The computing power sector is back and has entered a more discerning phase.

In the past two days, the trend in tech stocks has been quite interesting. The sectors seeing the most significant gains are no longer just traditional chip stocks, but also cloud services, data centers, and AI infrastructure. The signals from the market are clear: the mainline of AI computing power has not ended; it has simply entered a new phase. In previous market conditions, investments in computing power were largely driven by capital expenditure and shortages of GPUs. However, after the recent rounds of fluctuation, funds have clearly begun to apply a different set of criteria: its not that they no longer consider the scale of computing power, but they are now more concerned about whether this power can ultimately customer orders, deliverable projects, and sustainable income. In other words, the AI market is transitioning from a narrative centered around "selling shovels" to one focused on "who can reliably collect rent." This change is crucial. In the earlier phase, the market was willing to assign valuations based on "how much computing power was planned," "how many resources were tied up," and "how well one was positioned in the AI wave." But after a round of volatility, capital has become more discerning. It no longer just listens to abstraction; it has begun to query several more practical concerns: Are the orders real? Can the projects be delivered? Is the computing power operational? Can income be confirmed? Can profit and cash flow keep pace? These are precisely what the current market values most. Previously, discussions surrounding AI were met with enthusiasm for potential; now, the market is more interested in actual delivery. Order volume, customer quality, delivery pace, contract duration, revenue recognition, and cash flow conversion are becoming core metrics for pricing in this sector. A company that only has concepts without actual deliverables will find it hard to go far in a recovery market; conversely, if a company has orders, projects, and revenue, once market sentiment improves, valuation recovery is often swift. Following this logic, if we take a look at the computing power operators listed in the Hong Kong market, we can see that the value of certain companies is beginning to become clearer. Among these, one company is particularly worth highlighting. What sets it apart is not merely how much computing power it has planned, but how much verifiable business foundation it has already established. According to public information, GBA AI COMP has over 30 billion RMB in orders on hand, has delivered more than 15 billion RMB in orders, and operates over 50,000 P (FP16 dense) in computing power. Of the newly disclosed orders, over 95% are long-term contracts lasting up to five years. Looking at these numbers together reveals a different significance. Orders on hand indicate that demand is not merely a figment of imagination but has already settled through contracts, projects, and customer relationships; delivered orders show that the company is capable of not just receiving orders, but also executing projects; and operating over 50,000 P (FP16 dense) implies that this is not an asset stuck in the construction phase, but infrastructure that has already entered the operational stage. More importantly, this is just the portion visible from the current publicly available information. Many operations in the computing power industry are often delivered in batches, billed in phases, and release income on a cycle. In other words, the publicly available data is only one facet; the actual scale of subsequent business operations, pace of income release, and profit elasticity still hold considerable imaginative potential. This is also what makes GBA AI COMP (01396) quite recognizable. Many computing power companies talk about how much they will build in the future, but the market now wants to see how much has already been delivered, operated, and received. Computing power is not just a number on a PPT; realizing it involves a complex series of processes including equipment, data centers, networks, scheduling, operations, and customer adaptation. If any link in this chain is blocked, planning will be difficult to transform into billable assets. Thus, the core capability of computing power operators is not merely to tell grand stories but to effectively run the entire chain. From this perspective, GBA AI COMP offers more than a simple computing power concept; it provides a model of realization from orders to delivery, from delivery to operation, and from operation to revenue. The current market is actually undergoing a re-evaluation of AI companies. Companies without orders will face scrutiny; those with unclear deliveries will see discounts; those whose revenues take too long to materialize will be revalued. In contrast, firms that can secure long-term customers, continue delivering projects, and enter the revenue release phase will be re-identified amid sector recovery. Therefore, this round of recovery in the computing power sector is not just an emotional rebound; it seems more like the capital market's reaffirmation of the vibrancy of AI infrastructure. The adjustments in the sector's early phase have alleviated many concerns, and confidence in the supply-demand balance of computing power and the returns on AI investment is being restored. However, post-recovery, funding will not buy indiscriminately as it did in the first wave; it will place greater emphasis on realization capability. The computing power market has not cooled off; it has simply entered a more discerning phase. Moving forward, the market will not only reward those who tell stories but will prefer those who can deliver. Orders continue to come in, projects continue to be delivered, and operations continue to scalethat is the current most valuable mainline in the computing power sector. In the AI era, computing power is not a one-time business; it is new infrastructure. Those who can establish, manage, use, and lease out this infrastructure while generating stable returns are more likely to weather fluctuations. This also underlies the logic behind the renewed growth in the computing power sector. This article is reprinted from the "Liu Xiaobo Talks Finance" public account, edited by GMTEight: Liu Jiayin.