The rise of generative AI technologies, exemplified by platforms like ChatGPT, has ignited a technological frenzy, propelling NVIDIA to unprecedented market heightsThe demand for their pivotal product, GPU chips, has surged, providing concurrent opportunities for competitors like AMD.

Recent market analysis reveals that NVIDIA's market capitalization skyrocketed by nearly 1.9 trillion yuan (around 270 billion USD), boasting a leap of over 11%. Despite this impressive growth, industry experts are expressing concerns regarding NVIDIA's futureAccording to quarterly reports from tech giants like Google and Microsoft, AI-related revenues have not met expectations, raising fears about potential impacts on their significant capital investments

Additionally, competitors are on the rise, with AMD’s AI chip revenue reaching over 1 billion USD in the second quarter.

IExamining Financial Reports: Data Centers Shine

Analyzing the financial statements, AMD's latest report exceeded expectations, contrasting with NVIDIA's recent fiscal results, which have garnered substantial attention for their data center profitability.

01. AMD’s Second Quarter Report Surpasses Projections

On July 31, prominent chip manufacturer AMD unveiled its second-quarter earnings, which were better than previously anticipated (approximately 5.7 billion USD, with a variance of 300 million USD). The total revenue for the second quarter reached 5.835 billion USD, indicating a 9% increase year-over-year and a 7% rise from the first quarter, surpassing the expected 5.72 billion USD

• Poland Releases Butter Reserves to Combat Price Surge
• Hong Kong Privatizations Offer Investment Plays
• The Significance of Price
• U.S. Inflation Risks Rising Significantly Next Year
• The Fed's QT Approaches a Turning Point

The net profit stood at 265 million USD, reflecting an astonishing 881% year-over-year growth and a 115% jump from the first quarter.

Breaking down the specific business segments, AMD's data center division achieved record revenue of 2.8 billion USD, marking a 115% year-over-year and a 21% quarter-over-quarter increaseThe client segment generated 1.5 billion USD, up 49% year-over-year and 9% quarter-over-quarterIn contrast, the gaming division's revenue fell to 648 million USD, dropping 59% from the previous year and 30% from the last quarter, while the embedded segment saw revenue of 861 million USD, down 41% year-over-year but up 2% from the previous quarter.

AMD's Chairwoman and CEO, DrLisa Su, indicated that the AI sector is experiencing rapid growth, propelled by the demand for Instinct, EPYC, and Ryzen processors

The company is poised for robust revenue increments in the latter half of the year, leveraging the swift advancements in generative AI, which is amplifying the need for increased computation across various market sectors, thus opening vast growth opportunities.

Looking ahead to the third quarter, AMD anticipates revenues of between 6.4 billion and 7 billion USD, with a midpoint of 6.7 billion USD, projecting over 16% year-over-year and approximately 15% quarter-over-quarter growth, slightly above analysts' predictions of 6.62 billion USDAMD is also forecasting data center GPU sales to reach 4.5 billion USD in 2024.

02. NVIDIA's Total Revenue Surges Over 200% for Three Consecutive Quarters

NVIDIA's latest earnings report revealed that for the first fiscal quarter ending April 28, 2025, the company achieved revenues of 26.044 billion USD, translating to an 18% quarter-over-quarter increase and a staggering 262% year-over-year growth

The net profit reached 14.881 billion USD, showcasing an exceptional 628% increase compared to the previous yearThe gross margin stands at 78.9%, significantly exceeding forecastsThis marks the third consecutive quarter where NVIDIA saw revenue growth surpassing 200% year-over-year.

The data center computing revenue was reported at 19.4 billion USD, reflecting a 29% quarter-over-quarter and 478% year-over-year increase, primarily driven by the growth in shipments of NVIDIA's Hopper GPU computing platform used for training and inferring large language models, recommendation engines, and generative AI applicationsAdditionally, network revenue surged 242% year-over-year to 3.2 billion USD, bolstered by the robust performance of its InfiniBand end-to-end solutions.

