NVIDIA-公司周读04

NVIDIA Corporation provides graphics and compute and networking solutions in the United States, Taiwan, China, Hong Kong, and internationally. NVIDIA Corporation was incorporated in 1993 and is headquartered in Santa Clara, California.  NVIDIA has around 30,000 employees globally.

The Graphics segment offers GeForce GPUs for gaming and PCs, the GeForce NOW game streaming service and related infrastructure, and solutions for gaming platforms; Quadro/NVIDIA RTX GPUs for enterprise workstation graphics; virtual GPU or vGPU software for cloud-based visual and virtual computing; automotive platforms for infotainment systems; and Omniverse software for building and operating metaverse and 3D internet applications. The Compute & Networking segment comprises Data Center computing platforms and end-to-end networking platforms, including Quantum for InfiniBand and Spectrum for Ethernet; NVIDIA DRIVE automated-driving platform and automotive development agreements; Jetson robotics and other embedded platforms; NVIDIA AI Enterprise and other software; and DGX Cloud software and services. The company's products are used indata center,  gaming, professional visualization, and automotive markets. It sells its products to original equipment manufacturers, original device manufacturers, system integrators and distributors, independent software vendors, cloud service providers, consumer internet companies, add-in board manufacturers, distributors, automotive manufacturers and tier-1 automotive suppliers, and other ecosystem participants.

NVIDIA is modernizing the world’s trillion-dollar data center industry. By accelerating workloads with NVIDIA GPUs operating in parallel, we exponentially increase throughput while driving down the total energy used to complete a task and, ultimately, the total cost of ownership. The energy savings are incredible. That is precisely why, as CPU scaling has slowed, we must transition toaccelerated computing.We cannot continue scaling out the traditional way. Accelerated computing is essential, and every application that can be accelerated should be.

• Nvidia was founded on April 5, 1993,[28][29][30] by Jensen Huang (current CEO), Chris Malachowsky (current NVIDIA fellow), and Curtis Priem (retired former CTO 1993-2003).

• In 1996, Huang laid off more than half of Nvidia's employees—then around 100—and focused the company's remaining resources on developing a graphics accelerator product optimized for processing triangle primitives: the RIVA 128. By the time the RIVA 128 was released in August 1997, Nvidia was down to about 40 employees  and only had enough money left for about one month of payroll.

• Nvidia went public on January 22, 1999.

• From  late 1999, Nvidia started launching the GeForce series of consumer oriented GPUs for the PC gaming market (with variants for data center GPUs and working stations ). It has been upgrading every ~2 years for the GPU microarchitecture: GeForce 256 (microarchitecture: Celsius, 1999) , GeForce 2 series (Celsius, 2000), 3 series (Kelvin, 2001), 4 series (Kelvin, 2002), 5 series (Rankine, 2003), 6 series (Curie, 2004), 7 series (Curie, 2005), 8 series (Tesla, 2006), 9 series (Tesla, 2008), 200/300 series (Tesla, 2008), 400 series (Fermi 2010), 500 series, 600 series, 800M series, 900 series,  10 series (2016), 20 series (2018), 30 series (2020,Ampere,A100),  GeForce 40 series (2022, Ada Lovelace),H100 (Hopper, 2022), B100 (Blackwell, 2024), Rubin (2026)

• In 2001, NVDA entered the S & P 500 stock index.

• In June 23, 2007, NVIDIA releasedCUDA, a proprietary parallel computing platform and API that allows software to use certain types of GPUs for accelerated general-purpose processing.

• In July 2008, Nvidia took a write-down of approximately $200 million on its first-quarter revenue, after reporting that certain mobile chipsets and GPUs produced by the company had "abnormal failure rates" due to manufacturing defects.

• In November 2011, after initially unveiling it at Mobile World Congress, Nvidia released its ARM-based system on a chip for mobile devices,Tegra3. This product series is Mobile CPU for gaming consoles.

• In May 2017, Nvidia announced a partnership with Toyota which will use Nvidia'sDrivePX-series artificial intelligence platform for its autonomous vehicles.

