#Gate股票观点挑战 +$MRVL Nvidia is the bellwether, but the bigger hidden signal is in another company's earnings report—Marvell Technology reports tonight, marking the first real valuation test of the AI inference era

I. Two Earnings Reports, but the Market Is Watching This One More Closely
In the early hours of August 27 (Beijing time), Nvidia released its Q2 earnings report. All eyes were on it. But after the U.S. market close on the same day, another company will release an equally important but far less-watched earnings report. That company is Marvell Technology (Marvell, NASDAQ: MRVL).
If you understand that the real AI battleground is inference infrastructure, not the model layer—you will know that Marvell's earnings report can tell us more than Nvidia's about how far this competition has progressed.
II. The Results of Nvidia's Earnings Report In the early hours, Nvidia delivered this report card: revenue doubled year over year (+106%), and its Q3 guidance exceeded expectations by nearly $4B, with FY28 projected at +70%. The market reaction was a more than 5% rise in NVDA after hours.
This earnings report demonstrates one thing: demand for AI computing power is not slowing; it is accelerating. There is still no sign that hyperscalers' capital expenditures have peaked. Nvidia's figures are important, but they tell us about the overall scale of demand on the training side. Marvell tells us how far customization on the inference side has progressed. These are two different questions.

III. Who Is Marvell
Many people have heard of Nvidia but are unfamiliar with Marvell
Marvell Technology is a global leading semiconductor company focused on designing and developing chips for data infrastructure. Key information about the company includes:
1. Company Overview
· Founded in 1995 and headquartered in Santa Clara, California, it was co-founded by three Chinese-American engineers (Weili Dai, Sehat Sutardja, and Pantas Sutardja). It initially started with hard-disk controller chips before gradually transforming into a core supplier in the data infrastructure sector.
· Its current market capitalization is approximately $192 billion (as of 2026). It is one of the world's leading fabless semiconductor design companies, with products spanning data centers, network communications, storage, and other fields.
2. Core Businesses and Technological Advantages
· Optical interconnect chips: A global leader in optical DSP chips, holding over 60% of the core-chip market for 800G/1.6T optical modules. Its technology covers cutting-edge fields including PAM4, coherent DSP, and co-packaged optics (CPO), providing high-speed optical interconnect solutions for AI data centers.
· Custom ASIC chips: Designs customized AI chips for cloud providers such as Google, Amazon, and Microsoft (such as TPU and Trainium), optimizing power consumption and performance to meet specific AI training and inference requirements.
· Networking and switching chips: Provides data center Ethernet switching chips, PCIe Retimer, and CXL memory expansion technology, supporting networking and memory pooling for large-scale AI clusters.
3. Industry Position and Partnerships
· Works closely with Nvidia to jointly develop the NVLink Fusion architecture and advance the development of rack-scale AI interconnect technology.
· Has been called “the next trillion-dollar company” by Nvidia CEO Jensen Huang. Due to its core position in AI interconnects, it is regarded as a key participant in AI infrastructure.
4. Development History
· Starting with hard-disk controller chips, it gradually became an integrated supplier in the data infrastructure sector through acquisitions (such as Inphi and Innovium) and technological transformation, with data center business revenue accounting for over 75%. Leveraging its deep expertise in optical interconnects, custom chips, and networking technology, Marvell Technology has become a core enabler of data movement in the AI era, and its technological capabilities and industry influence continue to attract market attention.

ASIC—The Hyperscalers' “Exclusive Arms Dealer”

What does ASIC (custom inference chip) mean?
Nvidia's GPU is a general-purpose chip, like a universal tool—it can run any AI task, but many of its functions go unused, naturally resulting in lower efficiency. An ASIC (Application Specific Integrated Circuit) is a chip customized for a particular company and a particular AI model. It does only one thing, but does it exceptionally well. Its efficiency is 3-5 times that of a GPU, while its cost is only 20-30% of a GPU's.
The trade-off is a long customization cycle and high barriers to entry. Only hyperscalers such as Google, Amazon, and Microsoft can afford it. Marvell helps these hyperscalers design ASICs, making it their "exclusive arms dealer."

IV. What Marvell Has to Deliver Tonight
Let's clarify the numbers first.
After the U.S. market close today (early tomorrow morning in Beijing), Marvell will release its FY2027 Q2 earnings report. Wall Street's expectations are as follows:
In the previous quarter (Q1 FY27), actual revenue was $2.42B, up 27.6% year over year and already exceeding analysts' expectations of $2.41B.
The guidance for this quarter is $2.7B, representing continued acceleration.
For the full year: Management guided to FY27 full-year revenue of nearly $11.5 billion (+40% YoY), rising further to approximately $16.5 billion in FY28. For a company with a market capitalization of $7.5 billion, this growth rate represents an extremely steep growth curve.

