To The Moon

We have seen how $NVIDIA(NVDA)$ shares have been affected and decline by the multiple factors which DeepSeek is indicating to challenge.

I have been doing machine learning for many years and Nvidia have been the GPU chips that was seen and used by many in the machine learning sector, if you are someone who are using both on-premise and cloud platform for machine learning build, you will understand that there is always a way to train models cheaper because it has to do with the engineering part of the machine learning process, that can reduce the time (which is money) used to train a model and also inference the result.

So the concerns of DeepSeek breakthroughs causing some investors to start wondering whether Nvidia customers would not be needing more GPU and will be buying custom ones like what some of the hyperscalers are using (will be sharing more in the article). That could favour stocks like Broadcom (AVGO) and Arm Holdings (ARM).

But I would like to share in this article on what are the major hyperscalers planning for the large scale AI (machine learning) platform. From there, we should have an idea on what is the market share that Nvidia currently holds.

I hold position in Nvidia and will not be selling any as I foresee a long term advantage from Nvidia.

Hyperscalers Reliance On Nvidia GPUs

Major hyperscalers (AWS, Microsoft Azure, Google Cloud, Oracle Cloud, IBM, etc.) rely heavily on NVIDIA GPUs for their cloud-based machine learning (ML) platforms, though some are also developing or deploying custom AI accelerators to reduce dependence on NVIDIA.

NVIDIA GPUs (Dominant Across All Platforms)

Here are the Nvidia GPUs which are dominating across the different hyperscalers.

NVIDIA A100 Tensor Core GPU:

• Used by $Amazon.com(AMZN)$ AWS (P4d instances), $Microsoft(MSFT)$ Azure (NDm A100 v4), $Alphabet(GOOGL)$ Google Cloud (A2 instances), and $Oracle(ORCL)$ Oracle Cloud (BM.GPU4.8).

• Optimized for large-scale AI training and inference (e.g., transformers, recommender systems).

NVIDIA H100 Tensor Core GPU:

• Next-gen GPU with Transformer Engine for LLMs.

• Deployed in AWS P5 instances, Azure ND H100 v5, Google Cloud A3 instances, and Oracle Cloud (OCI Compute H100).

• 30x faster than A100 for certain AI workloads.

NVIDIA L40S:

• General-purpose GPU for AI training, inference, and graphics.

• Used in Google Cloud (G2 instances) and AWS (G5 instances).

Multi-GPU Systems:

• HGX H100/A100 systems (8x GPUs per node) power massive clusters like Microsoft’s NDv5 and AWS P5/P4d.

Custom AI Accelerators (Hyperscalers-Specific Chips)

When we talked about hyperscaler, we also need to understand that they also have their own custom chips which is hyperscaler-specific. But we need to understand Nvidia advantage is in its chips can handle both training and inferencing, and in a world where inferencing work becomes more important.

The below custom chips are mostly used by the hyperscaler, so hyperscalers customers are still consuming the Nvidia chips.

Google Cloud:

• TPU (Tensor Processing Unit): Custom ASIC optimized for TensorFlow/PyTorch.

• TPU v4/v5e: Used for LLM training (e.g., Gemini) and available via Google Cloud’s TPU v4 Pods.

• Faster and more energy-efficient than GPUs for specific workloads.

AWS:

• Inferentia (Inf1/Inf2): Custom chips for cost-efficient inference (e.g., GPT-3, Stable Diffusion).

• Trainium (Trn1): Competes with A100/H100 for training large models (e.g., in AWS EC2 Trn1n instances).

Microsoft Azure:

• Maia 100/200: Custom AI accelerators (co-designed with OpenAI) to power Azure’s Copilot infrastructure.

• Also partners with AMD for Instinct MI300X GPUs in Azure ND MI300 v5 instances.

Meta:

• MTIA (Meta Training and Inference Accelerator): Custom chip for ranking/recommendation models.

Competition From Other Chips Makers

While we could be seeing the major hyperscalers having a higher adoption of Nvidia GPUs in their platform, we need to also understand what are some of the other chips makers presence in these hyperscalers.

