## Detailed walkthrough
### [00:07](https://www.youtube.com/watch?v=ddApAM8iYmw&t=7s) Introduction: Cutting Through Macro Noise to Earnings Reality
Even with the formal earnings season technically behind Wall Street, the workload has hardly slowed down. Finishing the midnight review of two blockbuster earnings reports, host Xiao Tao dives straight into the analysis, urging viewers to show their support with likes, comments, and shares. Setting aside broader macroeconomic data and general market chatter, this breakdown focuses entirely on the wealth of structural insights buried inside the corporate earnings releases. In an environment saturated with headline noise, macroeconomic statistics pale in comparison to what these earnings reports reveal about capital flows, artificial intelligence infrastructure, and future industry demand.
[00:44](https://www.youtube.com/watch?v=ddApAM8iYmw&t=44s) Nvidia’s Shifting Customer Base: Hyperscalers vs. Emerging Clouds
Nvidia’s latest earnings report opens with staggering figures surrounding hyperscalers. The aggregate remaining performance obligations (RPO)—representing the future backlog of commitments across all major hyperscalers—stands at an enormous $2 trillion. Looking at capital expenditure, hyperscalers spent $800 billion in 2026, with 2027 projections climbing to $1.3 trillion. While 2026 capex nearly doubled, dividing the projected $1.3 trillion by $800 billion reveals that the year-over-year growth rate in hyperscaler spending will actually begin to decelerate heading into 2027.
To provide greater transparency, Nvidia has begun bifurcating its AI customer revenue into two distinct cohorts: hyperscalers and non-hyperscalers (emerging cloud providers and specialized AI firms). The underlying growth dynamics between these two camps present a stark contrast:
- **Non-hyperscalers:** Generated $4 billion in revenue, surging 25% quarter-over-quarter and 138% year-over-year.
- **Hyperscalers:** Grew at a comparatively modest pace of 14% to 15% quarter-over-quarter.
This divergence demonstrates that Nvidia's future revenue momentum is tilting heavily toward non-hyperscalers. It also explains Nvidia’s recent strategic initiative: partnering with six prominent debt and asset management funds to underwrite loans for emerging cloud startups. By acting as a guarantor for the residual value of the GPUs purchased by these neoclouds, Nvidia removes capital barriers, providing the balance sheet backing necessary for these burgeoning customers to become its primary revenue engine.
[02:15](https://www.youtube.com/watch?v=ddApAM8iYmw&t=135s) The Bombshell FY28 Guidance: Mobilizing the Supply Chain
The most shocking revelation of the earnings call arrived when management delivered its revenue guidance for fiscal year 2028, projecting a staggering 70% growth rate against prevailing Wall Street consensus estimates of just 40% to 45%. During the opening round of the Q&A session, analysts were stunned. Nvidia had never before offered guidance a year and a half into the future. With the company currently in the second quarter of fiscal 2027, two quarters still remain in the current fiscal year. Normally, FY28 guidance would be held until the final quarter of FY27; Jensen Huang delivered it half a year ahead of the customary calendar.
Jensen explained that providing visibility across six full quarters—up from the historical window of just two to three quarters (six to nine months)—was vital to ensure customers, investors, and the entire hardware supply chain are aligned with the same forward roadmap. The extraordinary 70% target is not a demand ceiling; it is explicitly a supply-constrained estimate. Had supply capacity been unrestricted, Jensen noted that revenue growth could comfortably exceed 100%. By anchoring expectations now, Jensen is forcing the broader manufacturing ecosystem to ramp up capital investments in lockstep. He cited his prior intervention with Micron: had he not pushed Micron two years ago to aggressively expand memory production, the current market would be facing catastrophic memory shortages and even higher price surges.
- **Consensus vs. Guidance:** Market expected 40%–45% growth for FY28; Nvidia guided to 70%.
- **Historical Trajectory:** Revenue grew over 60% in FY26, accelerated past 80% in FY27, and remains exceptionally high at 70% in FY28 despite an increasingly massive revenue base.
- **Supply Constraint:** Guidance reflects the absolute limit of what the supply chain can physically produce, not the total addressable end-market demand.
[05:33](https://www.youtube.com/watch?v=ddApAM8iYmw&t=333s) Blackwell Ramp and the CoreWeave Revenue-Sharing Architecture
Addressing investor worries about the product transition cycle, Nvidia confirmed that Blackwell represents the "fastest product ramp in history." Virtually all incremental data center revenue over the coming quarters will be powered by this next-generation architecture, putting lingering execution concerns to rest.
Alongside hardware progress, management detailed the financial mechanics underpinning Nvidia’s deep partnership with CoreWeave. Rather than simply selling hardware off the shelf, Nvidia has structured a unique revenue-assurance model:
```mermaid
flowchart TD
accTitle: Nvidia and CoreWeave Revenue Assurance Model
accDescr: Diagram illustrating the two-tiered revenue model and downside protection between Nvidia and CoreWeave.
subgraph CoreWeave["CoreWeave Infrastructure Buildout"]
A["Constructs AI Data Center"]
B["Sells Compute to End Customers"]
end
subgraph NvidiaModel["Nvidia Dual-Revenue Model"]
C["Hardware Sales Revenue"]
D["Minimum Lifetime Revenue Guarantee"]
E["Revenue Sharing on Upside"]
end
A -->|"Purchases Hardware"| C
D -.->|"Downside Revenue Floor"| A
B -->|"Baseline Reached"| E
```
Under this arrangement, Nvidia guarantees a baseline lifetime revenue floor for a newly constructed CoreWeave data center (for instance, underwriting a $3 billion minimum revenue target). In exchange for absorbing this operational downside, Nvidia claims two distinct income streams:
1. **Upfront hardware sales:** Revenue collected from the direct purchase of GPUs, networking, and server racks.
