The most popular phrase in the market at the beginning of the year was: AI will kill SaaS. But now I increasingly feel that cybersecurity might be the biggest exception to this logic. The reason is actually very simple: The stronger AI gets, the larger the attack surface companies face. Previously, a company protected employee computers, email, and servers. In the future, they will also need to protect AI Agents, APIs, cloud databases, machine identities, model permissions, and various systems that automatically execute tasks. So for ordinary software companies, AI might mean: "Will customers still need so many seats in the future?" But for security companies like $CRWD, $PANW, $FTNT, the question might instead become: How many new security capabilities will customers need to buy? This is what makes cybersecurity particularly interesting to me. AI can replace some software functions, but it’s very hard for AI to make companies say: "I won’t spend on security budgets anymore." On the contrary. The more AI can automate tasks, the greater the potential loss from a single permission leak or attack. So if I were to pick a long-term direction in AI software least likely to be wiped out by the "AI replacement theory," I would prioritize: $CRWD $PANW $FTNT $ZS $OKTA One of AI’s biggest byproducts might be an increasingly large cybersecurity market.

The market fears the last leg the most in this round: NEAR suddenly accelerates, UNI surges over the week, BCH remains flat; U.S. software stocks only recognize profit-taking. $BTC remains steady as ever #BTC兑黄金比率升至1月以来高位，强势能否延续？ $NEAR is around $2.25, up over 10% in 24 hours, about 22% in seven days. The AI narrative and volume bring back resilience, but 2.30 is already resistance; if it holds above, look to 2.45, a drop back to 2.10 indicates weakening chasing demand, I prefer to wait for a pullback. $UNI is about $6.32, up more than 40% in seven days, the risk is that profits are too thick. Support at 6.00, breakout at 6.45; with volume, if it holds above, look to 6.80, if it falls below 6, watch for profit-taking, this is not a place for blind chasing. $BCH is about $250, almost flat in 24 hours, clearly not keeping up with altcoin enthusiasm. 245 is support, 254–260 is resistance; only a volume breakout above 260 counts as a catch-up rally, otherwise continue to treat it as a range-bound, wait for funds first. On the U.S. stock side, $DOCU revenue up 9.4%, EPS beats expectations, closing up 3.7% at 68.41; $ASAN revenue up 9.9% but guidance lacks surprises, plunging 12.8% to 8.81; $ORCL rises 3.1% to 158.78 on OpenAI cloud collaboration expectations, the real test is the September 10 earnings report. Growth alone is not enough now, it must continue to beat expectations.

The four companies newly added to the S&P 100: $DELL, $ANET, $PANW, $SNDK. I reviewed their latest financial reports again. The conclusion is straightforward: These four companies perfectly represent the four directions in which AI infrastructure is expanding: servers, networking, security, and storage. Let's start with $DELL. The latest quarter's revenue is about $47 billion, up 58% year-over-year. AI server revenue is $16.4 billion, and new AI server orders in a single quarter reached $60.9 billion, with the backlog at the end of the period directly hitting $95 billion. Dell's biggest problem now is no longer "whether there is AI demand," but: How fast can they fulfill so many orders? AI servers are also driving traditional server, networking, and storage businesses, so my positioning of Dell is simple: The performance certainty for the next few quarters is very high. But going forward, we can't just look at revenue; we need to start focusing on how much profit AI servers can retain. Next, $ANET. The latest quarter's revenue surpassed $3 billion for the first time, up 37.7% year-over-year. Even more impressive is the adjusted operating margin approaching 50%. This is why among the four companies, I like ANET's business model the most. When AI clusters scale from tens of thousands to hundreds of thousands of GPUs, the real growing challenge is not the chips themselves, but: How these chips communicate at high speed. 800G is upgrading to 1.6T; the larger the AI cluster, the more valuable the switches and Ethernet Fabric become. So ANET is not just capturing one-time server orders but the entire AI data center network expansion. High growth + high profit margin + AI essential demand—this kind of business is very rare. The third is $PANW. The latest quarter's revenue is about $3.4 billion, up 34% year-over-year. Next-Generation Security ARR reached $9.1 billion, up 63% year-over-year, and RPO reached $21.2 billion. The logic behind PANW's real benefit is simple: Previously, enterprises only had employees logging into systems; Now, there could be thousands or tens of thousands of AI Agents simultaneously calling APIs, accessing databases, and executing tasks. The stronger the AI capability, the larger the enterprise's attack surface. So I have always believed that cybersecurity is not a niche track in the AI era but a layer that must increase budget once AI is truly deployed at scale. PANW may not be the most aggressively rising among the four, but: Recurring Revenue + high cash flow + cybersecurity essential demand Make it increasingly look like a true AI blue chip. Finally, the one I am most familiar with, $SNDK. The latest quarter's revenue is about $8.97 billion, up 51% quarter-over-quarter; full-year revenue is about $20.25 billion, up 175% year-over-year; data center business surged 437% for the year. But the biggest difference between SNDK and the other three companies is: It is not only benefiting from AI demand now. It is benefiting from: AI demand + NAND price increases + enterprise SSD expansion + storage cycles. So the profit elasticity will be extremely volatile. But conversely, once NAND prices reverse in the future, its profit fluctuations will be much larger than ANET and PANW. So my current positioning for these four is simple: $DELL: strongest order certainty $ANET: best business model $PANW: most like a mature AI blue chip $SNDK: greatest earnings elasticity If judging only by financial report quality, my personal ranking is: ANET > PANW ≈ DELL > SNDK But if judging only by earnings explosive potential in the next one to two years: SNDK > DELL > ANET > PANW These two rankings are not contradictory. Because SNDK is currently earning from "AI + cycle resonance" money, while ANET earns from more stable, higher-margin AI infrastructure growth. What is truly worth watching is not who entered the S&P 100. But the signal the market gives when these four financial reports are combined becomes clearer: DELL tells you server orders are still booming; ANET tells you networking is becoming a bottleneck; SNDK tells you storage prices are rising; PANW tells you security budgets must continue to increase after AI deployment. This round of AI CapEx is no longer just buying $NVDA GPUs. It is expanding layer by layer from: Computing power → servers → networking → storage → security This is the most valuable aspect of these four companies entering the S&P 100 together.

