Last week, Menlo Ventures partner Deedy Das posted on X, triggering a "spiritual earthquake" in Silicon Valley and the global tech world.
He said: the vibe in San Francisco is "extremely manic." The AI industry's polarization is the most severe he has ever seen. Only about 10,000 people—employees and founders of OpenAI, Anthropic, xAI, and NVIDIA—have achieved "retirement-level wealth," over $20 million. Everyone else is worried that with an annual salary under $500,000, they can work their whole lives and never catch up.
Note that he said "people with an annual salary under $500,000." In any country, that is a high-income group. But in San Francisco, these people are anxious.
One netizen commented something very piercing: "In this wave, the same technology is both a lottery ticket and the thing that eats away your way out."
You use AI to make money, and AI is also eating your job. This is the most real portrayal of the AI industry in 2026.
To understand this "spiritual earthquake," let's first look at what happened in the AI industry this week.
Event 1: Cerebras IPO, surging 108%.
On Wednesday, chip company Cerebras Systems went public at a price of $185, raising $5.5 billion. The opening price directly soared to $385—more than doubled. At close, its market value reached $66 billion. This is the first major tech IPO of 2026. You should know that just over a year ago, many people still thought it might never go public.
A company making AI chips became a $66 billion giant in a single day. This is not called gold mining; this is called "digging up the gold mine itself."
Event 2: ChatGPT connected to your bank account.
OpenAI launched a personal finance tool. Through a partnership with payment infrastructure company Plaid, users can connect accounts from more than 12,000 financial institutions to ChatGPT. Want to know where this month's money went? Ask ChatGPT. Want to know the risk of your investment portfolio? Ask ChatGPT. Want to plan for retirement? Ask ChatGPT.
This news has two highlights: First, AI is moving from "chatting" to "acting"—it has begun managing your real money. Second, OpenAI chose Plaid. Plaid is one of Silicon Valley's most famous API companies, connected to almost every U.S. bank. This means ChatGPT is becoming a financial super gateway.
Event 3: Greg Brockman returns to the product front line.
OpenAI co-founder Greg Brockman has taken over product strategy. Over the past year, OpenAI has experienced a roller coaster—from Sam Altman being fired and returning, to the lawsuit with Elon Musk, to the continuous loss of core talent. Now, the founding team is regaining its footing.
At the same time, OpenAI is reportedly preparing to take legal action against Apple. The two companies that "cooperated" for less than a year are now turning against each other. The reason? Apple's layout in the AI field may have already crossed OpenAI's bottom line.
Event 4: AI begins to "build itself."
TechCrunch published a major article titled: "What happens when AI starts building itself?" Researchers at security company Calif used Anthropic's cybersecurity AI Mythos and took only 5 days to write code, successfully exploiting two macOS vulnerabilities to achieve privilege escalation.
You should know that Apple previously claimed its MIE (Memory Integrity Enforcement) technology was an "unprecedented design and engineering feat that took five years." And AI broke it in only 5 days.
Event 5: Runway wants to challenge Google.
Runway—the company that helps filmmakers do AI video editing—now says its goal is not video, but to challenge Google. Runway's CEO believes that multimodal models are the future of AI, and Google's layout in short video and multimodal directions has given Runway an opportunity. A video AI company saying it wants to challenge a search giant sounds very irrational, but in the AI era, "crossing boundaries" is the norm, and "crossing boundaries to take down giants" is no longer new.
Event 6: Cisco lays off 4,000 people, invests the money in AI.
Cisco announced layoffs of about 4,000 people while reporting "record quarterly revenue." Layoffs and record revenue appearing at the same time—this is not self-rescue during losses, but strategic restructuring while profitable. Cisco is saying: I replaced the budget for people with the budget for AI.
Event 7: ArXiv bans AI-generated junk papers.
Academic preprint platform ArXiv announced that if researchers submit large numbers of AI-generated "junk papers," they will be banned for one year. Behind this is a grim reality: AI-generated academic papers have already become so numerous that the platform cannot bear them.
Put these seven things together. What can you see?
What we see is a world accelerating toward polarization. On one side are soaring IPOs, exploding valuations, and wealth myths. On the other side are layoffs, anxiety, academic pollution, and security threats.
This AI "gold mine" is enriching its wealth and also eliminating its outsiders.
The "10,000 people getting rich" that Deedy Das mentioned is actually a structural phenomenon.
Think back to the internet wave. From 1995 to 2000, how many people became rich because of the internet? Many. At that time, the internet was "flood irrigation"—as long as you were connected to .com, you could raise funds.
But the AI wave is different.
AI is a "super capital-intensive" industry. Training a large model requires hundreds of millions of dollars to buy GPUs, thousands of megawatts of electricity, and a top PhD team. This is not a project a college student can do in a garage.
So AI wealth is concentrated in a few companies: OpenAI, Anthropic, xAI, Google DeepMind, Meta FAIR, NVIDIA. And NVIDIA's GPUs are the "shovels" for all AI companies, so it has instead become the biggest winner.
This has formed a "winner-takes-all" pattern. AI has not created a new middle class; it has created an extremely small "super-rich club" and a huge "anxious group."
This is also why Deedy Das's post caused such a big reaction. The "anxious people" he mentioned are engineers in Silicon Valley with annual salaries of $400,000-$500,000. They were originally the "gold-collar" workers of this industry, but now they realize: they may never get into that club.
And what makes them even more uneasy is—they are being replaced by the technology they helped build. Software engineers find that "the skills they spent a lifetime building are no longer useful." This feeling of "being eliminated by what you created yourself" is more unsettling than any form of competition.
