Data centers are increasingly challenged by the rapid growth of data, rising demand for computational power, and the need for sustainable practices. Among the most pressing issues is energy consumption; data centers are notorious for their high electricity usage, driving up operational costs and environmental impact. Additionally, the exponential growth in data necessitates scalable storage and processing solutions, requiring careful resource allocation to avoid over-provisioning or underutilization. The computing market is growing with a focus on reducing carbon footprint through renewable energy integration and enhanced energy efficiency. These challenges are compounded by rising operational costs, including those associated with advanced technology, energy, cooling, and maintenance.
Artificial Intelligence's (AI) rapid growth demands immense computational power, leading to higher energy consumption in data centers. As AI models become more complex, they require increased processing and storage capabilities, straining existing infrastructure. The need for scalable resources to support AI workloads can lead to over-provisioning, underutilization, or increased carbon footprints, making it harder for data centers to balance efficiency, cost, and environmental impact.
The uniqueness of Intensivate's solution lies in their Intensivate Processing Unit (IPU), which can process 20W of energy usage, equivalent to a standard 1U server. Their IPU's microarchitecture is tailored for big data workloads, employing multiple hardware threads to tolerate long latencies effectively. Unlike traditional Out-of-Order (OoO) CPUs, which rely on complex machinery to optimize performance, Intensivate's design uses 16 hardware threads per core, allowing it to handle parallel workloads more efficiently. This approach results in lower power consumption and no modification of software, hardware, or infrastructure, making it a highly efficient solution for data-intensive applications.
Intensivate has significantly de-risked the scale-up of their technology through extensive testing using FireSim, an open-source emulation system from the creators of RISC-V at UC Berkeley. FireSim uniquely synchronizes all FPGA-emulated CPU clocks, ensuring accurate emulation of network and disk I/O during the correct CPU cycles. By programming their final implementation into FireSim, Intensivate can replicate the exact instructions per cycle and simulate real-world scenarios, including network latency and memory usage.
Intensivate’s design has demonstrated significant advantages in performance per watt (PPW), being approximately seven times more efficient than Intel cores and six times more efficient than AMD cores on the same process node when running non-target applications. The company’s approach, which eliminates the need for complex OoO machinery, offers a compelling alternative in the competitive landscape of data center processors, particularly for big data workloads.
As Intensivate moves forward, their focus on industry-standard manufacturing processes, extensive testing, and strong partnerships positions them well for their planned commercialization in 2027. Their innovative approach to processing unit design could redefine efficiency and performance in data centers, offering a unique solution in a rapidly evolving industry.
Intensivate has just successfully received their first silicon chip, marking a significant milestone as they prepare for testing. The chip’s manufacturing process poses minimal risk, as it is being produced by one of the world’s most established chip manufacturers. Their rigorous testing environment has allowed Intensivate to run real applications’ scenarios, ensuring that their architecture would perform as predicted before the actual silicon was manufactured.
Intensivate's technology enables a significant reduction in this environmental impact by replacing 1 in every 11 racks, leading to a 90.91% energy savings. That is also a 12x physical space increase in order to process future large data.
In 2022, global data centers consumed 460 TWh of energy, with forecasts predicting a rise to 1,000 TWh by 2026 and 2,967 TWh by 2030. This growth translates to substantial carbon emissions, as each kilowatt-hour generates 0.86 pounds of CO2.
With an average data center spanning 100,000 square meters and housing 100,000 servers, energy use per square meter averages 1,000 kWh. Intensivate could reduce emissions by approximately 40 million metric tons annually based on today’s Big Data Processing. Based on 2030 data processing assumptions, Intensivate could reduce GHG emissions by 315 million metric tons per year.
Intensivate is targeting a significant market opportunity by initially focusing on a $15 billion segment involving traditional machine learning, Kafka, and Hadoop/Spark, with plans to expand into a $33 billion global market for Big Data servers.
Their business model revolves around selling a compact server module called an IPU, integrating their technology with low-power DDR5 memory into a standard EDSFF form factor. This design allows for seamless adoption by server OEMs, cloud providers, and enterprises, eliminating the need for new software development.
Intensivate’s strategic approach includes collaborations with major OEMs like Dell, HPE, and Supermicro, and partnerships with chipmakers and service providers, enhancing their market presence and scalability. Additionally, Intensivate is engaging with hyperscalers such as Azure and eBay, and industry giants like IBM and SK Hynix, to validate and expand their technology’s impact.
The Intensivate team, led by newly appointed CEO Zach Deiri, combines extensive industry experience and domain expertise. Zach, with a background in engineering and notable leadership roles at companies like Motorola, STMicroelectronics, and SanDisk, brings a proven track record in scaling and commercializing technology.
Sean Halle, the CTO and inventor of the technology, previously served as CEO and has extensive experience in parallel computation and chip design. Supporting them are Jim Aralis, with 40 years in electronic circuits; Scott McCormack, a semiconductor chip design veteran; and Serge Sirisena, with expertise in server sales and business development.