• In 2018, Google announced that Nvidia's Tesla P4 graphic cards would be integrated into Google Cloud service's artificial intelligence

• On March 11, 2019, Nvidia announced a deal to buyMellanoxTechnologies for $6.9 billion to substantially expand its footprint in the high-performance computing market.

• On May 14, 2020, Nvidia officially announced their Ampere GPU microarchitecture and the Nvidia A100 GPU accelerator

NVDA stock start taking off by AI (2016, 20x from 2016 t0 2020) and by Generative AI (2023, 10x from 2022 to 2024), stock increased around200xin the past 10 years.

管理层/CEO
Jensen Huang

‘Founder mode’ 管理模式（vs.职业经理人模式）
Just CEO, CFO, COO and Consel. No CTO, CPO, CSO etc
CEO have 60+ direct reports, no 1:1s
Back in 2016, Huang personally delivered (donated) the first DGX-1 AI supercomputer to OpenAI, in support of democratizing AI technology
Nvidia CEO Jensen Huang personally delivered the first Nvidia DGX H200 to OpenAI's office in San Francisco on April 24, 2024
Huang has been directly talk to top AI scientists when there are breakthroughs in AI (2012 AlexNet, 2017 Transformers, etc)

董事会

<!--{"config":{}}-->

⚙️Data Center was 88% of overall revenue(+1pp Q/Q). It grew 154% year-over-year and 16% sequentially. As part of Data Center:

• ⚡ Compute: Up 162% year-over-year and 17% sequentially to $22.6 billion. Like previous quarters, the driver was strong demand for theHopper GPUcomputing platform used for training and inferencing of large language models (LLMs), recommendation engines, and Gen AI apps.

• 🔌 Networking: Up 114% year-over-year and 16% sequentially to $3.7 billion. The Spectrum-X Ethernet networking solution optimized for AI doubled sequentially, andInfinitibandremained a crucial driver.

🎮Gaminggrew 9% sequentially, partially boosted by back-to-school season. GeForce RTX PCs have an installed base of over 100 million.

👁️Professional Visualizationgrew 6% sequentially, driven by AI and graphic use cases, including model fine-tuning and Omniverse-related workloads.

🚘Automotive grew +5% sequentially, driven by the ramp of AI cockpit solutions and self-driving platforms.

Nvidia's product families include graphics processing units, wireless communication devices, and automotive hardware and software, such as:

• GeForce, consumer-oriented graphics processing products

• RTX, professional visual computing graphics processing products (replacing GTX and Quadro)

• NVS, a multi-display business graphics processor

• Tegra, a system on a chip series for mobile devices

• Tesla, line of dedicated general-purpose GPUs for high-end image generation applications in professional and scientific fields

• nForce, a motherboard chipset created by Nvidia for Intel (Celeron, Pentium and Core 2) and AMD (Athlon and Duron) microprocessors

• GRID, a set of hardware and services by Nvidia for graphics virtualization

• Shield, a range of gaming hardware including the Shield Portable, Shield Tablet and, most recently, the Shield Android TV

• Drive, a range of hardware and software products for designers and manufacturers of autonomous vehicles. The Drive PX-series is a high-performance computer platform aimed at autonomous driving through deep learning, while Driveworks is an operating system for driverless cars.

• BlueField, a range of data processing units, initially inherited from their acquisition of Mellanox Technologies

• Datacenter/server class CPU, codenamed Grace, released in 2023

• DGX, an enterprise platform designed for deep learning applications

• Maxine, a platform providing developers a suite of AI-based conferencing software

• 芯片厂：AMD(bought ATI), Intel, Broadcom, Qualcomm, Apple GPU, Companies from China (Huawei)

