Applied Digital Boosts Stock with Cutting-Edge AI Data Centers

Key Points

• Applied Digital develops and operates next-gen AI data centers, which require exponentially more computing and electrical power than traditional data centers.
• As large language models (LLMs) increase, the number of parameters increases, as evidenced by GPT-3, which had 175 million in 2020, and GPT-4, which had 1.76 trillion in 2023, and it continues to grow.
• Next-gen AI data centers have high-density support of up to 120kW versus traditional data centers with low-power density designs ranging from 12kW to 15kW.

Applied Digital Co. NASDAQ: APLD designs, develops and operates next-generation data centers accommodating artificial intelligence (AI) service providers in the high-performance computing (HPC) industry. The AI boom has been driving the demand for more data centers, which is driving demand for next-gen data centers that fall right in Applied Digital's wheelhouse. The company also provides blockchain colocation services for cryptocurrency miners and GPU cloud services through its Applied Digital Cloud segment. They are an NVIDIA Co. NASDAQ: NVDA Elite Partner.

Applied Digital operates in the computer and technology sector, competing with data center companies, including data center REIT Digital Realty Trust NYSE: DLR and giant Equinix NASDAQ: EQIX.

The company has two blockchain data centers. Its JMS01 data center is located in Jamestown, North Dakota, and has a capacity of 100 MW. Its ELN01 data center is in Ellendale, North Dakota, and has a capacity of 180 MW. Its 2 HPC customer-built AI data centers are JMS02 in Jamestown and ELN02 in Ellendale, North Dakota, and have capacities of 9 MW and 100 MW, respectively.

Applied Digital's Ellendale Data Centers: Power Outages and Growth Plans

In fiscal Q3 2024, Applied Digital suffered several power outages in its Ellendale data centers. In Q2 2024, Applied Digital broke ground on its first 100 MW HPC facility in Ellendale. The 342,000-square-foot building is designed to provide highly efficient and ultra-low-cost HPC applications with a liquid-cooled infrastructure.

The company secured an exclusive letter of intent (LOI) with a US-based hyperscaler for a 400 MW capacity lease of its Ellendale HPC campus as an anchor tenant. This includes the 100 MW facility, which is still under construction, and plans for two additional buildings. The company continues to witness outsized demand for HPC hosting at its differentiated Ellendale campus. Ellendale has 600 MW of future capacity.

Applied Digital's Data Center Hosting Updates

The 100 MW Jamestown facility operates at full capacity. The 180 MW Ellendale facility experienced a power outage that started in January 2024. Its utility provider installed equipment that enables them to power down affected portions of the site. They discovered the power failures were due to transformers that didn't meet industry standards. The company procured the correct transformers and components to start re-energizing to around 14% of its capacity. Its blockchain data centers total 280 MW capacity hosting infrastructure (power and maintenance) for Bitcoin miners and blockchain infrastructure companies. Its largest Bitcoin miner is Marathon Digital Holdings Inc. NASDAQ: MARA.

Applied Digital Cloud Services Update

Applied Cloud Services enable customers to execute and scale critical HPC workloads cost-effectively, including AI, rendering, and machine learning (ML). It enables AI/ML companies to access cloud servers to run and train AI applications. While the initial customers were startup AI companies with significant funding, Applied Digital is now experiencing demand from medium to large-size enterprises. This is essentially GPU-as-a-service (GaaS), which enables AI companies to perform AI applications without having to invest in buying the equipment.

Understanding Parameters: Fueling the AI Boom and LLM Complexity

The evolution and upgrade of ChatGPT models bring a significant rise in the number of parameters and higher energy consumption. Parameters are pieces of information learned from training data for large language models (LLMs). As they encounter sequences of words, they change the value of the parameters to create connections and ultimately predict the next word or phrase in the sequence. The more parameters are used, the better its performance can be.

Parameters are the numerical foundation of LLMs representing learned relationships between training data. The more parameters that are used, the more power in terms of computational and energy is needed. Chat-GPT 3, released in 2020, had 175 billion parameters. GPT-4 in 2023 had 8X 220 billion or 1,760,000,000,000 or 1.76 trillion parameters, which require exponentially more computing power and energy consumption. The number of parameters continues to climb, requiring even more power. Data center electricity consumption is expected to grow 125% by 2030, using 9% of all electricity in the US.

Comparing Traditional Existing Data Centers to Next-Get AI Data Centers

AI hardware requires much more power than traditional devices. A conventional CPU in a traditional data center uses 300 watts per hour, while an NVIDIA H100 GPU uses 700 watts per hour. A typical Google request/query takes 0.3 watt-hours, compared to a Chat-GPT request, which is 10x at 2.9 watt-hours.

With the elevated power needed to operate AI applications, the traditional data center is lacking. Traditional data centers have low IT MW load and low power density design (12 kW to 15 kW). They are located in major cities and are optimized for high-speed and ultra-low latency. They are used for the internet backbone, centralized data, and instantaneous streaming apps that are evolving for Web 2.0.

Next-gen AI data centers are purpose-built to support significantly higher energy consumption. They are better suited to remote locations with high-density support up to 120kW and are latency insensitive due to training. They are used for HPC functions like AI, ML, language processing, drug discovery, and graphics rendering. Applied Digital has over 6,000 H100 GPUs deployed as clusters across 3 data center footprints.

APLD Stock is Forming a Potential Cup and Handle Pattern

The daily candlestick chart for APLD depicts a cup and handle pattern. The cup line has been tested at $6.34 and is awaiting a retracement to form the handle. The daily relative strength index peaked at the 70-band and is pulling back to the 61-band. Pullback support levels are at $5.57, $4.82, $4.35, and $3.84.

Applied Digital analyst ratings and price targets are at MarketBeat. The consensus price target suggests a 41.18% upside.