2. **Compute revenue share:** A contracted percentage cut of all compute capacity CoreWeave sells above the baseline floor.
This mechanism acts as an insurance policy for data center buildouts. Nvidia functions as the insurer by providing a guaranteed revenue floor, and the "premium" is paid through CoreWeave sharing its excess earnings. In return, CoreWeave benefits from significantly reduced financing costs and derisked infrastructure expansion.
[06:49](https://www.youtube.com/watch?v=ddApAM8iYmw&t=409s) Gross Margin Compression: Memory Cost Inflation and the Apple Parallel
The immediate market reaction to the release was volatile: the stock initially slipped before recovering all of its losses and turning positive. The initial knee-jerk sell-off was triggered by gross margin guidance for the fourth quarter of fiscal 2027, which is projected to dip to a cyclical trough of 71%. Full-year gross margin is expected to settle near 74%, down from roughly 75% in the prior year, before rebounding to a stable 72% to 73% range in fiscal 2028.
This short-term margin compression stems from sharp, unexpected price increases in memory components (such as high-bandwidth memory). Because Nvidia's shipment volumes and contracted pricing for the current fiscal year were finalized well in advance, the company could not immediately pass through the full brunt of the memory cost surge to buyers. However, margins will stabilize once repriced GPU contracts take effect in Q1 FY28.
Xiao Tao draws a direct parallel to Apple's business dynamics: when memory component costs surge, hardware giants can pass the raw dollar cost along to the end customer, but they cannot legitimately tack on their standard massive profit markups on top of someone else's inflated component costs. Attempting to extract premium profit margins on third-party raw component inflation is neither practical nor ethical in enterprise contracting. Consequently, while total dollar profits remain protected, the gross margin percentage naturally experiences a temporary dilution.
针对企业客户对于数据安全的重重顾虑，马克·贝尼奥夫在业绩会上主动抛出了重磅定心丸——“零数据保留协议”（Zero Data Retention）。所谓零数据保留，是指企业在使用大模型运行任何内部业务数据时，大模型厂商绝对不得在后台留存、沉淀这些数据，更不允许用于模型的后续预训练，从而在架构层彻底隔绝了核心企业数据外泄的风险，打消了B端上云用AI的防御心理。
[17:55](https://www.youtube.com/watch?v=ddApAM8iYmw&t=1075s) How AI Accelerates the Enterprise Software Cycle: The Salesforce Case Study
The true structural advantage that artificial intelligence brings to established software incumbents is not replacement, but dramatic efficiency: "trial-and-error costs are drastically reduced, and what we call the product cycle—the time to launch new products—has shortened significantly." Once an enterprise software platform identifies a hit application or feature, it does not face the cold-start problem of a startup. Instead, it can immediately deploy that innovation across existing, deeply entrenched distribution rails, global enterprise sales forces, and forward-deployed engineering (FDE) teams. This structural edge translates directly into rapid deployment and accelerated top-line monetization.
Salesforce serves as an illustrative benchmark for this resilience. Despite persistent doom-and-gloom narratives around generative AI destroying legacy software moats, Salesforce’s adjusted operating margins across the past several quarters have remained remarkably stable—printing sequential figures of roughly 32%, 27%, 34%, 35%, 34%, 34.8%, and 34%—demonstrating that profitability has "remained completely unaffected." While its recent quarterly top-line expansion touched 11%, marking a slight deceleration, the more decisive operational metric is Remaining Performance Obligations (RPO). RPO captures contracted future revenue, and this quarter saw RPO expansion re-accelerate from 13% up to 14% year-over-year. This clear upward inflection indicates that enterprise client commitments are building rather than deteriorating.
```mermaid
flowchart TD
accTitle: AI Commercialization Engine in Enterprise Software
accDescr: Flowchart illustrating how AI reduces experimentation costs and leverages established distribution to drive financial re-acceleration.
subgraph R_and_D ["Product Development"]
A["Trial-and-Error Costs Reduced"] --> B["Shortened Product Launch Cycles"]
B --> C["Blockbuster AI Feature Emergence"]
end
subgraph Commercial ["Enterprise Distribution Engine"]
C --> D["Existing Sales Infrastructure"]
D --> E["Forward-Deployed Engineers (FDE)"]
end
subgraph Financials ["Financial Outcomes"]
E --> F["AI Upsells & Pricing Premiums"]
F --> G["RPO & Revenue Re-acceleration"]
G --> H["Valuation Multiple Mean Reversion"]
end
```
- Adjusted operating margins have held steady across recent quarters within the 32% to 35% band, showing zero margin degradation from AI disruption.
- Revenue growth registered at 11% in the latest quarter, while forward-looking RPO growth climbed to 14% year-over-year, up from 13% in the prior quarter.
- Traditional software revenue models historically relied on periodic acquisitions to maintain a baseline 10% trajectory; integrated AI offerings now provide organic upsell levers.
[19:00](https://www.youtube.com/watch?v=ddApAM8iYmw&t=1140s) Software Valuations and the Flawed "Terminal Value" Bear Thesis
Current consensus forecasts project Salesforce's top-line revenue to expand by 11%, 9%, and 10% over the next two and a half years—an average pace right around 10% that merely mirrors its historical acquisition-driven baseline. Bottom-line EPS growth is expected to pace at 13%, 11%, and 17%, averaging roughly 13% to 14%. However, these estimates underestimate how direct AI feature integration allows enterprise vendors to capture meaningful pricing premiums and upsells. With a trailing price-to-earnings ratio near 13, the resulting PEG ratio hovers around 1, which represents an exceptionally discounted multiple for cash-generative enterprise software platforms.