After oil prices climbed back above $90, I think many people's first reaction was: "Buy oil stocks." But the energy sector isn't that simple. If oil prices keep rising, I would categorize companies into three tiers. The first tier is those directly selling oil: $XOM, $CVX, $COP. The higher the oil price, the more directly they benefit through upstream profits and free cash flow. The second tier is oil services: $SLB, $HAL. What they really want to see is not just a few days of rising oil prices, but oil companies believing that high prices will persist, prompting them to increase drilling and equipment CapEx. The third tier is where I am most cautious: refining, aviation, transportation, and many high-energy-consuming enterprises. Because rising crude oil prices are not necessarily good for them; more often, it means higher costs. So after oil prices rise to $90 or $100, I wouldn't simply interpret it as: The entire energy sector benefits. What really deserves study is: Is the oil price spike just a short-term pulse caused by war, or is it high enough to make energy companies reopen capital expenditures? The former is driven by sentiment. The latter could form a real industry trend. Right now, what I care about more is not how much oil prices have risen today. But: $XOM

Storage forever! $SNDK

This time's S&P 100 adjustment, I find it more interesting than just $SNDK being included. Because the four companies entering together this time are: $DELL, $ANET, $PANW, $SNDK. They basically cover the most important layers of AI infrastructure. $DELL — AI servers $ANET — AI networking $SNDK — data center storage $PANW — network security This is not just a simple index reshuffle; it feels more like the core structure of American blue chips is being reshaped by AI. Let's start with $DELL. The latest quarter's AI server revenue has reached $16.4 billion, with single-quarter AI server orders at $60.9 billion, and backlog even hitting $95 billion. The biggest change for Dell now is that the market no longer sees it just as a PC company. As AI clusters grow larger, servers, networks, storage, racks, and liquid cooling all need to be delivered together. Dell is becoming the "final assembly plant" for AI data centers. Next, $ANET. It plays in the second layer I am very optimistic about: AI Networking. As GPUs/XPUs increase, the real challenge is not "whether there are chips," but how to transmit data at high speed among hundreds of thousands of chips. 800G is continuing to upgrade to 1.6T; the larger the AI cluster, the higher the value of switches. So among the four, if I had to pick the one with the most comfortable long-term business model, I would actually pay more attention to ANET. Then there's $SNDK. Its logic is the most aggressive and also the most cyclical. With AI data centers continuously expanding, enterprise-grade SSD and NAND demand is clearly rising, coupled with a storage price increase cycle, SNDK's stock price elasticity naturally is the greatest. But it’s different from ANET. ANET earns from long-term network expansion; SNDK, besides AI demand, also depends on: NAND prices, supply, inventory, and cycles. So among the four: SNDK is the easiest to surge sharply and also the most volatile. Finally, $PANW. It actually represents another direction in the AI era that is easily overlooked: The stronger AI gets, the more important network security becomes. In the future, as AI Agents in enterprises increase, accounts, permissions, APIs, and data access all become more complex. Attackers will also use AI. So AI doesn’t necessarily weaken security companies; it might actually increase enterprise security budgets. If I put the four together, my own ranking is simple: Short-term elasticity: SNDK > DELL > ANET > PANW Long-term certainty: ANET ≈ DELL > PANW > SNDK What’s most interesting is that among those replaced this time, there are traditional consumer and real estate blue chips like Nike, Colgate, Simon Property. The ones coming in are: Servers, networking, storage, security. This is what I think is the most worth watching about this S&P 100 adjustment. Previously, the 100 most core American companies represented consumption, finance, and industry. Now AI is gradually taking over their positions. The next round of companies truly entering the core American blue chip circle will very likely continue to come from AI infrastructure.

