover, the surge in oil prices has a significant impact on the stock market.
Bitcoin will follow the decline of the Nasdaq index.
#比特币 $BTC

Some people say that Bitcoin has already dropped to the right level and should start rising.
But I don't think so.

There are three types of people who buy and sell Bitcoin:
(1) People who need Bitcoin not for investment purposes, but to buy or sell.
(2) People who treat Bitcoin as something similar to gold for investment. Building positions or closing positions.
(3) Speculators.

If there were no speculators, at the current price level, Bitcoin should be rising.
But speculators manipulate the market.
The market is unreasonable; just because Bitcoin is already very cheap doesn't mean it won't drop.
Currently, although the price of Bitcoin is significantly lower than it should be, the pace of capital entering the market is slow.
At this time, large speculators can easily make money by shorting. Speculators don't care whether the Bitcoin price is reasonable; as long as they can make money by shorting, they will maliciously crash the market.
Therefore, I think there is a high probability it will go towards $45,000.
#比特币 $BTC

#AI $TAO Well-known VC fund Pantera says that this year's biggest opportunity in the WEB3 track has also seen explosive growth points:

The year of AIAGENT, it is not that Crypto needs AI, but rather AI needs Crypto,

To be more precise: what AI truly needs is a payment system between machines.

1️⃣ Payment issues:

AI Agents will certainly require a large amount of interaction, the trading market for AI assistants may reach trillions in the coming years, but if automatic trading, API calls, purchasing computing power, and buying data are needed, the traditional banking system is almost unusable.

To achieve permissionless + programmable payments, currently only crypto can solve this.

2️⃣ Global settlement issues

AI is a global network,

At that time, your AI agent in Germany and my AI agent in South Korea, the payment system must be part of the international financial system.

Blockchain is inherently a cross-border settlement layer.

3️⃣ Digital scarce assets

The biggest problem in the AI era is: proving ownership of data, models, and computing power.

Token / blockchain is the solution.

Jane Street pays traders $400k to achieve this
Top quant experts at Polymarket use the same method
Immediately stop trading when calibration fails
**Brier Score** can accurately tell you when to stop.
Formula: \bm{\sum (\text{predicted\_probability} - \text{actual\_outcome})^2}
• Perfect prediction = 0.00
• Random blind guessing = 0.25
Why Wall Street values it so much
• Citadel tests this in interviews.
• Top traders at Polymarket self-check every 50 trades.
Even if you have a 70% win rate, your probability calibration can still be terrible.
For example:
You claim to have 90% confidence in 10 trades.
If you only win 6 in the end, you were overconfident.
Your so-called 90% confidence is actually only 60%.
The Brier Score will immediately expose this illusion.
Discipline of top experts
• Quant firms will fire traders who cannot calibrate probabilities.
• The same goes for the top 20 traders at Polymarket: they track the deviation of predicted probabilities from actual outcomes.
• When the Brier Score starts to rise, they will stop trading.
This is not out of fear, but because math tells them: your edge has disappeared.
The dividing line between amateurs and professionals
Imagine this: you made $5000 in a week, but your Brier Score jumped from 0.12 to 0.19.
• Quant experts: decisively exit.
• Amateur retail traders: continue trading and lose back all their earnings.
Why do most traders ignore this?
Because tracking the Brier Score feels like doing extra homework. You might think, "Whether I win or lose, don't I have a sense of it?"
Wrong.
• Calibration is poor but you won: it means you were just lucky.
• Calibration is good but you lost: it means you were just unlucky.
Only the Brier Score can tell you the truth.
The real moat
Trading is not about "predicting correctly," but about **"knowing when you predicted incorrectly."**
The moment your prediction no longer matches reality, you should stop betting. Quant experts have established a system that measures not only profit but also their own accuracy. #预测市场