NVIDIA stated that its data center business achieved record-breaking results due to the shipment of Hopper series GPUs, generating 22.6 billion USD in revenue for that quarter, with an impressive year-over-year growth of 427% and a 34% rise from the previous quarter.

NVIDIA projects a revenue of 28 billion USD for the second quarter, with a margin of 2% fluctuation, which exceeds the market expectation of 26.8 billion USD

However, it anticipates a decline in non-GAAP gross margins to 75.5% (with a fluctuation of approximately 50 basis points), with the annual gross margin narrowing to around 70%, aligning closely with market outlooksAdditionally, it expects its total operating expenses to rise by 40% to 45% for the year.

IIAMD's Determination to Catch Up with NVIDIA

NVIDIA, which holds a significant share of the market, is diligently pursued by AMD, as both companies have outlined their product release schedulesCurrently, NVIDIA's powerful AI chip, Blackwell, is set to hit the market by the end of 2024, while AMD's MI325X is expected to be released in the fourth quarter of 2024—a highly anticipated moment for the industry.

Industry analysts widely believe that AMD's MI300 development roadmap has become more competitive due to software optimization and the advanced configuration of high-bandwidth memory (HBM3e). Compared to NVIDIA's more closed technical approach, AMD's open strategy is expected to yield positive returns.

AMD has revealed that companies like Microsoft, Meta, Dell, HP Enterprise (HPE), and Lenovo have started adopting its MI300 chips

The next-generation MI325X is set to launch in Q4 of this year, with the updated MI350X expected in 2025, and subsequently, the MI400 slated for a 2026 release.

In terms of GPU memory upgrades, AMD's MI325X is transitioning from 192 GB of HBM3 to 288 GB of HBM3e, with the speed of production ramp-up potentially determining its competitive standing against NVIDIA's current Hopper GPU and the upcoming Blackwell GPUIndustry experts speculate that AMD MI325X could achieve production increases more readily than the MI300, given that there are no substantial architectural changes.

AMD has committed to launching new AI processors or chips annually to keep pace with NVIDIA's innovation, which previously accelerated its own product launch timeline from every two years to annually.

NVIDIA's Blackwell platform is scheduled for mass production in 2025, replacing the existing Hopper platform and becoming the primary offering for NVIDIA’s high-end GPUs, comprising nearly 83% of the company's overall premium products.

CEO Jensen Huang of NVIDIA has stated that the company is well-prepared for the next wave of growth, with the Blackwell chips set to ship in the second quarter, entering data centers in Q4 this year, promising substantial revenue

Furthermore, NVIDIA plans to introduce new products annually.

IIIA “Chip” Boost for HBM and CoWoS...

According to TrendForce, in terms of the distribution of AI chip suppliers for AI servers, NVIDIA maintains a near 90% market share for AI servers equipped with GPUs, while AMD holds roughly 8%. However, when considering all AI chips in AI servers, including GPUs, ASICs, and FPGAs, NVIDIA's overall market share stands at approximately 64% for this year.

TrendForce's survey indicates a strong demand for high-end AI servers through 2025, particularly with NVIDIA’s next-generation Blackwell series (including GB200, B100/B200, etc.) expected to replace the Hopper platform as the market mainstream, thereby driving demand for CoWoS and HBM.

As for NVIDIA's B100, its chip size will double compared to the H100, requiring significantly more CoWoS materials

It is estimated that TSMC, the main supplier, will reach a total production scale for CoWoS of 550k to 600k by the end of 2025, marking a growth rate approaching 80%.

Regarding HBM consumption, the mainstream H100, expected in 2024, will feature 80GB HBM3, while NVIDIA's Blackwell Ultra and AMD's MI350, among other key chips, will incorporate up to 288GB of HBM3e by 2025, reflecting a threefold increase in per-unit usageWith a robust demand for AI servers expected to persist, the overall supply of HBM is anticipated to double by 2025.

Additionally, the demand for GPU chips will drive the AI server marketAs the need for high-speed computing grows, more effective cooling solutions for AI servers are becoming a focal point