The sector's disconnect is even more glaring when analyzing cash generation. Across the software complex, price-to-cash flow multiples have compressed to approximately 10 times, with forward projections pointing to multiples of 11, 10, and 8.8 times over the next three years. Put into plain perspective, "if everything remains unchanged, the cash they earn over 10 years would be enough to buy back their entire company." This multiple compression stems from the thesis peddled by Wall Street short-sellers six months ago claiming that software companies' "terminal values were affected—specifically, that their revenue growth five or 10 years down the line would suffer." Yet that thesis remains purely speculative and unproven: after two full years of generative AI adoption, established enterprise software margins and revenues have remained rock solid. As long as these firms continue to execute, Wall Street's doomsday narrative cannot logically sustain itself for three or four years against contradictory financial realities.
- Consensus bottom-line EPS projections stand at 13%, 11%, and 17% over the coming three years (averaging 13% to 14%), generating a PEG ratio near 1.
- Price-to-cash flow multiples sit near 10x (forward estimates of 11x, 10x, and 8.8x), creating an extreme margin of safety where a decade of cash flow equals total market capitalization.
- Realistic mean reversion targets for software valuations lie between 18 to 20 times earnings, providing substantial upside from current compressed levels without requiring a return to historical peaks of 25 to 30 times.
[21:29](https://www.youtube.com/watch?v=ddApAM8iYmw&t=1289s) Sentiment Layering and Upcoming Member Livestream
A decisive turning point for software sentiment has quietly begun, led by Microsoft’s resilient earnings print and reinforced by broader peer reports. The market's constructive reaction reflects two dynamics: institutional investors were severely under-positioned across the sector, and market participants are waking up to the reality that software is not being cannibalized wholesale by foundation models. Much like the multi-year rally in AI hardware infrastructure, restoring software valuations will not occur overnight off a single earnings report; it demands "a layering of sentiment and a layering of fundamentals" over several consecutive quarters of strong operational execution.
Looking ahead, a member-exclusive livestream is scheduled inside the APB app for this Thursday at 9:30 PM, following up on the recent public streams held on Monday and the preceding week. The dedicated session will provide space to dissect key executive interviews, field detailed macro questions, and analyze after-the-bell quarterly results from Marvell Technology (MRVL) if time permits. Viewers are encouraged to download the application via the on-screen QR code to explore its suite of free analytical tools before deciding whether to upgrade to a full membership tier.
- Market sentiment requires consecutive quarters of compounding data; hardware's sustained bull run was built on continuous execution and sentiment layering, which software is now beginning to replicate.
- Upcoming member livestream: Thursday at 9:30 PM on the APB app, featuring deeper dives into macro questions and post-market earnings analysis for Marvell Technology (MRVL).
- Viewers can scan the on-screen QR code to download the platform, which includes free core features accessible without a paid subscription.
## Notable quotes
> [05:33](https://www.youtube.com/watch?v=ddApAM8iYmw&t=333s) fastest product ramp in history
> [17:55](https://www.youtube.com/watch?v=ddApAM8iYmw&t=1075s) trial-and-error costs are drastically reduced, and what we call the product cycle—the time to launch new products—has shortened significantly
> [17:55](https://www.youtube.com/watch?v=ddApAM8iYmw&t=1075s) remained completely unaffected
> [19:00](https://www.youtube.com/watch?v=ddApAM8iYmw&t=1140s) if everything remains unchanged, the cash they earn over 10 years would be enough to buy back their entire company
> [19:00](https://www.youtube.com/watch?v=ddApAM8iYmw&t=1140s) terminal values were affected—specifically, that their revenue growth five or 10 years down the line would suffer
> [21:29](https://www.youtube.com/watch?v=ddApAM8iYmw&t=1289s) a layering of sentiment and a layering of fundamentals
[5:33] fastest product ramp in history
[10:39] 软件末日论
[10:39] 大脑层
[10:39] 工具层
[13:28] 马克·贝尼奥夫（Marc Benioff）: 底层大模型本质上是可替代的大宗商品（Commodities）
[13:28] 双轨战略
[13:28] 管道与被调用的螺丝钉
[13:28] 以攻代守
[15:16] 马克·贝尼奥夫（Marc Benioff）: 零数据保留协议
[15:16] 软件大年
[17:55] trial-and-error costs are drastically reduced, and what we call the product cycle—the time to launch new products—has shortened significantly
[17:55] remained completely unaffected
[19:00] if everything remains unchanged, the cash they earn over 10 years would be enough to buy back their entire company
[19:00] terminal values were affected—specifically, that their revenue growth five or 10 years down the line would suffer
[21:29] a layering of sentiment and a layering of fundamentals
Original transcript
[0:07] Welcome everyone to a professional US
[0:08] stock channel that goes beyond just
[0:10] reporting the news. I am your market
[0:11] analyst, Xiao Tao. Even though the
[0:13] earnings season is behind us, I don't
[0:15] know why I still feel incredibly busy.
[0:16] It was already midnight when I finished
[0:18] listening to the earnings reports of
[0:19] these two companies to record this
[0:21] video for you, so given this level of
[0:22] passion, I feel a great sense of
[0:23] responsibility, right? To come here and
[0:25] put together this episode for you all
[0:26] today. And because of that, shouldn't
[0:28] everyone start by hitting like, leaving
[0:29] a comment, and sharing? Alright, I
[0:31] won't waste any more time; let's get
[0:32] straight into it. Today I will only
[0:34] discuss company earnings; I won't cover
[0:35] other things or general market trends,
[0:37] including certain data points. I
[0:38] believe macro data isn't as important
[0:39] as the wealth of information revealed
[0:41] behind these two companies 'earnings
[0:42] reports. Let's start by looking at
[0:44] Nvidia's earnings. Let's break it down
[0:46] point by point. First, in Nvidia's
[0:47] report, they mentioned hyperscalers,
[0:49] with an RPO of 2 trillion—that is,
[0:51] the combined orders for all
[0:53] hyperscalers to be fulfilled in the
[0:55] future amount to 2 trillion. Capital
[0:57] expenditure for '26 is 800 billion,
[0:59] with '27 capital expenditure projected
[1:00] to reach 1.3 trillion. Looking at the
[1:02] year-over-year growth rate, the
[1:04] year-over-year growth in capital
[1:06] expenditure for '27 will actually be
[1:08] slower compared to '26. '26 is nearly
[1:10] doubling, but if you calculate the
[1:12] year-over-year growth for '27 by
[1:14] dividing 1.3 trillion by 800 billion
[1:16] and comparing it to '26, you will see a
[1:18] slowdown in the year-over-year growth
[1:20] rate. Furthermore, in Nvidia's earnings
[1:22] report, they have now separated their
[1:24] AI orders, categorizing them into
[1:25] hyperscalers and non-hyperscalers.