• 云厂/客户：Google TPU, Meta, Amazon, Microsoft

• 初创企业(inference chip): Groq, SambaNova, etc

• 开源社区(CUDA): ROCm, Triton, VLLM, JAX, SYCL, etc

<!--{"icon":"https://mdn.alipayobjects.com/huamei_ptjqan/afts/img/A*DPzkQ5dfwDQAAAAAAAAAAAAADkN6AQ/original","abstract":"这是文章的摘要，我还要想想怎么做","create_datetime":"Fri, 20 Sep 2024 23:13:58 GMT","folder_id":150,"id":461,"name":"web_clip_20240920_231356.pdf","real_path":"/mnt/datadisk/spaces/43f5ae52-183c-448f-922c-5ce49fd5a020/Saved Web Pages/web_clip_20240920_231356.pdf","space_id":null,"tags":null,"user":{"create_datetime":"2024-05-23 20:54:38","email":"[email protected]","id":8,"is_admin":true,"last_login":"2024-08-20 00:20:38","name":"Rui Ma","oid":"43f5ae52-183c-448f-922c-5ce49fd5a020","space_id":8},"collaborators":[{"Rui Ma":1}],"updateTime":"2024-09-20 16:13:58","type":"card","url":"https://www.know2.co/file/461","description":"这是文章的摘要，我还要想想怎么做"}-->

web_clip_20240920_231356.pdf

<!--{"icon":"https://mdn.alipayobjects.com/huamei_ptjqan/afts/img/A*DPzkQ5dfwDQAAAAAAAAAAAAADkN6AQ/original","abstract":"这是文章的摘要，我还要想想怎么做","create_datetime":"Fri, 20 Sep 2024 23:04:47 GMT","folder_id":158,"id":460,"name":"解读神秘的华为昇腾910.pdf","real_path":"/mnt/datadisk/spaces/43f5ae52-183c-448f-922c-5ce49fd5a020/NVDA/解读神秘的华为昇腾910.pdf","space_id":null,"tags":null,"user":{"create_datetime":"2024-05-23 20:54:38","email":"[email protected]","id":8,"is_admin":true,"last_login":"2024-08-20 00:20:38","name":"Rui Ma","oid":"43f5ae52-183c-448f-922c-5ce49fd5a020","space_id":8},"collaborators":[{"Rui Ma":1}],"updateTime":"2024-09-20 16:04:47","type":"card","url":"https://www.know2.co/file/460","description":"这是文章的摘要，我还要想想怎么做"}-->

解读神秘的华为昇腾910.pdf

Local Interference by consumer level GPUs:

竞争优势

产品护城河:

1. CUDA software (18 years)

2. Networking (Mellanox)

3. GPU architecture experienced (Hopper: 2022, Blackwell: 2024, Rubin: 2026)

4. Partnership with Microsoft, Meta, Oracle, OpenAI etc

最好的赛道：

1. AI

2. Automotive, Professional visualization

3. Gaming

优秀的管理层：

1. Founder lead

2. Founder with lots of product vision

3. Founder with great sales skills

4. The more you buy, the more you save

5. And demand is so great that delivery of our components and our technology and our infrastructure and software is really emotional for people because it directly affects their revenues, it directly affects their competitiveness.

6. For every dollar that they spend with us translates to $5 worth of rentals, And that's happening all over the world, and everything is all sold out.

7. Lean workforce

8. 30k employees, much smaller than other Mag 7 companies.

9. Competitive culture: top 30% performers got 70% of stock refreshes

   强大的商业模式：

   1. Gross margin mostly >60% in the past 10 years (~75% in 2024)

   2. 产品，解决方案，软件与服务 一体

竞争劣势

1. Weak on CPU (failed buying ARM)

2. Weak on Memory (e.g.,AMD MI300 has more memory than the Nvidia H100)

3. Core is still hardware, less on software, cloud or subscription revenues

4. Easy to fall into cyclic, be commoditized (e.g., Intel CPU is commoditized)

5. Core is still server side, not much on edge side (mobile, AI pc, etc)

机会

• Generative AI, LLM is still booming

• Big tech is still spending more on GPUs, less on hiring employees

• Government is spending a lot on AI (middle east countries)

• Help revolutionize traditional industries (biotech, energy, industrial etc)

• Develop their own cloud

威胁

• AI slow down, LLM performance plateau (scaling law is slowing)

• Competition, declining margin

• Semiconductors depends on the 半导体产业链，跟竞争对手上下游公司/技术都类似，半导体垂直领域分工明确, depends on TSMC onsupply

• EDA 软件: SNPS, CDNS

• Fab: TSMC, Samsung

• Semi Equipment: ASML, AMAT, LRCX, etc

• AI progress from training to inference

• Anti-trust, national security (policy on China)

• Jensen Huang/Founder dependent

• New computing / algorithm paradigm

Q2 FY25 (2024 Q2)

Income statement:

Revenue jumped +15% Q/Q to 1.3 billion beat).