In the AI era, I observe three kinds of people moving toward completely different fates.
The first type of person: the shovel makers.
NVIDIA, Cerebras, TSMC—these companies don't directly make AI applications, but everyone doing AI depends on their hardware. Cerebras' IPO rose 108%, with a market value of $66 billion. This is the power of "selling shovels."
In a gold rush, the ones who truly make big money are often not the gold diggers, but the shovel sellers. In the AI era, this rule still holds, and even more obviously so.
The second type of person: the gold miners.
OpenAI, Anthropic, Google DeepMind—they are training the most powerful foundation models. This is the most expensive, most risky, but potentially highest-return track.
But note: the threshold for this track is already so high that it is almost insurmountable. The training cost of GPT-4 is about $100 million. GPT-5 is rumored to exceed $1 billion. The next generation of models may cost on the order of tens of billions of dollars. This is no longer a game startups can play.
The third type of person: those who use gold to make jewelry.
This is where the opportunity lies for most people and companies. You don't need to mine gold yourself, and you don't need to manufacture shovels. You only need to use AI, this "gold," to make better products and services than what exists now.
Runway uses AI to make video tools, and its valuation has multiplied dozens of times. Replit uses AI to enable "vibe coding"—you say one sentence, and it writes code for you, already making even Apple feel threatened. A Saudi startup uses AI to make voice robots and keeps raising funding.
OpenAI's ChatGPT connecting to bank accounts is also "making jewelry"—it inlays AI, this "gold," into the scenario of financial services.
People who "make jewelry" do not need GPU clusters worth tens of billions of dollars. What they need is deep understanding of scenarios and the ability to act quickly.
Deedy Das's post made me think of a bigger trend: AI is accelerating the "disappearance of the middle class."
Let's look at the news of Cisco laying off 4,000 people. At the same time as the layoffs, revenue hit a record. What does that mean?
It means these jobs may never come back. Cisco is not short of money; rather, it believes these jobs are "not worth having people do" anymore. AI can do them better, faster, and cheaper.
This is not just Cisco's logic. In finance: Goldman Sachs' trading floor shrank from 600 people in 2000 to just a few dozen now. In law: AI can handle more than 50% of contract review and due diligence. In media: ArXiv has already begun cracking down on "AI papers," because there are too many AI-generated academic articles.
Jack Antonoff—a famous musician and Taylor Swift's longtime collaborator—raged on Instagram this week against users of AI tools, calling them "godless bitches." He said: "All those people excited about new ways to fake art, go ahead, drive off a cliff. We're happy to see you leave."
Anger is anger. Trends will not change because of anger.
The real question is: if "middle-class" jobs are disappearing en masse, then where are the new "middle-class" jobs?
The answer is: in the things AI cannot yet do.
What can AI currently not do? Based on this week's news, we can summarize:
First, AI is not very good at judging what is valuable. ArXiv wants to ban AI-generated papers, not because AI writes badly, but because AI writes too much, and the vast majority has no value. Judging whether "this thing is worth doing" is still a core human ability.
Second, AI is not very good at building trust. If OpenAI wants to connect to your bank account, you need to trust that it will not leak information or make wrong investments. Building this kind of "trust" requires brands, reputation, regulations, insurance—all of which involve a human role.
Third, AI is not very good at creating "non-consensus" breakthroughs. When AI accelerates wildly in popular directions, real breakthroughs often happen in unpopular directions. After Runway reached the extreme in video, it now says, "I want to challenge Google Search." This is a non-consensus judgment—the direction of AI's development is not decided by AI itself.
After reading all this, you may ask: so what should I do?
I'll give you a "three dos and three don'ts" action list. Not profound, but useful.
✅ Do "make jewelry," don't just be an audience member of the "gold mine."
Don't spend every day scrolling through AI news and marveling at how fast technology is changing. That is called "watching the gold mine." Use AI to do one thing—even if it's just writing a WeChat public account article. Turn yourself into "a person who uses AI to do things," not "a person who discusses AI."
✅ Do go deep into scenarios, don't chase technology.
ChatGPT has the ability to connect to 12,000 banks, but if you don't understand finance, you cannot use this ability to make a good financial product. OpenAI is a tool provider, not an expert in every scenario. Your understanding of a certain industry is a moat that AI cannot replace.
✅ Do believe in the trend, don't bet on the trend reversing.
Jack Antonoff curses AI, but the music industry has already been changed by AI. Print media cursed the internet, but print media has almost disappeared. Every era has "angry defenders of the city," but history never stops because of their anger.
Instead of being angry, learn. Instead of being anxious, act. Instead of sighing while watching the myth of 10,000 people getting rich overnight, start making your own "jewelry."
The AI industry in 2026 is like a jungle just after a heavy rain. Thick trees are growing wildly, competing for sunlight. Low shrubs are being shaded and gradually withering.
But there is also a third role in the jungle—vines climbing on tree trunks, orchids blooming among branches, birds perching in the canopy. They do not compete with the big trees for sunlight, but attach themselves to the big trees and find their own ecological niche.
In the AI gold rush, 10,000 people have already gotten rich. This is a fact. But this does not mean everyone else has no opportunity.
The bigger opportunity lies precisely with everyone else—using AI's gold to make your own jewelry.
Don't be an audience member. Be a player.
Room 1409, Building C, Shenzhen Century Industrial Center, No. 743 Zhoushi Road, Hezhou Community, Hangcheng Sub-district, Bao'an District, Shenzhen