[1:27] Let's take a look: non-hyperscaler
[1:29] revenue growth is 4 billion, with a
[1:30] quarter-over-quarter growth of 25%and a
[1:32] year-over-year growth of 138%, while
[1:34] the growth for hyperscalers is around
[1:36] 14%to 15%quarter-over-quarter. So,
[1:37] clearly, the quarter-over-quarter
[1:39] growth of non-hyperscalers has
[1:40] surpassed that of hyperscalers. What
[1:42] does this indicate? It indicates that
[1:44] moving forward, Nvidia's customer
[1:46] revenue growth will primarily come from
[1:48] non-hyperscaler companies. This is also
[1:50] how everything connects, and why Nvidia
[1:52] recently joined forces with six major
[1:54] fund companies to provide loans, with
[1:57] Nvidia acting as a guarantor for the
[1:59] residual value of GPUs for these
[2:00] emerging cloud companies—why is that?
[2:03] It is because they want to support
[2:05] these new companies from a financing
[2:07] perspective, as these companies will
[2:09] become the main driver of their revenue
[2:11] growth in the future. Then, the third
[2:15] point from this earnings call might be
[2:17] considered a bombshell news or guidance
[2:19] : the revenue growth forecast for
[2:21] fiscal year 2028 is 70%, while the
[2:23] market expectation was only...so why
[2:26] would he do that? Looking at the first
[2:29] point in the Q&A, Jensen indeed
[2:31] mentioned—when asked by an analyst—
[2:33] that they had never provided guidance
[2:35] for a year and a half out before. We
[2:37] have just entered their second quarter
[2:39] of fiscal year 2027. Their fiscal year
[2:41] differs from the calendar year, so
[2:43] there are two quarters left in FY27,
[2:45] yet they are already providing guidance
[2:47] for FY28. Normally, you would provide
[2:50] guidance for FY28 in the final quarter
[2:52] of FY27, but they are essentially doing
[2:54] it half a year in advance. Why would
[2:56] they choose to do that? Jensen said he
[2:58] wanted their customers, supply chain,
[3:01] and investors to all have the same
[3:03] information, because it is extremely
[3:05] important. Why? Because many people
[3:07] across the entire supply chain are
[3:09] currently committing a significant
[3:11] amount of resources. My thought is that
[3:13] the market expected 40%, but Jensen
[3:15] provided 70%. In reality, he could have
[3:18] easily followed the market expectations
[3:20] and delivered even higher results. But
[3:22] why did he choose to give this 70%
[3:24] guidance directly now? It is first
[3:26] because, as he noted, demand is massive
[3:28] . He mentioned that this 70%guidance is
[3:31] still constrained by supply-side
[3:32] factors, because our demand will exceed
[3:34] this supply capacity. If there were no
[3:37] supply-side constraints whatsoever,
[3:39] their growth rate could easily reach
[3:41] over 100%. The reason he is doing this
[3:43] is because he wants the entire supply
[3:45] chain to keep up, which is exactly what
[3:47] Jensen has been working on for the past
[3:49] two years. In this earnings call, he
[3:52] mentioned that if he hadn't brought
[3:54] Micron along to expand production back
[3:56] then, at this current stage, there
[3:57] would definitely be an even greater
[3:59] shortage of memory, and the price
[4:01] increases we’ve seen would be even
[4:03] higher than what we’ve experienced
[4:05] recently. So, he is very glad that he
[4:07] led the entire supply chain to move
[4:09] forward together at that time. The same
[4:11] logic applies now; since NVIDIA is the
[4:13] leader in GPUs, he wants to signal to
[4:15] the entire industry what the projected
[4:18] growth rate is for fiscal year 2028,
[4:20] hoping the whole supply chain and
[4:22] industry will expand alongside them.
[4:24] This point is crucial and quite rare,
[4:26] because since 2023, Jensen usually only
[4:28] provides guidance for the next six to
[4:30] nine months—two to three quarters—
[4:32] because their visibility on order
[4:33] demand typically only extends about
[4:35] half a year to three quarters. Yet this
[4:37] time, how much did he reveal? He’s
[4:39] looking a year and a half out,
[4:40] revealing he already knows the demand
[4:42] for the next six quarters. So, this is
[4:44] indeed a very important signal, a
[4:45] bombshell that has left the market very
[4:48] surprised. Because market expectations
[4:49] for revenue growth are currently at 45%
[4:51] ; for fiscal year 2026, which just
[4:53] ended, revenue growth was over 60%, and
[4:55] this year, growth is expected to be
[4:56] over 80%. You’ll notice that despite
[4:58] such massive scale, their revenue
[5:00] growth is actually accelerating;
[5:02] although 2028 might show a slight dip
[5:04] compared to 2027, the reason for that
[5:06] decrease is solely due to supply-side
[5:08] constraints. If one had to pick a key
[5:10] highlight, it would be the 2028 revenue
[5:13] guidance, which is truly incredible—
[5:15] 70%revenue growth at that scale; which
[5:17] other company in the world could
[5:19] achieve that? No one, only Jensen, and
[5:21] the reverse is also true. If he weren't
[5:24] leading the entire supply chain forward
[5:26] , Jensen himself could not achieve 70%
[5:28] revenue growth; it would be absolutely
[5:30] impossible. Moving on, the fifth point:
[5:33] Blackwell will become the fastest
[5:35] product ramp in history, which also
[5:37] alleviates some market concerns, as
[5:39] essentially all revenue for the coming
[5:41] quarters will come from Blackwell, the
[5:43] new generation of GPUs. Next, the sixth
[5:45] point regarding the cooperation details
[5:47] between NVIDIA and CoreWeave has also
[5:49] been disclosed this time. They stated
[5:51] that they guarantee a minimum revenue
[5:53] for the data center; for you, CoreWeave
[5:55] , if you build a data center, we can
[5:57] guarantee a minimum total revenue for
[5:59] that data center. For example, if we
[6:01] guarantee this data center will
[6:02] generate 3 billion in lifetime revenue.