Players like AMD, Intel and Arm also have their shares in the hyperscalers.

AMD GPUs (Emerging Alternative)

AMD Instinct MI300X:

• Used in Azure ND MI300 v5 and Oracle Cloud (OCI Compute MI300X).

• Competes with NVIDIA H100, offering 1.6x more memory (192GB HBM3) for large LLMs.

AMD Instinct MI250/MI210:

• Deployed in AWS EC2 Hpc6a instances for HPC and ML workloads.

Other Accelerators

Intel Gaudi 2/3:

• Used in AWS EC2 DL1 instances for training and inference (though less common than NVIDIA/AMD).

AWS Graviton:

• Arm-based CPU with ML optimizations, often paired with GPUs for cost savings.

Key Trends Of Major Hyperscalers Multiple ML GPUs/Accelerators

We have seen the different GPUs or accelerators that hyperscalers have, the reason why we are seeing existence of both Nvidia GPUs and custom chips are because of the following factors.

• Hybrid Deployments: Hyperscalers increasingly pair NVIDIA GPUs with custom chips (e.g., AWS uses H100 + Trainium).

• Diversification: To cut costs and reduce reliance on NVIDIA, hyperscalers are investing in custom silicon (e.g., TPU, Maia).

• Memory Matters: GPUs like H100 and MI300X are prioritized for LLMs due to massive HBM3/HBM3e memory (up to 192GB).

Why NVIDIA Could Still Dominates

• CUDA Ecosystem: NVIDIA’s software stack (CUDA, cuDNN, Triton) is deeply integrated into ML frameworks like PyTorch.

• Scale: Hyperscalers need 10,000+ GPUs for AI clusters, and NVIDIA is the only vendor with capacity and reliability.

However, competition from AMD, Intel, and custom chips will intensify in 2025 as hyperscalers seek cost and performance optimizations.

Technical Analysis - Key Moving Averages and RSI

If we have been following Nvidia trades over last few weeks, we are seeing how NVDA is trading below the 50-day MA period but try to stay above the 200-day MA period, this signal that NVDA is still have a pretty good momentum in the long term.

As of yesterday’s trading, we are seeing NVDA trending above both the 50-day and 200-day moving averages which is signalling strong momentum. And RSI levels is also not near 70 (which usually suggest a overbought short-term), so we could be seeing another upside movement moving forward.

Nvidia (NVDA) Sentiment Ratings

Based on the investors portfolios gathered from TipRanks, the sentiment is positive and we are also seeing a Strong Buy signal from the overall consensus which comprises of technical analysis and moving averages.

Can the Uptrend Continue?

Short-term (6–12 months): Likely yes, assuming:

• AI spending continues unabated (earnings beats).

• No major supply chain disruptions.

• Fed rate cuts boost tech valuations.

Long-term (2–5 years): Depends on:

• NVIDIA’s ability to maintain its AI leadership against rivals.

• Diversification into software/services to reduce hardware cyclicality.

• Navigating geopolitical risks and avoiding regulatory crackdowns.

Summary

As we can see that Nvidia GPU chips remains and important part in hyperscaler deployment of the platform for AI machine learning, and there is also the need for hybrid co-existence of hyperscalers in-house chips. So we can see that the demand for Nvidia chips should be ongoing and can continued in the long term.

NVIDIA’s uptrend is supported by secular AI tailwinds and unmatched innovation, but sustainability hinges on execution in a high-stakes environment. While near-term momentum favors bulls, we as investors should watch for competitors product launches and also macro factors (interest rates, recession risks).

Appreciate if you could share your thoughts in the comment section whether you think Nvidia is able to prove that their chips demand should calm investors fear of DeepSeek challenge.

@TigerStars @Daily_Discussion @Tiger_Earnings @TigerWire appreciate if you could feature this article so that fellow tiger would benefit from my investing and trading thoughts.

Disclaimer: The analysis and result presented does not recommend or suggest any investing in the said stock. This is purely for Analysis.