New Trends to Power Faster AI/ML Adoption?
In 2012, Google X lab created a neural network that can identify cats. Since then, technology companies have been increasingly adopting AI/ML on a large scale to build better applications and services for consumers (ToC). On the other hand, AI/ML's adoption on the enterprises' side (ToB) has yet to see the same growth trajectory due to the costs and complexities in both hardware and software. However, Since 2020, we started noticing three emerging tech trends that can help accelerate enterprises' adoption of AI/ML.
Breakthrough in semiconductors: In 2020, Nvidia debut the concept of "Data Processing Unit," a new class of programmable processors that combine high-performance CPU with SmartNiC (network interface controller). Data centers can deploy DPUs to optimize computing offload and frees up CPUs to focus on intended tasks, such as machine learning. DPUs help resolve a significant bottleneck for ML training, where models, sometimes with billions of parameters, are way too big for traditional CPUs and GPUs to handle. Other leading semiconductor players, such as Marvell and Xilinx, follow suit with their in-house or partner-designed DPUs. Industry analysts have forecasted that the market size for DPUs in data centers alone could reach $50 billion by 2025.
Close 
Breakthrough in software: we also saw significant progress of "Conversational AI," a new form of AI that can understand and speak languages with human-like accuracy, in 2020. Conversational AI allows two-way interactions and provides a much better user experience than traditional AI-powered Chatbot, mostly a one-way response system. The secret of conversational AI is its ability to handle lots of human conversation variance. Developers have designed innovative algorithms such as "Switch transformers" and "Sparse training" to enable models to handle vast amounts of data. The size of conversational AI training models is enormous and keeps expanding. For example, in February 2021, Google Brain announced a model with 1.6 trillion parameters, nine times the size of the famous Open AI GPT-3 (175 billion parameters) unveiled in July 2020. GPT-3 is 100+ times bigger than GPT-2 introduced in 2019.
Close
Cloud giants are expanding machine-learning platforms for developers. Andy Jassy famously said that "AI is shifting from a niche experiment inside technical departments to becoming more mainstream in business processes." in the 2020 AWS reInvent. During the conference, AWS rolled out many AI products across the technology stack, including AI chips (AWS Trainium), database (Aurora Machine Learning), and vertical solutions (Amazon Healthlake), etc. However, the most significant development is the drastic expansion of "Amazon SageMaker," one of the largest cloud machine-learning platforms. SageMaker has been offering new features to make it easier for developers to automate machine learning workflow. Microsoft Azure and Google Cloud are also growing their ML developer platforms.
Close
Witnessing these breakthroughs in semiconductor and software, coupled with cloud giants' effort to democratize AI, we see a coming inflection point of accelerated AI adoption in both ToC and ToB markets. So how do we benefit from this megatrend? In semiconductors, we like companies with DPUs exposure. In AI development and processing, we favor multi-cloud AI platforms such as Databricks. In enterprise software, we believe there will be a strong wave of new AI-based enterprise applications that can be creative and efficient in solving real-world problems.
Esoterica's statements are not an endorsement of any company or a recommendation to buy, sell or hold any security. For full disclosures, click here.