[6:05] Then, if the revenue ultimately exceeds
[6:07] those 3 billion, we will share the
[6:08] excess with CoreWeave based on a
[6:10] certain ratio. So, from NVIDIA's
[6:11] perspective, if the data center is
[6:13] profitable or meets the minimum revenue
[6:15] requirement, it can actually collect
[6:17] two streams of income: the first is the
[6:19] money from selling the hardware, and
[6:21] the second is the money from CoreWeave.
[6:23] For all the computing power that
[6:25] CoreWeave sells, NVIDIA gets a cut, so
[6:27] the entire cooperation detail can be
[6:29] understood as what? It can be
[6:31] understood as NVIDIA providing
[6:32] insurance for the data center's total
[6:33] revenue, offering this minimum revenue
[6:35] guarantee, so what is the premium? The
[6:37] premium is that CoreWeave needs to pay
[6:39] a portion of the excess revenue to
[6:41] NVIDIA, and for CoreWeave, this is
[6:42] still a positive, why? Because with
[6:44] NVIDIA's guarantee, CoreWeave's cost of
[6:46] building data centers will be reduced.
[6:49] Moving to the seventh point, which is
[6:51] also something the market might be
[6:52] concerned about. To be honest, if
[6:54] NVIDIA hadn't released the revenue
[6:56] guidance for fiscal year 2028, the
[6:57] market might have fallen after this
[6:59] quarterly earnings report was released.
[7:01] Of course, right after the earnings
[7:02] report or the data came out, the stock
[7:04] did drop, but then the decline was
[7:06] erased, and now it has risen back. The
[7:08] initial decline was because the gross
[7:09] margin for the fourth quarter, the
[7:11] fiscal year 2027 fourth quarter, will
[7:13] see a decrease. Why would it decrease?
[7:14] It is because of the rise in storage
[7:16] prices, as the magnitude of this
[7:18] increase exceeded expectations. Since
[7:20] it exceeded expectations and their
[7:21] shipment volumes and prices for this
[7:23] year are already set, they are unable
[7:25] to fully pass on the storage price
[7:27] increases to their customers. That is
[7:29] also why the gross margin in the fourth
[7:31] quarter will reach a low point of 71%,
[7:33] while for the full year, the market
[7:35] expectation is around 74%, and last
[7:37] year it was around 75%, so the gross
[7:39] margin shows a certain degree of
[7:40] decline. Then the company expects gross
[7:42] margins to stabilize between 72%and 73%
[7:44] in fiscal year 2028. Why? It is because
[7:46] starting from the first quarter of
[7:48] fiscal year 2028, their new GPU pricing
[7:49] will be reflected in their revenue
[7:51] reports. So, after their explanation, I
[7:53] think this short-term dip in gross
[7:55] margin is acceptable. It’s actually
[7:58] the same as Apple; if you look at Apple
[8:00] , their gross margins will also decline
[8:03] soon. Why? Because they simply cannot
[8:05] —it's not just about passing the
[8:07] memory costs onto consumers; if you
[8:09] want to maintain the original gross
[8:11] margin, what must you do? You would
[8:13] have to make a profit on top of the
[8:15] memory cost itself. In other words,
[8:18] even though you didn't produce the
[8:19] memory, you would be treating it as
[8:21] your own product and selling it at a
[8:23] premium. That isn't reasonable, and
[8:25] it’s not ethical, right? So that is
[8:28] why for both companies, whether it’s
[8:30] Nvidia or Apple, they can pass the rise
[8:32] in memory costs entirely to their
[8:34] customers, but they cannot earn higher
[8:36] premiums while these costs are rising.
[8:39] Therefore, once the memory costs are
[8:41] added to their own product costs, a
[8:43] short-term dip in gross margin will
[8:46] occur, okay? And the last point is some
[8:48] industry data that we can update, as
[8:50] Jensen just provided some new figures.
