Data shows that over the next six months, compute growth will surpass AI-related revenue growth. Since October, the shortage of compute has significantly improved; by then, top AI companies will be forced into price wars, as too much excess compute has already been pre-ordered. Global AI-related revenue is projected to rise from approximately $110 billion in 2025 to $300 billion in 2026 and $550 billion in 2027—annual increments of +$190 billion and +$250 billion, with growth rates of 170% and 83%, respectively. The projected growth rate is already beginning to slow. From 2026 to 2027, new installed capacity is expected to reach 32 GW and 49 GW. At $10 billion per GW, annualized cost increases amount to +$320 billion and +$490 billion. Shipment volumes are rising by +54% and +27%, but due to increased power consumption per GPU, the equivalent GW growth rate is +67% and +44%. Although Anthropic’s revenue continues to grow rapidly—with its annual recurring revenue (ARR) more than doubling in the first half of the year—it can still absorb the compute demand. However, from the second half onward, growth trends plateaued in July, August, and September, and the compute shortage has significantly eased. What’s alarming is that GPU shipments are still growing at such a rapid pace, while revenue fails to keep up—and compute efficiency continues to improve. With these three factors converging, the inflection point for compute appears imminent: inefficient GPUs will sit idle, while new, high-efficiency GPUs will remain in short supply—reminiscent of the Bitcoin miner upgrade cycle. If compute and land remain scarce, operators will prioritize deploying the latest machines. By late 2027, as committed installations come online, U.S. data centers may have surplus compute capacity—this scenario aligns logically with the above data trends. On another front, let’s examine Anthropic’s compute growth and how much it has pre-ordered. Its compute growth rate far exceeds its revenue growth rate. Eventually, equilibrium will be reached: since Anthropic has signed compute commitments, excessive idle capacity will likely trigger price reductions. Anthropic has recently relaxed its usage constraints slightly, and advanced intelligence capabilities are being democratized, challenging the cost-efficiency model of Chinese models. Although China benefits from lower electricity costs, without access to the most advanced chips, its cost per generated token is likely still higher than in the U.S.

Next week is another blockbuster earnings week for U.S. stocks, with earnings reports from Apple, AMD, Palantir, Cloudflare, Shopify, and others already compiled. Even more critical are two major macro events: the U.S. midterm elections on November 3 and the non-farm payrolls report on October 6. Pay close attention to capital expenditures—tech companies’ earnings reports and conference calls will provide guidance on our investment direction for the next six months. If the market survives the month following the midterm elections, it is likely to enter an upward trend. Since 1946, in all 20 midterm election cycles, the S&P 500 has never declined over the subsequent 12 months, with an average return of +14.8% and a median return of +14.2%.

Apple's intelligent integration with Alibaba's Qwen has begun testing, with features launching on the iOS Developer Beta version on November 4. Chinese users can directly access Qwen’s text and image understanding, content generation, and other capabilities through Siri, writing tools, and visual intelligence. $BABA

After Liu Bo was forcibly liquidated, he’s likely headed back to prison—given his leveraged position, there’s probably nothing left but residual value. https://t.co/AAq5pNCeOv

After securing substantial funding, will Anthropic initiate a price war? I found the answer in its prospectus. Amazon, Google, Broadcom, and AMD have collectively provided or committed approximately $133.3 billion in funding to Anthropic, in exchange for roughly $405.7 billion in compute orders—roughly $3 in orders for every $1 invested. Anthropic has used minimal cash to lock in compute capacity far exceeding its own scale, while the cloud giants have converted their investments in Anthropic into guaranteed orders for their own cloud services and chips. The contracted compute capacity must ultimately be monetized by selling tokens; after going public, Anthropic will have no shortage of cash—approximately $131.7 billion in cash and short-term investments as of end-June, plus an additional $77 billion in undrawn strategic notes and a $15 billion revolving credit facility. However, the $405.7 billion in binding compute payment obligations mean that if Anthropic faces a competitive downturn similar to the third quarter, it will almost certainly lower prices—especially since post-IPO financials will be transparent. Currently, with locked-in capacity and growing demand, token prices can be maintained. But the contracted compute capacity represents a hard cost regardless of utilization—if tokens aren’t sold, they must still be sold at a discount, just like clearing inventory. At that point, no one will defend prices—market penetration will be the priority. If infrastructure investment has been overdone, this dynamic will be validated within the next two years.