V. Why This Earnings Report Is Not Ordinary
The numbers are the result. What truly deserves attention is what has happened over the past six weeks.
On August 19, Marvell completed something the market had underestimated. Google issued Marvell warrants to purchase 58.97 million shares at an exercise price of $206.58, with a total value of approximately $12.2 billion and an expiration date in 2033. This is not an ordinary supplier contract.
There is a hidden clause behind it: The warrants vest in stages, with vesting contingent on Google actually purchasing a certain volume of Marvell chips. The purchase volume corresponding to full vesting is reportedly close to $120 billion.
In other words, Google used its own equity upside to align its incentive to purchase Marvell chips. Every chip it buys increases the value of the options it holds. This is an unprecedented incentive-alignment mechanism.
But the more important implication is another one: This transaction completed Marvell's three-hyperscaler lineup. Amazon → Trainium (AI training/inference accelerator)
Microsoft → Maia 200 (inference chip, 30% cost advantage)
Google → TPU v5/v7 Ironwood (agreement finalized on August 19)
All three major hyperscaler customers are now locked in.
Broadcom's customers are Google and Meta, not Amazon and Microsoft.
This landscape was only established just before today's earnings release.

VI. What the Market Has Priced In, and What It Has Not
Marvell's stock has risen steadily from its August low of $220. After Nvidia's earnings report far exceeded expectations early this morning, it jumped directly to $256 in overnight trading, with its gain for the month already exceeding 16%.
The market's reaction is that it recognizes the direction but has not yet fully priced in the scale.
The reason is a mathematical issue: Marvell's current annual AI ASIC revenue is approximately $11 billion (actual FY2026). But this figure was achieved while its partnerships with Amazon and Microsoft were still in the early stages and its Google partnership had only just been confirmed.
Oppenheimer's view is that the Google agreement supports management's forecast that custom AI ASIC sales will double next year. The base is $11 billion, so doubling would bring it to $22 billion—while at that point, Marvell's total revenue would only be $16.5 billion (FY28 guidance).
This means ASICs will account for the majority of Marvell's total revenue. Marvell has effectively already become an AI inference chip company; the earnings figures simply have not fully reflected it yet. Tonight's earnings report is the first opportunity for this thesis to be reflected in the numbers.

VII. The Three Signals That Truly Matter in the Earnings Report
The numbers are certainly important, but whether they beat expectations is the result. The following three signals are the key to judging the direction:
① AI ASIC revenue share and growth rate Management typically provides segment data for AI ASICs during the earnings call. If AI ASIC revenue accelerates sequentially this quarter—not merely meets guidance—that would be the strongest positive signal.
② The strength of third-quarter guidance Will Marvell proactively raise its guidance, or conservatively provide a range of $2.8-3.0B? The strength of the guidance will reflect management's assessment of future order visibility better than the actual figures.
③ The first quantification of the Google partnership This is what the market is most anticipating. Will management provide a specific timeline for the Google partnership's revenue contribution during the call? Even a directional statement would lead the market to reprice the practical significance behind these $12.2 billion warrants.

VIII. A Framework for Investors
This is an asymmetric window of opportunity.
Over the past two years, Marvell has had "a compelling thesis without execution"—everyone said ASICs were the future, but earnings growth was not yet fast enough to support a high valuation.
This moment is different: The three-hyperscaler lineup is complete, the $12.2 billion warrants are locked in, and AI inference capital expenditure is still accelerating. The next step is for the "thesis to begin showing up in the earnings figures." For investors holding Marvell: Tonight's earnings report is the first real test of execution. There is only one core question: Has AI ASIC revenue growth begun to reflect the order structure following the completion of the three-hyperscaler lineup?
If so, this earnings report deserves to be repriced.
If not, then wait for Investor Day on October 6—that will be the first time management formally presents the long-term revenue path following the completion of the three-hyperscaler lineup.

Afterword
The five major hyperscalers will spend a combined more than $650 billion on AI infrastructure capital expenditures this year. Ultimately, 60-80% of this money will flow to the inference side.
On the inference side, custom ASICs are 3-5 times more efficient than general-purpose GPUs and cost 70-80% less. Only two companies control more than 95% of this market. Marvell is one of them. Today's earnings report is the first exam. $MRVL ‌