[8:52] Five years ago, a one-gigawatt data
[8:55] center cost about 30 billion. Now,
[8:57] after they launched Blackwell, the cost
[8:59] for a one-gigawatt data center is about
[9:01] 60 billion. So, a while ago I heard
[9:03] JPMorgan say a one-gigawatt data center
[9:05] now costs 100 billion; that figure is
[9:07] too high. Jensen just gave the latest
[9:09] data, which is around 60 billion, or 60
[9:11] billion dollars. After this revenue
[9:13] guidance was provided, let’s update
[9:15] the valuation. In yesterday's video, I
[9:17] talked about the title of my video,
[9:19] which was that Nvidia’s valuation is
[9:20] cheap. Now, if you look at the fiscal
[9:22] year 2028 revenue guidance provided,
[9:24] you will find that if you look at its P
[9:26] /E ratio, Nvidia’s valuation has
[9:28] become even cheaper, it has become even
[9:29] more affordable. Based on market
[9:31] expectations, just like the EPS growth
[9:33] I mentioned earlier, the revenue and
[9:35] EPS growth are actually the same. The
[9:37] market assumption for fiscal year 2028
[9:39] is a 45%revenue growth and 46%EPS
[9:41] growth, assuming no gains in
[9:43] operational efficiency—meaning profit
[9:46] and EPS grow in line with revenue. Thus
[9:49] , the market projects 46%EPS growth for
[9:52] FY28. Now, if we factor in that 70%
[9:54] growth rate, the original FY28 P/E
[9:56] ratio was around 16 times, or 15.9
[9:59] times to be exact. If we plug that 70%
[10:01] growth rate in, what does the FY28 P/E
[10:03] become? 13.7 times. 14 times, what does
[10:07] that imply? It means if we assume that
[10:09] over the next year and a half the stock
[10:11] price remains flat and the earnings
[10:13] reports perfectly match market
[10:14] expectations, its static P/E ratio is
[10:16] 14. What kind of growth rate does a 14x
[10:19] static P/E match? A 14x static P/E
[10:21] matches a future EPS growth rate of
[10:23] around 10%. If that growth stays around
[10:25] 10%in the long term, you could argue
[10:28] that a 14x static P/E is reasonable.
[10:30] Otherwise, for any EPS growth rate
[10:32] above 20%, a 14x static P/E is an
[10:34] undervaluation, or looking at it now, a
[10:37] 14x P/E is significantly undervalued.
[10:39] Okay, now let's talk about the recent
[10:41] CRM earnings report. First, a rather
[10:43] important piece of news from their
[10:45] earnings call was the partnership with
[10:47] Anthropic. This collaboration marks a
[10:50] new model for the future of AI and
[10:52] software companies: AI won't completely
[10:54] disrupt software companies; instead,
[10:57] software will become part of the AI
[10:59] ecosystem. The AI will handle the "
[11:01] brain" layer, and software will handle
[11:03] the "tool" layer, where the brain calls
[11:06] upon these tools. This is the direction
[11:08] we previously envisioned for AI agents
[11:10] and their underlying tools. This is
[11:13] also why, after checking the news
[11:15] regarding the CRM and Anthropic
[11:16] partnership, I believe it's reasonable
[11:19] that the stock price of WEN dropped.
[11:21] The reason is that from CRM's
[11:22] perspective, they believe Anthropic
[11:24] will become the orchestration layer for
[11:27] future AI agents—it will become a "
[11:29] brain" that sits on top of the software
[11:31] , controlling the software below. So,
[11:33] if Anthropic can achieve that position,
[11:36] it will certainly pose a certain level
[11:38] of threat to Microsoft. I think it is
[11:40] reasonable that Microsoft's stock price
[11:42] saw some decline after hours. Of course
[11:44] , as to whether this will ultimately
[11:46] cause massive disruption or impact for
[11:48] Microsoft, I don't think it will,
[11:50] because Microsoft will also have its
[11:51] own AI agents. Let's get back to the
[11:53] CRM side of things. Regarding the
[11:55] partnership between CRM and Anthropic,
[11:57] they have launched a new product called
[11:58] ClaudeForce, and they have developed 37
[12:00] skills. How should we interpret this?
[12:02] It means that within Anthropic's Claude
[12:04] , they have developed 37 CRM products,
[12:07] meaning if you enter Claude now, you
[12:09] can fully utilize CRM products through
[12:11] Claude. So, as I mentioned earlier, CRM
[12:14] believes that the future gateway for
[12:16] the B-end market is very likely to be
[12:18] Claude. So, who is the gateway right
[12:21] now? The current gateway is CRM's own
[12:23] user interface. The partnership between
[12:26] CRM and Anthropic stems from their
[12:28] belief that a portion of future B-end
[12:30] traffic will directly come through
[12:32] Claude, which is why they chose to
[12:33] collaborate with them. Therefore, these
[12:36] two companies will have a division of
[12:38] labor going forward: CRM will be
[12:39] responsible for sales and distribution,
[12:41] leveraging their many salespeople to
[12:43] promote this product. CRM and Anthropic
[12:46] will then enter a state of coexistence,
[12:48] with Anthropic serving as the brain and
[12:50] the gateway, and CRM acting as the tool
[12:53] and distribution channel—for now, it
[12:55] appears to be a partnership based on
[12:57] supply. However, as I said, there is a
[13:00] risk that in the future, CRM might lose
[13:03] this UI and lose this gateway. If they
[13:06] lose the gateway, there is a certain
[13:08] risk, considering that the trust
[13:10] currently built between CRM and their
[13:12] clients exists because they have a
[13:14] direct connection. But if they lose
[13:17] that gateway in the future, it
[13:19] essentially means the clients
[13:20] themselves will form an interdependent
[13:22] relationship with Anthropic. So, I
[13:25] think this partnership between CRM and
[13:27] Anthropic can be viewed from two angles
[13:29] ; one perspective is that internally,
[13:32] AI itself is not solely reliant on
[13:34] their partnership with Anthropic.
[13:36] It’s because we know their internal
[13:38] view, Mark’s consistent belief, is
[13:40] that models are commodities and are
[13:42] interchangeable. So, their internal
[13:44] products offer a choice; whether their
[13:46] customers want to use open-source
[13:48] models, Claude, or OpenAI, Salesforce
[13:50] provides these options. You can choose
[13:54] the underlying model; this is on the
[13:56] product side, within their own user
[13:58] interface. Within their existing
[14:01] products, you can select different
[14:03] models to perform different tasks.
[14:06] Building on this, they chose another
[14:08] approach: they believe Claude could
[14:10] become a gateway for enterprise
[14:11] endpoints. They don’t want a
[14:15] situation where, if a client’s
[14:17] primary entry point is Claude, they
[14:19] can’t access Salesforce products—so
[14:21] they must ensure the client can still
[14:23] choose Salesforce. Do you understand?
