AI Perspectives on NVIDIA: From Financial Statements to Future Forecasts
“You’ve copied me on the company’s financial statements for 17 years. You just assumed I couldn’t read.”
– David (Greg Kinnear) in Sabrina (1995)
Exploring NVIDIA’s Financial Performance with AI Tools
Financial reports of companies serve as an invaluable resource for gaining insights about the respective organization and the broader industry landscape. In my book ‘Directing Business’, I talk about how important it is to use these helpful resources. You can also ask questions about it using my MoviSvami GPT tool here.
My curiosity was sparked when I learned about NVidia’s outstanding quarterly results through the nocode.ai newsletter curated by Armand Ruiz.
NVIDIA announced a record-breaking Q4 financial performance for the period ending January 28, 2024, with quarterly revenue soaring to $22.1 billion, a 22% increase from the previous quarter and a 265% surge from the previous year. The Data Center segment was a standout, contributing $18.4 billion, up 27% from Q3 and up 409% year-over-year. The fiscal year 2024 saw revenues hit $60.9 billion, marking a 126% growth. Earnings per diluted share also saw significant gains, with GAAP earnings at $4.93, up 765% from the previous year, and Non-GAAP earnings at $5.16, up 486% from last year.
I wanted to explore further and also use this chance to test various AI tools. Honestly, I thought I wouldn’t need anything beyond the superstar ChatGPT4, but I was surprised.
This unexpected response, along with a few more attempts, prompted me to explore other AI tools for comparison.
Perplexity.ai…
….no good.
Copilot in Edge.
I had to repeatedly close and reopen the browser session to receive a somewhat satisfactory response. However, even that wasn’t from the document.
The actual value we are looking for is the one in green.
I don’t often use claude.ai, but I thought I’d give it a go this time.
It provided the right answer.
I would have continued with my analyst role if claude.ai had humoured me with a few more queries. However, that was not the case.
So, I had to bring out the rookie techie in me.
As I regularly use ChatGPT4, I thought I would give it another chance.
First, via ChatPDF-GPT
Second, via OpenAI Assistant
In both cases above, we got perfect answers!
So, as Sam Altman has been suggesting (no, not about Sora), it seems like ChatGPT4 is not performing as well as expected or there’s something else going on.
The fact that we see many more options in the user feedback section (as shown below) proves this point.
So, the key takeaway from this exercise for me was the necessity of a third-party tool that uses the API of any popular LLM, such as ChatGPT4, Perplexity, Claude, etc., or a local chatbot using an open-source model.
Enter Embedchain and Gemma.
Embedchain simplifies the development of RAG (Retrieval-Augmented Generation) pipelines, handling complexities such as integrating and indexing data from diverse sources, determining optimal data chunking methods, and synchronizing with updated data sources.
Here’s a simple code snippet to work with LM Studio and gemma-it-2B q4_k_m gguf version.
from embedchain import App
from openai import OpenAI
# Initialize the embedchain application
ln_bot = App()
ln_bot.add("pdf_file", "NVDA-Company-Overview-2024-02-21.pdf")
#ln_bot.add("web_page", "https://investor.nvidia.com/news/press-release-details/2024/NVIDIA-Announces-Financial-Results-for-Fourth-Quarter-and-Fiscal-2024/")
# Fetch aggregated content
aggregated_content = ln_bot.query("What is the free cashflow for FY2024?")
# Set up local language model client
client = OpenAI(base_url="http://localhost:1234/v1")
#response = ln_bot.query(aggregated_content)
# Make a request to the local LM using the aggregated content
completion = client.chat.completions.create(
model="local-model", # this field is currently unused
messages=[
{"role": "system", "content": "Summarize and provide insights based on the following content"},
{"role": "user", "content": aggregated_content},
],
temperature=0.3,
)
# Print the response from the local language model
print(completion.choices[0].message.content)
“What is Operating Income and Margin as per GAAP for FY 2020?”
Sure, here's a summary and insights based on the content you provided:
**Summary:**
* Operating income for FY 2020 was $2,846 million.
* The margin of the operating income was 34.2%.
**Insights:**
* The operating income was significantly higher than the previous year ($1,904 million).
* The increase in operating income can be attributed to various factors, including increased sales, cost optimization, and effective management of operating expenses.
* The high margin indicates that the company was able to generate a significant amount of profit from its core operations.
* It would be beneficial to analyze the underlying drivers of the operating income increase, such as market conditions, customer behavior, and competitive landscape.
* The company's financial performance should be closely monitored to ensure continued stability and growth in the future.“What is Operating Income and Margin as per non-GAAP for FY 2020?”
Sure, here's a summary and insights based on the content you provided:
**Summary:**
* Operating income for FY 2020 was $3,735 million.
* Margin percentage was 34.2%.
**Insights:**
* The operating income was significantly higher than the GAAP (Generally Accepted Accounting Principles) amount of $1,884 million. This could be due to the inclusion of non-GAAP items or the use of different accounting principles.
* The margin percentage of 34.2% suggests that the company achieved a high level of profitability. It would be beneficial to understand the specific non-GAAP items included in the operating income calculation to assess the company's financial performance in more detail.
* The comparison with the GAAP amount provides a benchmark for evaluating the company's performance and the impact of non-GAAP items on the overall financial results.
**Additional Context:**
Without further context or information, it is difficult to provide additional insights or draw conclusions based on the given content.“Tell me about NVIDIA Gross Margins Reflect Value of Acceleration”
As you can see, the results are spot on if the questions are related to plain text. However, the accuracy decreases a bit when dealing with tables or images.
A free Streamlit chatbot is available on GitHub, which can be deployed locally (with an OpenAI api key). It further simplifies the process of uploading documents and chatting over them. The results are pretty much the same as code execution.
When you fork the above repository or manually copy the files to your local machine, you might encounter the following error.
`KeyError: 'st.session_state has no key "api_key". Did you forget to initialize it? More info: https://docs.streamlit.io/library/advanced-features/session-state#initialization'If you do encounter this error, simply declare a random value (`export api_key=1') before executing the Streamlit application. This will resolve the error.
In simpler terms, if you need a chatbot that’s very accurate, go for closed source models accessed via an API. But if you need to keep your data confidential and you can’t use tools like Open AI assistants (even though Open AI does promise not to use this data for training), then your best bet is to use local or private cloud options. These can be made using open-source models or smaller language models.
Regardless of the approach, it remains crucial to systematically experiment across the entire RAG pipeline.
References:
- NVIDIA Financial Presentation: https://s201.q4cdn.com/141608511/files/doc_presentations/2024/Feb/NVDA-Company-Overview-2024-02-21.pdf
- LM Studio: http://www.lmstudio.ai
- Embedchain Chat Pdf demo: https://ec-chat-pdf.streamlit.app/
