Yevgeny Gelfand
It has been over six months since ChatGPT’s release took the world by storm. This seminal technology turned more than a few heads, and for good reason: the AI chatbot marked a material improvement over previous iterations and unlocked applications–for commercial use and personal amusement alike–previously relegated to the world of make-believe.
But what seems like “magic” to the uninitiated is matter-of-factly explained by insiders as the natural, even inevitable, evolution of the eponymous law Gordon Moore pioneered nearly six decades ago (suspending for the moment the ongoing debate about its applicability in the present day): reductively summarized as computers finally becoming powerful enough to support the vast neural networks that ChatGPT and other Large Language Models (“LLMs”) demand.
It is precisely this consistent, steady increase in computing power that answers the question Venture Capitalists (“VCs”) love to ask: “why now?” (as in, why is now the right time for this technology to come to market and bear fruit). Why is 2023 writing itself into history books as the year that decades of dreaming about programmable, useful AI came within reach.
While LLMs, such as those from current leaders OpenAI/Microsoft, Anthropic/Google, and a handful of others (including emerging open source models) understandably receive the lion’s share of the credit along with seemingly endless media attention, there are two unsung heroes that make this all tick.
“Why is 2023 writing itself into history books as the year that decades of dreaming about programmable, useful AI came within reach.”
One is the vast amounts of accessible online data–coupled with the supporting web crawler & scraper infrastructure–that has been invaluable to training these advanced algorithms, which is beyond the scope of this article.
The second, where I’d like to draw focus, is the reason this article is published in Semiconductor Review magazine: advances in the chips themselves! As the saying goes, “in a gold rush, sell picks & shovels.”
Make no mistake: given the sheer volumes of investor capital flowing into AI these days, fortunes will be made and lost in the glitzy “gold layer” also (i.e., the most headline-grabbing companies and applications being built atop the LLMs).
According to PitchBook, the global Generative AI market is projected to hit $42.6 billion this year, and continue growing at 32% annually, to $98.1 billion by 2026. $4.5 billion of investor capital was already invested in 2022 (the vast majority of that before the broad release of ChatGPT). To the impartial observer, it may feel like the onset of another potential hype cycle (though calling bubbles is notoriously difficult: always clear in hindsight but near-impossible to discern ex-ante).
So what’s a sober-minded, disciplined investor to do? Seek refuge where others fear (or lack the expertise to) tread, parse through the noise, and avoid the crowds (which might be a relative judgment call in a sector as hyped as Generative AI seems to be). Done right, these simple investing precepts can be similarly rewarding: according to Precedence Research, the global AI chip market is expected to reach $227 billion by 2032 – more than enough “gold” to go around for patient capital willing to make bets on emerging semiconductor technologies, just a hair beneath the limelight.
In the private markets–the traditional preserve of VC investors–opportunities abound. In the last several years alone, dozens of upstarts have emerged with novel chip technology to challenge established incumbents and improve upon the status quo. To attract venture dollars, these up-and-comers target improvements across one or multiple key dimensions (e.g., cost, performance, power consumption, latency, scalability, etc.), and need to be materially superior–sometimes by an order of magnitude–to existing technologies to be considered for investment (in order to justify the outsized risks taken by the VCs, and to compensate for the high failure rate of venture investments).
Examples of areas that have garnered significant VC attention include:
● AI for Edge computing (e.g., for automotive, sensor, and other IoT applications), such as the analog in-memory computation being developed by Analog Inference
● AI Accelerators (high-performance parallel computation machines specifically designed for the efficient processing of AI workloads) such as the Wafer-Scale Engine chip pioneered by Cerebras
● Neuromorphic Computing (in which chip design is modeled after systems in the human brain and nervous system), such as the Akida AI Processor kits and boards manufactured by Brainchip.
Notwithstanding the above, one need not be confined exclusively to privately-held company investing to make a handsome return, as Nvidia’s (ticker: NVDA) recent meteoric rise clearly demonstrates, becoming the first chip maker to reach a market capitalization of $1 trillion, on the back of insatiable demand for their Graphics Processing Units (“GPUs”) and a redoubled commitment to dominating the LLM market.
Of course, the competition is not standing still, with other industry juggernauts following suit and expanding their own in-house chip manufacturing capabilities. Google, in addition to a deepening partnership with OpenAI competitor Anthropic on the LLM front, recently launched a hardware salvo directed at Nvidia, with Google claiming its Tensor Processing Unit (TPU) chips deliver superior speed and energy efficiency.
Microsoft, for its part, has been developing its own chip (the “Athena”) since early 2019, and already making it available to a small group of Microsoft and Open AI employees. In addition, Microsoft is reportedly partnering with AMD to present their own challenge to Nvidia, despite Microsoft’s Azure platform being among Nvidia’s largest customers, and a still-extant collaboration between Microsoft and Nvidia to jointly build the market-leading cloud AI supercomputer.
Lastly, other prominent incumbents like Amazon and Facebook are also investing billions into the technology and entering the arms race to develop their own in-house AI chips.
So continue to watch this space: while silicon wafers don’t garner the same degree of media attention and widespread adulation as the chatbots composing perfect prose, original music, and intricate works of art, it is precisely these picks & shovels that will usher in the advances of tomorrow.