[14:26] They view Claude as a distribution
[14:28] channel. If Claude becomes a gateway in
[14:31] the future, they want to ensure that
[14:32] Salesforce products are available on
[14:34] that platform. There is a risk here:
[14:40] currently, customers rely on Salesforce
[14:42] products, but in the era of AI agents,
[14:45] customers might rely on the agents
[14:47] themselves rather than the Salesforce
[14:49] product, right? So, if that happens, it
[14:54] might not necessarily be a long-term
[14:56] benefit for Salesforce. While this
[14:59] partnership is definitely a positive in
[15:01] the short to medium term, it’s not
[15:03] necessarily a win from a long-term
[15:05] perspective. Of course, this could also
[15:07] be seen as a defensive move for
[15:08] Salesforce rather than an offensive one
[15:10] , because they need to defend. Defend
[15:12] against what? Defending against the
[15:13] possibility that Anthropic could become
[15:15] a giant in the future. Next, let’s
[15:16] look at the second point, which wasn't
[15:18] asked during the earnings call, but
[15:19] Mark brought it up himself. In their
[15:21] partnership with Anthropic, there is a
[15:23] crucial issue that has been widely
[15:25] discussed recently: data privacy in AI.
[15:27] Are customers willing to share their
[15:29] data with Claude? Mark touched on this,
[15:30] which is why I’m asking: why do
[15:32] customers feel safe using it? Why are
[15:33] they confident enough to use it? Aren't
[15:35] they afraid that their data might be
[15:36] taken by Anthropic? So, Mark himself
[15:38] mentioned it. He asked, what does zero
[15:40] data retention mean? Zero data
[15:42] retention means that any internal
[15:44] company data you run through their
[15:46] large models will absolutely not be
[15:47] retained by the model. The model won't
[15:51] keep any of our customer data, which is
[15:53] why it's called Zero Data Retention.
[15:55] CRM ensures that data won't leak. And
[15:58] lastly, the second point in the Q&A
[16:00] where their management mentioned that
[16:02] currently, only 5%of customers have
[16:04] upgraded to the latest version, called
[16:06] the Premium tier. This version of the
[16:08] product has a 60%to 80%price premium,
[16:10] so they believe there is huge room to
[16:12] help existing customers upgrade, as
[16:13] they are very bullish on the current
[16:15] version—that is, the version upgraded
[16:18] with AI integration. So, in fact, up
[16:19] until now, we have already looked at
[16:21] this quarter, we have already looked at
[16:23] the earnings reports of many software
[16:25] companies. By now, I am slowly
[16:27] realizing this issue, and I hope
[16:29] everyone realizes it too: it's only
[16:30] starting from this quarter that we are
[16:32] gradually discovering the integration
[16:34] between AI and these software companies
[16:36] ' products, and we have already seen it
[16:38] land—we are actually seeing customers
[16:40] willing to pay for these AI software
[16:41] products. Therefore, I believe this
[16:43] quarter is truly the first quarter
[16:45] where software companies are monetizing
[16:46] their integration with AI. It can be
[16:48] considered the first quarter of the
[16:49] inaugural year, just getting started.
[16:51] To some extent, the market actually
[16:53] thought back in late 2023 and early
[16:55] 2024 that 2024 would be the year of
[16:57] software, because software would
[16:59] benefit from AI, given that AI's
[17:00] implementation is centered on software.
[17:02] As a result, software kept falling from
[17:04] 2024, falling for two and a half years,
[17:06] until roughly the middle of this year
[17:08] —from the beginning of 2024 all the
[17:10] way down to June 2026. So, throughout
[17:12] these two and a half years, the market
[17:14] clearly misjudged, anticipating the
[17:16] benefits of AI for software too early.
[17:18] In reality, it is only this quarter
[17:20] that we are truly seeing that AI seems
[17:22] not to be a headwind for software
[17:24] companies, but a tailwind. Of course,
[17:26] this benefit is predicated on the
[17:27] company's ability to effectively
[17:29] integrate AI into its own products and
[17:31] then upsell them to customers at a
[17:32] higher price. This entirely depends on
[17:34] the company's execution and its
[17:36] capability to seamlessly merge its
[17:38] existing products with AI and
[17:39] successfully bring them to market.
[17:41] Therefore, the cost of trial and error
[17:43] has also dropped significantly. You'll
[17:45] notice that every software company now
[17:47] launches new products much faster; what
[17:49] used to be two or three features a year
[17:51] is now a dozen features per quarter.
[17:53] Through this continuous trial-and-error
[17:55] process, once they hit on a blockbuster
[17:57] product, they push it out aggressively.
[17:59] With their existing distribution, sales
[18:01] teams, and FDE, the deployment speed is
[18:03] incredibly fast, leading to rapid
[18:04] revenue growth for these software
[18:06] companies. This is the true benefit
[18:08] that AI brings to these software
[18:09] companies: trial-and-error costs are
[18:11] drastically reduced, and what we call
[18:13] the product cycle—the time to launch
[18:15] new products—has shortened
[18:16] significantly. For example, Salesforce
[18:18] is a great case. I've plotted their
[18:20] adjusted margins below; look at their
[18:23] profit margins over the past five
[18:25] quarters: 32, 27, 34, 35, 34, 34.8, 34.
[18:27] They have remained completely
[18:29] unaffected. As for their revenue growth
[18:31] , they have been making acquisitions,
[18:33] so as a relatively traditional company,
[18:35] CRM’s revenue growth is largely
[18:36] driven by those acquisitions. Now, in
[18:38] this past quarter, the revenue growth
[18:40] rate was 11%, which is a slight
[18:42] slowdown compared to the previous
[18:44] quarter, but their RPO is more
[18:45] important because RPO represents future
[18:48] revenue growth. RPO saw a 14%
[18:49] year-over-year growth, up from 13%in
[18:51] the previous quarter, so we are seeing
[18:53] a re-acceleration in RPO. So, my
[18:55] expectation after listening to this
[18:56] earnings call is that their RPO growth
[18:58] will continue to trend upward. With
[19:00] these perspectives in mind, let’s
[19:02] finally take a look at their valuation.
