Are we going ‘soft’ on Hardware?
In the startup space, for reasons of scalability, discussions about ‘tech’ revolve around software. But software or AI algorithms run on a device made up of hardware components, most critical of them being the processing unit, commonly referred to as the chipset.
Beyond the tech world, chips or semiconductors (building material of chipset) have become a discussion point in the geopolitics arena, as nation states strategize to secure the supply-lines of rare-earth elements, and develop design & manufacturing competence in them.
Simplistically, the AI hardware can be classified in two ways, based on type of AI activity – training or inference.
Training refers to the AI component where an algorithm mines through a massive amount of data for long durations in order to ‘learn’ what it is set to learn. Thus, high memory, high computation, longer cycle life would be the required hardware characteristics. Often there is a disconnect between the training device and the eventual deployment device.
Inference refers to the real-time data crunching component of the AI algorithm to generate output. Hardware for this type of activity would demand speed & efficiency over capacity. Inference or output is required at every end deployment device which necessitates the hardware to be of low power consumption.
Hardware can also be classified based on the end application specificity. Industry jargons are – FPGA, GPU, ASIC.
Field programmable gate array (FPGA) can be simply understood as those chips which are flexible, configurable, power efficient, required to perform parallel tasks. Think of chips in our smartphones or laptops. Graphic processing unit (GPU) refers to the chips initially designed for graphics but are now also used for computationally heavy tasks such as video processing, image analysis etc. Application specific integrated circuit (ASIC) refers to a custom designed chip optimized for end application which cannot be used elsewhere.
Another industry jargon worth mentioning is the ‘leading’ edge / ‘trailing’ edge semiconductor manufacturing nodes. In simple terms, a chip is a circuit made up of micro transistors. Over generations, transistors have become smaller and more efficient. Leading edge denotes the smallest manufacturing possible(10-12nm), thus packing more transistors in a unit space for more computing power. But not all chips are meant for advanced computations or the latest laptops. There are less sophisticated devices and applications which rely on ‘trailing’ edge nodes. (older set of chips)
Given that the applications of AI are unfolding and expanding their reach in every sector, there exists a latent opportunity of hardware led innovation. By 2025, AI-related semiconductors could account for almost 20 percent of all demand, translating into about $67 billion in revenue. The Traditional Non AI semiconductor market is estimated to be ~$300 billion.
GPU market (NVIDIA, AMD) and FPGA market (Intel) are dominated by US firms. The US also controls the majority of semiconductor fabrication units (known as fabs) worldwide. China is catching up with the US in manufacturing of the ‘trailing edge’ chips.
India has certain strengths in the form of large human capital which can dominate design capabilities and recent policy initiatives are a step to harness them. The weaknesses of India are the lack of supply-side manufacturing infrastructure, and linkages with larger fab units.
Large-scale low-cost trailing edge chip design & production, geared towards AI hardware might be an interesting space to look out for in the Indian context, provided the market is shifting from general-purpose to application-specific chipsets.
We at Ecosystem Ventures aim to be at the ‘leading’ edge of upcoming startup ‘nodes’. Your insights and valuable comments are no less than the ‘training’ hardware for our ‘inferences’. Please feel free to share your thoughts.
Startup Funding Summary
Udaan, Bangalore-based e-commerce platform, has raised $120 Mn in convertible notes and debt from Existing Investors – Read More
boAt, Delhi-based consumer electronics company, has raised $60 Mn in a funding from an affiliate of Warburg Pincus and Malabar Investment – Read More
Kinara Capital, Bangalore-based fintech platform, has raised $25 Mn in funding from British International Investment, Nuveen and ASN Microkredietfonds – Read More
Devtron, Gurugram-based open-source DevOps platform, has raised $12 Mn in funding from Insight Partners, Leo Capital and Angel Investors – Read More
EMotorad, Pune-based EV brand, has raised $3 Mn in Pre-Series A funding from Green Frontier Capital, LetsVenture and Ivy Growth Associates – Read More
M&A Snippets
Ghaziabad-based FMCG company Dabur will acquire Mumbai-based foodtech company Badshah Masala for $73 Mn – Read More
Leave a Comment