⚙️ Data Center grew +16% Q/Q to $26.3 billion.

🎮 Gaming grew +9% Q/Q to $2.9 billion.

👁️ Professional Visualization grew +6% Q/Q to $0.5 billion.

🚘 Automotive grew +5% Q/Q to $0.3 billion.

🏭 OEM & Other grew +13% Q/Q to $0.1 billion.

Gross margin was 75% (-3pp Q/Q), in line with guidance.

Operating margin was 62% (-3pp Q/Q), 1 point ahead of guidance.

Non-GAAP operating margin was 66% (-3pp Q/Q).

Non-GAAP EPS 0.04 beat).

Cash flow:

Operating cash flow was $14.5 billion (48% margin).

Free cash flow was $13.5 billion (45% margin).

Balance sheet:

Cash and cash equivalents: $34.8 billion.

Debt: $8.5 billion.

Q3 FY25 Guidance:

Revenue +8% Q/Q to 0.8 billion beat).

Gross margin 74.7% (-0.5pp Q/Q).

As of 9/20, stock price of ~116, market cap is ~2.85 trillion, P/E ratio around 54, forward P/E is ~ 41

Earnings estimate (from seeking alpha):

Revenue estimate:

<!--{"config":{}}-->

Income statement:

Balance sheet:

Cashflow:

Nvidia has been involved in many acquisitions over the years, including:

• 3Dfx Interactive. In 2000, Nvidia acquired the graphics card maker 3Dfx Interactive.

• MediaQ. In 2003, Nvidia acquired MediaQ for about $70 million to gain expertise in wireless multimedia.

• iReady. In 2004, Nvidia acquired iReady, a provider of high-performance TCP offload engines and iSCSI controllers.

• ULI Electronics. In 2005, Nvidia acquired ULI Electronics, which provided third-party southbridge parts for chipsets to ATI.

• Hybrid Graphics. In 2006, Nvidia acquired Hybrid Graphics.

• Mellanox.In 2019, Nvidia purchased Mellanox for $6.9 billion, a supplier of networking technology.

• Arm. In 2020, Nvidia attempted to acquire chip designer Arm for $40 billion, but regulators blocked the deal. Nvidia later purchased a minority stake in Arm after it went public.

• Run:AI and Deci. In 2024, Nvidia recently acquired two small Israeli startups, Run:AI (700M) and Deci (300M), to help lower the cost of developing AI models.

Semiconductor:

• 计算芯片: need iterate every year/2 years, highly competitive

• 半导体周期属性Cycles:

• Hardware ( different from software: no continuous subscription rev after a sale, marginal cost is not zero, supply demand constraint)

AI 发展3要素: 算力, 算法, 数据，互相促进，互相制约

• 算法路线：多模态，scaling law (10万卡+），强化学习

• 算力路线：存算一体，短距离通信，内存

• 数据：sythetic data, 标注, 图像/视频

• 中长期 AI 怎么看？

• 中期：可能会有bubble破裂，近期大模型相对停滞，没有杀手级应用，大模型公司在等待blackwell硬件更新

• 硬件-基础设施-应用

• 长期：逐渐代替文科工作（文案，翻译，客服），理工科 （coding，画图）, 蓝领（自动驾驶, 人形机器人humanoid）

• 中长期  NVDA 怎么看？

• 中期：需求>Supply，training 下一代大模型需要，竞争对手需要很长时间赶上

• 长期: AI瓶颈，竞争，training 到 inference 的转移