Quick Bytes

🔥 Hot New Capacity for 2025 – New data center capacity across U.S. regions with high-density, low-PUE options. Let’s talk capacity! 🚀

• Southeast US: 10 MW, 20,000 sq ft, 40 kW/rack, available now.

• Central US: 40 MW, 15 MW mid-2025, 25 MW early-2026, 40 kW/rack.

• Mid-West US: 4 MW, 150 kW/rack, PUE < 1.2, 9 MW more by early-2026.

🚀 DeepSeek’s AI Breakthrough – Activates only 37B of 600B+ parameters, cutting compute costs & boosting efficiency by 15-20%.

💰 Big Tech's AI Spend – Amazon, Microsoft, Google & Meta investing $310B in AI despite cost-saving innovations.

⚖️ DeepSeek vs. OpenAI & Gemini – Competitive pricing, but Gemini Flash 2.0 leads in cost per million tokens & speed.

⚡ Natural Gas Powers AI Data Centers: Gas turbines, engines, and fuel cells provide scalable solutions for AI infrastructure.

Does DeepSeek Impact the Future of AI Data Centers?

China’s DeepSeek has made innovations in the cost of AI and innovations like mixture of experts (MoE) and fine-grain expert segmentation which significantly improve efficiency in large language models. The DeepSeek model activates only about 37 billion parameters out of its total 600+ billion parameters during inference, compared to models like Llama that activate all parameter. This results in dramatically reduced compute costs for both training and inference.

Others have been using mixture of experts (MoE) but DeepSeek R1 aggressively scaled to the number of experts within the model.

Other Key Efficiency Improvements in DeepSeek's Architecture Include:

1. Enhanced attention mechanisms with sliding window patterns, optimized key-value caching and multi-head attention.
   

2. Advanced position encoding innovations, including rotary position embeddings and dynamic calibration.
   

3. A novel routing mechanism that replaces the traditional auxiliary loss with a dynamic bias approach, improving expert utilization and stability.

These innovations have led to a 15-20% improvement in computational efficiency compared to traditional transformer implementations.

Amazon, Microsoft, Google and Meta are still proceeding with large data center buildouts for several reasons:

The surge in AI compute for reasoning and AI agents requires more compute and the increased efficiency enables more value to be delivered. Jevons paradox (economics) occurs when advancement make a resource more efficient to use but the effect is overall demand increases causing total consumption to rise. This was seen with cheaper personal computers meant the demand for computers increased 100 times from tens of millions to billions of units. The top 4 companies plan to spend $310 billion on AI infrastructure and research.

Deepseek came out at prices per million token that was far cheaper than OpenAI but OpenAI and Google Gemini have competitive and even better pricing.

The AI inference price improvements have been consistent but the surprise from Deepseek is that this latest push was not by OpenAI or Meta.

Google Gemini Flash 2.0 is lower cost per million tokens and gives faster answers than Deepseek.

Those who are building AI data centers and training models know that AI will continue to get much better and cheaper. The expectation is the demand for really good AI will increase despite cost improvements. There is energy efficiency and design choices that Deepseek has highlighted. They optimized coding by directly accessing the hardware of Nvidia GPUs. There are many companies exploring FPGA hardware encoding of logic.

OpenAI o3-mini has competitive pricing. It higher on input but output is twice as expensive.

There is scaling of pre-training, post training and test time training. There is also key competition for hardware efficiency and efficiency and optimization of all aspects of the hardware and software stacks.

There will be specialized AI models and agent systems that keep only the necessary specialized knowledge needed for particular use cases. Energy and cost efficiency will be another area of competition beyond improving the base models.

There are increases in performance from test time compute. The OpenAI O3 model used 30,000 H100 GPU hours to answer the toughest math and reasoning problems. This type of AI inference will only be used when there is great value to push the boundaries of AI capability for a vastly superior and urgently needed answer. There needs to be some form of question pre-analysis or routing done to estimate how much effort is worthwhile.

Deepseek shows AI continues to improve rapidly and we are getting better results for less energy and less cost. It shows that AI will be profitable where answers and value will become lower and lower cost. Even virtually free with more and more capable local models. The World will change and more AI Data Centers will be needed to give the best answers or agent actions for the most valuable and challenging needs.

Natural Gas: A Scalable Power Solution for AI Data Centers

Natural gas can generate electricity through three primary methods: gas turbines, gas engines, and gas fuel cells. Each of these technologies serves different use cases, depending on the power requirements and scalability needs of a data center. Gas turbines are ideal for large-scale data centers, offering high efficiency when used in a combined cycle power plant that recaptures waste heat for additional power generation. Gas turbines are best suited for hyperscale AI facilities, where hundreds of megawatts of power are required to support thousands of GPUs.

Gas engines, on the other hand, function similarly to traditional internal combustion engines, using spark plugs to ignite a compressed air-fuel mixture. They provide a lower-cost, modular solution for smaller-scale or edge data centers. While less efficient than turbines, gas engines are easier to integrate and scale as localized power solutions.

The most advanced natural gas power source for data centers is gas fuel cells, which use an electrochemical reaction rather than combustion to generate electricity. By extracting hydrogen from natural gas, fuel cells offer higher efficiency, lower emissions, and quieter operation, making them ideal for urban environments and grid-independent AI deployments.

As AI workloads continue to grow, natural gas—especially when paired with carbon capture and hydrogen integration—will play a critical role in the future of data center power.

New Capacity Opportunities For 2025

10 MW: Southeast Region of United States

• 20,000 sq ft data hall (contiguous space)

• Maximum rack density: 40 kW/rack

40 MW: Central Region of the United States

• Approximately 15 MW mid 2025 and 25 MW in early 2026

• Maximum rack density: 40 kW/rack

4 MW: Mid-West Region of the United States

• Another 9 MW online by early 2026

• Supports over 150 kW/rack density

• PUE < 1.2

The above sites are great for large-scale deployments and we have plenty of inventory for those that need a few megawatts or hundreds of kilowatts of capacity. What kind of infrastructure are you looking for? Book time here with our team to learn about what’s available.

Your Feedback Matters

At Infra Insider, we’re here to give you an edge—exclusive insights into the evolving infrastructure world that others miss. While most newsletters skim the surface, we go deeper, spotlighting the trends shaping AI, data centers, and tech capacity. Think behind-the-scenes intel, key players driving change, and actionable updates straight from the field. It’s not just another industry briefing—it’s your front-row seat to the future.

Your feedback is crucial in helping us refine our content and maintain the newsletter's value for you and your fellow readers. We welcome your suggestions on how we can improve our offering; [email protected]. 

Nina Tusova
Director of Customer Relationship & Success // Team Ignite