The most outrageous scene of the AI era may have just begun: AI is "eating up" the memory of ordinary people. FT today directly gave this round of the market a name: RAMageddon. In the past year, the price of some DRAM has already risen to about 5 times the original. The reason is simple: SK Hynix, Samsung, and Micron now want to produce not ordinary computer memory at all. But: HBM. Because the value of the same wafer sold to AI data centers is far higher than that sold to ordinary PCs and phones. So production capacity has been continuously shifting towards: HBM → AI server DRAM The result? Ordinary consumer electronics are starting to be out of stock. Manufacturers like Apple, Microsoft, and Nintendo are already facing obvious increases in memory costs, with some product prices even rising close to 20%. So the truly scary part of this AI market is not just: How many GPUs $NVDA has sold. But that it is reallocating the entire semiconductor industry's production capacity. For every extra unit of memory AI servers buy, consumer electronics may lose a unit of capacity. This is also why I have always been optimistic about: SK Hynix, $MU. The storage industry is no longer the old model of: Poor demand → price cuts → production cuts → waiting for the cycle to reverse. Instead, AI has directly turned high-end memory into a strategic resource. Even more interestingly, this shortage may continue until 2027. So what I’m most focused on next is not: "Can DRAM prices still rise?" But: When AI is willing to pay higher prices to compete for capacity, who in traditional consumer electronics will ultimately bear this cost? The answer may be simple: Consumers. Before, AI drove up tech stocks. Now AI is starting to raise the price of your next computer and phone.

Weekend market is not a broad rally: LTC surges, AVAX sideways, AAVE pulls back; US stocks have played out risks from earnings, leadership changes, and regulation all at once. $LTC currently around $53, intraday rose over 5%, still outperforming the market over seven days. Buying pressure effective below 50, 54 has entered resistance zone; holding above 54 then look to 56, falling back to 51 indicates more of a short-term catch-up rally, I won’t chase the highs. $AVAX around $7.44, basically sideways in 24 hours. Watch 7.30 for support, 7.55 for breakout; before volume expands, breakout credibility is limited. Above 7.55 then look to 7.80, if it falls below 7.30 avoid first. $AAVE around $129.5, down about 3% in 24 hours, but still up nearly 7% over seven days. V4 integration and deposit growth support mid-term logic, but short-term is shrinking volume pullback. 128 must hold, reclaiming 136 counts as strengthening; if broken, look at 124 first. On US stocks side, $IOT revenue up 30% YoY and raised guidance, but stock price fell from 45 to 40.2, 39 is defense level; $ADBE down 6.7% due to CEO transition, watch 264 for support; $TSLA Cybercab just launched and is under investigation, down nearly 6% to 354, 350 must hold. Now stories are not lacking, what’s missing is buying that can hold the gains through close. #LTC catch-up rally sustainability

Many people understand AI data centers as: Buying GPUs, then starting to train models. But it's not that simple at all. When tens of thousands or hundreds of thousands of chips work simultaneously, the real troublesome question becomes: How do they transmit data between each other? A table to understand the AI network industry chain: Company｜Core Segment $NVDA｜InfiniBand, Spectrum-X networks $AVGO｜Switch chips, SerDes, DSP $ANET｜AI data center Ethernet switches $MRVL｜DSP, interconnect chips $LITE｜Lasers, optical communication devices $COHR｜Optical devices, transceiver components $AAOI｜High-speed optical modules You can imagine an AI data center as a city. GPUs are factories. Switches are highways. Optical modules and fibers are roads. DSP and SerDes are responsible for correctly sending signals. The larger the AI cluster, the more data needs to be exchanged between chips. So the real bottleneck of future AI infrastructure may gradually shift from: "Whether there are GPUs" to: "Whether hundreds of thousands of GPUs can be efficiently connected together." This is also why when I look at AI hardware now, I don’t just focus on computing power. Networking is becoming the second main line.

The craziest IPO in AI history might be coming. Anthropic is preparing to go public, with a target valuation that could reach as high as $2 trillion. This is definitely good news for hardware. For example, my favorite $SNDK No mistake. An AI company that hasn't even gone public yet is aiming for a valuation on the scale of super giants like Apple, Microsoft, and Google. Even more astonishing, Anthropic expects its revenue by 2028 to be: $190–200 billion. So Wall Street isn't really betting on how much Claude makes today. What they're betting on is: Whether enterprise AI will become a super market worth hundreds of billions of dollars. One of the biggest differences between Anthropic and OpenAI is that Anthropic has been more enterprise-focused from the start. Writing code, enterprise agents, APIs, data analysis, secure deployment... If in the future large companies really start having millions of AI agents working for employees, then every model call essentially generates revenue. So a $2 trillion valuation sounds outrageous, but the logic behind it is simple: If Anthropic can really achieve $200 billion in revenue, even with a 10x PS ratio, that’s: $2 trillion. But the problems are also huge. The biggest risk for AI now is no longer "whether there are users." But: Whether the money earned can cover the terrifying costs of GPUs, data centers, electricity, and model training. So I think this Anthropic IPO will be a real big test for the AI market. If the market is willing to give $2 trillion: It means Wall Street is still willing to pay for the future of AI. If the market is not willing, Then the entire AI sector might for the first time truly start discussing: How much growth is really worth. I will definitely be watching this IPO closely.