链研社

Meta's big gamble on AI has been a disaster; the mother of all losses. While investing in GPU computing power at least shows some results, the acquisition of the team cost $16.3 billion, which is about one-third of Meta's annual net profit. So far, no particularly impressive成果 has emerged, and Meta still lags far behind competitors in AI capabilities.
In June 2025, Meta spent $14.3 billion to acquire a 49% stake in Scale AI. Based on the latest 2026 valuation of around $12 billion, this is nearly a 60% decline from when Meta first invested.
With self-evolving agents like openclaw, the reli

Alibaba's models are leading, but in AI commercialization, they are far behind competitors. The slogans are there, but very few have actually been implemented, which is the biggest regret. Because Alibaba Cloud's commercialization has been unsuccessful, they blame the Qwen team, a team responsible for large model development, for the commercial results—it's truly absurd.
Let's see how ByteDance is doing; they are the most thoroughly AI-ized in the internet industry:
1. ByteDance's Volcano Engine consistently ranks first domestically in token call volume, far surpassing Alibaba. ByteDance's log

Anthropic Partner: 80% of AI Companies Will Fail, Only 3 True Moats Remain
1. 80% of AI Companies Will Fail
The AI industry is at a turning point from reckless funding to brutal selection. The so-called "failure" mainly refers to the following three types of companies:
API Wrappers: Companies that only layer a UI on top of OpenAI or Anthropic interfaces, without proprietary technology or unique use cases.
Functional Tools: Solutions that address a small pain point but will soon be overtaken by built-in features of large models (like GPT-5, Claude 4).
Inefficient Competitors: In the same niche

Vibecoding now reminds me of the days when I was controlled by programmers at the company, it was so painful. Everything was not good enough, and ridiculous things like waiting half a year for feature requests kept happening.
If AI had been as good as it is now back then, I wouldn't have been so frustrated that I couldn't sleep at night. In the future, using AI to write code will become a basic skill, just like using Word, Excel, and PowerPoint.

Who can accurately calculate the gross profit margin of these AI large model companies' coding packages? Anthropic's gross profit is approximately 40%.
Domestic companies like Zhipu and Minimax are open-source models; those with computing power can deploy them themselves. Therefore, it's unlikely that the pricing for coding packages will be too high, and the gross profit margin may be very low or even result in losses. Moreover, the business model of selling APIs differs from SaaS and other internet applications; each token incurs a cost, and there is no near-zero marginal cost.
Zhipu has

Investing enters a world of extreme compression; the cycle has not disappeared but has been violently folded.
1. Narrative is Justice: Assets are being tokenized
Gold/Silver are no longer just cold metals; they are also narrative tokens for de-dollarization. They complete historically significant market moves in a very short time.
Memory/Hard drives are no longer just hardware for cycles; they are the next generation after AI chips.
The first AI IPO is no longer judged solely by financial reports but replaces SaaS, disrupting traditional business models with a huge market potential.
In the sho

After updating to the new version of openclaw 2.19, I crashed the gateway. I spent the whole night trying to initialize it without success, but finally I cured my AI anxiety over the past few days.
Actually, most people who are unsure whether openclaw is useful for them can try Youdao Lobster LobsterAI first, which is more suitable for domestic use. It can be configured with Feishu, email, and also comes with commonly used skills that are ready to use out of the box. The only thing left is to configure the API( which only supports domestic models). For long-term use, I recommend purchasing a

Qwen3.5 ：Architectural innovation + open source + ultimate cost performance
Seed-2.0 ：More like a digital employee, emphasizing workflow and delivery
Kimi K2.5 ：More like a research-oriented beast, emphasizing reasoning limits and Agent Swarm
Gemini 3 Pro ：Ecosystem and long-term stable state king
GPT-5.3：Pushed programming intelligent agents to the extreme, but at a high cost
Opus-4.6：Precise agent deconstructor + dynamic thinking pioneer