The Korean stock market surged to its circuit breaker. A-share fundamentals are becoming increasingly solid. Hong Kong stocks are still digging deeper to strengthen their foundation. U.S. stocks edged up, holding firm with a strong backbone. Why is Bitcoin still in its bottoming phase? Was all that previous strength just an act?

The Bitcoin cycle top has been identified, and Lianyanshe's website immediately incorporated the Wave Cycle Index into its data dashboard.

The market’s concern is no longer just inflation—can the AI investment boom be sustained? 1. Rising oil prices can no longer push up U.S. Treasury yields. Last night, Treasury yields continued to decline, while tech stocks plunged, signaling that market capital is reassessing returns and reallocating assets. The old trading logic was: rising oil prices → fears of higher inflation → expectations of further Fed rate hikes → selling stocks and Treasuries. This was also why U.S. Treasury yields remained stubbornly high recently. The current logic reflects concerns over the return on AI infrastructure investments—lower yields and defaults could have a more severe impact on economic growth, prompting large-scale buying of U.S. Treasuries. The 10-year Treasury yield falling below 5% is essentially a reconfiguration of risk. 2. The tech stock crisis has not yet evolved into a uniform, global sell-off. Last night, semiconductor and tech stocks were sold off, but defensive sectors and low-P/E defensive stocks continued to rise, indicating a shift in market sentiment. 3. Next week, Microsoft, Amazon, and Alphabet will release their earnings reports. After Meta’s results, Wall Street has changed its earnings evaluation criteria. It is highly likely all three cloud giants will exceed expectations—but this does not necessarily mean AI stocks will rebound. If AI investments erode free cash flow and force companies to issue debt to fund operations, it could further fuel market concerns. Meanwhile, the Fed will announce its interest rate decision. One positive development: the expected timing for the next Fed rate hike has been pushed back from December to January (due to the stock market decline, the urgency to hike has diminished).

It’s still a good time to continue buying regularly, dollar-cost averaging, and using grid LP strategies. Why? Because most people have extremely poor timing and price-selection skills—it’s common to buy and immediately see prices drop. Don’t overestimate your ability to time the market. The advantage of dollar-cost averaging, grid strategies, and LP is reducing the probability of making poor decisions. If you want to maximize returns, you could go all-in, but the margin for error is extremely low. People waiting for the “final drop” or to buy at $50,000 have been waiting from $60,000 all the way to $80,000—and now buying still feels uncomfortable, fearing they’ll buy right before another drop, leaving them with costs 30% higher than others. But regardless of whether there’s still one final drop, the risk of missing out entirely remains. Bitcoin has traded sideways between $80,000 and $87,000 for the past month. While buying spot or using grids above the absolute bottom means you’ll earn less, it’s still better than missing out completely. Gradual buying on dips, with the downside support level unchanged at just a 2.4% drop below the current range, allows you to enter steadily under three scenarios: sideways movement, upward trends, or no breakdown of the range. Even if your judgment is wrong, losses are minimized. Sitting on the sidelines with a bearish outlook may feel satisfying—but you’ve already waited from $60,000 to $80,000. Waiting is fine, but holding cash is equivalent to holding a put option on the market. Missing out is itself a risk.

I keep trying to catch the bottom in U.S. stocks and BTC—laughing all the way, only to realize there’s no end to the bottoms. This time, I bought too soon; the risks hadn’t gradually eased, but seeing prices hit new highs, I followed historical patterns and started buying at the slightest dip in the index. I’ve always maintained over 80% exposure in my U.S. stock portfolio, never dropping below 50%. This time, my account pulled back by about 10%. U.S. stocks and crypto are pretty similar—I primarily use a share-based and coin-based position strategy. After years of losing money, I’ve broken down bottom-fishing into four steps to ensure I always have capital to buy: 1. When it dips slightly, fully load your spot position using sell puts to accumulate. 2. If it drops further, add margin leverage—but keep total leverage under 1.5x. 3. When a crash hits and everyone starts screaming, shift some positions into long-dated options; options require low margin and typically offer 10x+ capital efficiency. 4. If it drops further, replace the underlying stock with a 2x leveraged ETF (I haven’t reached this step yet). This strategy still carries significant drawdowns—normally, just executing steps 1 and 2 is enough. Being too aggressive comes at a cost: in crypto, I’ve been liquidated at low leverage during the 10-11 drop; in stocks, adding leverage meant getting margin-called during a modest pullback—even when the risk clearly wasn’t that high.