[19:03] The market currently expects their
[19:05] revenue growth for the next two and a
[19:06] half years to be 11%, 9%, and 10%,
[19:07] which is very much in line with CRM's
[19:09] typical revenue profile. Their revenue
[19:10] basically stays around 10%, and
[19:12] whenever growth slows down, they just
[19:13] make an acquisition, right? But things
[19:15] are different now; it’s about the
[19:17] integration between products and AI to
[19:18] drive upsells. By selling these AI
[19:20] products, the bottom-line growth is
[19:22] projected at 13%, 11%, and 17%. Well,
[19:24] one of my expectations is that their
[19:26] revenue growth will see an acceleration
[19:28] . Revenue growth is currently averaging
[19:30] about 10%, exactly 10%over the last two
[19:32] and a half years, and I expect that
[19:34] growth to re-accelerate. Because the
[19:37] integration of AI with their existing
[19:40] products will lead to upsells and a
[19:42] premium on the product side, which
[19:44] aligns with their revenue and EPS
[19:47] growth—EPS averages about 13%to 14%,
[19:49] and with a trailing P/E of 13, the PEG
[19:52] ratio is around 1. It is considered to
[19:54] be on the low side. If you look at the
[19:56] price-to-cash flow, it's 1.7 times, and
[19:59] for the next three years, it's roughly
[20:01] 11, 10, and 8.8 times; most software
[20:03] companies currently have a
[20:05] price-to-cash flow ratio around 10.
[20:07] Basically, if everything remains
[20:08] unchanged, the cash they earn over 10
[20:10] years would be enough to buy back their
[20:12] entire company. My expectation is that
[20:14] I still believe software companies are
[20:16] currently undervalued in terms of their
[20:18] valuations. I expect that companies
[20:20] like CRM, along with many other
[20:22] software firms, can at least see their
[20:24] valuations return to former levels—
[20:25] not necessarily 25 or 30 times, but at
[20:27] least back to the 18 to 20 times range,
[20:29] which I consider a reasonable zone.
[20:31] Building on that, since their
[20:33] fundamentals continue to change, if
[20:35] they keep improving, I don't think
[20:37] it’s impossible for them to return to
[20:39] previous valuation levels or former P/E
[20:42] ratios. I said half a year ago that the
[20:44] reason the market was selling off
[20:46] software was the belief that these
[20:48] companies 'terminal values were
[20:50] affected—specifically, that their
[20:51] revenue growth five or 10 years down
[20:53] the line would suffer—but that view
[20:56] itself is something you cannot prove or
[20:58] verify. Over the past two years,
[21:00] we’ve seen that for many software
[21:02] companies, their revenue hasn't been
[21:04] impacted at all, and their margins
[21:05] remain unaffected, so on what grounds
[21:07] do you assume your assessment is
[21:09] correct? That's why I say as long as
[21:11] companies continue to execute, the
[21:12] story Wall Street is currently selling
[21:14] will eventually be proven wrong. You
[21:16] can't keep running this kind of
[21:17] narrative for three or four years. It's
[21:19] impossible for software companies to
[21:21] keep executing, where their revenue
[21:22] shows no slowdown, shows no disruption
[21:24] from AI, and margins aren't hit by AI,
[21:25] and then expect this story to keep
[21:27] running for three or four years.
[21:29] Therefore, I think this quarter,
[21:30] starting with Microsoft releasing their
[21:32] earnings first, followed by many other
[21:34] software companies, you can see that
[21:35] the market's feedback has been
[21:37] relatively positive. Why? First, the
[21:39] primary reason is a lack of positioning
[21:41] , and second, the market is slowly
[21:42] realizing that it seems not all
[21:44] software companies are being affected
[21:46] or disrupted by AI. As I said, this
[21:47] requires a layering of sentiment and a
[21:49] layering of fundamentals; it won't be
[21:50] just one company's earnings report that
[21:51] immediately brings the software sector
[21:53] back to life. That won't happen. It’s
[21:55] the same with hardware. The reason
[21:56] hardware has performed so well is due
[21:58] to a continuous layering of market
[21:59] sentiment and earnings results. So, I
[22:01] think most software companies have
[22:02] reported decent earnings this quarter,
[22:04] and I hope this continues into the next
[22:06] quarter, as I believe valuations in the
[22:07] software sector will gradually recover.
[22:09] Alright? That’s all for today's
[22:10] sharing. A quick heads-up: this
[22:12] Thursday at 9:30 PM, I will be hosting
[22:14] a livestream exclusively for members
[22:16] within the APB app. We did a public
[22:17] livestream this past Monday, and one
[22:19] the week before that, but this Thursday
[22:21] will be for members. The main reason is
[22:23] that I can dive deeper into that
[22:24] interview, and address any questions
[22:26] you might have, including the MVEL
[22:28] earnings after the bell on Thursday. If
[22:29] I have time, I’ll listen to the MVEL
[22:31] earnings call and share my thoughts
[22:33] with everyone during tomorrow's
[22:34] livestream, along with other macro
[22:36] topics and questions you may want to
[22:37] ask during the member stream, okay? The
[22:39] QR code on the screen is the download
[22:41] link, and even if you don’t join
[22:42] tomorrow's livestream, that’s
[22:43] perfectly fine. I encourage everyone to
[22:45] download it first. Once downloaded,
[22:47] you'll find that many of the features
[22:48] inside are actually very useful and
[22:49] completely free. After you've
[22:50] downloaded it and given it a try, then
[22:52] you can consider joining as a member.
[22:54] Alright? That’s it for today's
[23:01] content.
