The semiconductor industry is attracting more investor attention now than at any time in its storied history. This is because silicon is a key enabler for creating differentiated products for emerging applications and unlocking further industry transformation. Compute is at the center of the modern economy, with a high rate of growth in use cases across all our essential industries. Established incumbents have been unable to move fast enough to meet the urgent need for next-generation silicon needed to transform these industries, which opens the door to semiconductor startups to disrupt existing markets or to introduce brand-new products.
Digging into these topics was the focus of a lively VC panel debate at the recent TSMC 2022 Open Innovation Summit in Santa Clara, CA. The four panelists included Eclipse Partner Kushagra Vaid, who talked about how investor interest in silicon started to pick up six or so years ago driven primarily by the rise of AI workloads, and demand for highly efficient cloud infrastructure. Investments in chip startups reached an estimated record high of $19.4 billion in 2021, up 8% from the previous year.
“Everybody started to realize that silicon is needed to differentiate,” Vaid said. “If you want to differentiate your application and to differentiate on the economics, you have to go and build custom silicon, whether it’s in the cloud, mobile, automotive or IoT.”
Fellow panelist Pete Rodriguez, CEO, Silicon Catalyst agreed. “We see all kinds of solutions that can really benefit mankind, such as gas sensing, all sorts of biological markers, and very unique applications of AI to data on the edge, and security cameras with orders of magnitude better than what’s available today,” he said. “You may need your software, your system and your use model to be a differentiator, but at the heart of it is the semiconductor.”
Every decade or so, there’s a new workload era, which then becomes the economic engine for investments and evaluations, according to Vaid. “In 2015/2016, the time was ripe for that new driver to be the AI workload, which is when VCs started getting back into making silicon investments,” he said.
The panel discussion was moderated by Lucas Tsai, director, market development and emerging business management at TSMC North America. Joining Vaid and Rodriguez were Cheng Wu, General Partner, Tech Fund, Taiwania Capital Management and Andrew Kau, Partner, Walden Catalyst Ventures.
The Future Relies on Silicon Innovations
Tsai asked the panelists to talk about where they’re currently investing and their plans for future funding.
“One of our theses is that digital transformation will eventually impact all of the so-called legacy industries, which encompass about $85 trillion of the world economy,” Vaid said. He pointed to the five semiconductor investments Eclipse has made so far in the advanced compute sector: Cerebras, DataPelago, Flex Logix, Oxide and Tenstorrent.
Looking ahead, Vaid is most excited to invest in startups in three major areas:
- Advanced wireless communications, where 5G technologies are just “the tip of the spear” to transform industrial sectors like manufacturing, transportation and supply chain.
- The disaggregation of resources to completely redefine data centers as interconnected pools of memory, compute and storage resources instead of the current hard-wired racks of servers.
- The massive compute infrastructure needed to support next-generation Web3 technologies, the cryptocurrency economy and the fully immersive experiences promised by the metaverse.
With the amount of data being created or replicated set to rise to an estimated 180 zettabytes in 2025 up from 64 zettabytes in 2020, Tsai asked the panel to talk about the silicon opportunities around data generation.
Kau noted the critical role that next-generation processors will play in helping to turn vast amounts of data into actionable information. “The GPU world, which is dominant now in the AI space, is not necessarily the best way to process a lot of this data,” he said. “We think new architectures may be better for that.”
According to Vaid, currently less than 1% of raw data is eventually used for performing data analytics to drive business insights. Silicon could shift that percentage by providing the compute power to examine data closer to its source. “That’s where the innovation opportunities lie for building new accelerators for moving to the left of the data pipeline and doing much more analysis at the edge,” Vaid said.
To SPAC or Not to SPAC
Given the popularity of SPACs in 2020 and 2021, Tsai was eager to gauge the panel’s opinion on the investment vehicle’s ability to provide a viable exit for semiconductor players versus traditional IPOs or M&A.
“I think you’ve got to be really careful because, like anything else, SPACs have been overhyped and misused to a certain extent,” Vaid said. He noted that over 700 SPACs were created between June 2020 and November 2021, and that since January 2000, SPACs have raised $235 billion.
In analyzing SPACs, Vaid cited the findings in a recently updated academic research paper from economists at Stanford University and New York University. For every $10 raised by a SPAC, on average only $5.40 to $6.60 went to helping the company raise money, the rest went towards sponsor and other fees. If an organization may lose 34% to 46% of its value with a SPAC, it’s worth considering if it’s the right approach, especially since on average, SPACs underperformed on the Nasdaq by 20%.
“You have to ensure that you really have a business case and a good story for why you want to use SPAC to go IPO versus the traditional approach,” Vaid said. “SPAC is going to be a path that a lot of companies will pursue, but the hype needs to settle down.”
Kau pointed to the tension or misalignment of interests between sponsors and investors, especially PIPE investors. “It’s easy to do a SPAC from a lot of dry powder,” he said. “You’ve got a ticking time bomb to find a deal. It’s hard to raise a PIPE now. If you have a choice, I would go through the traditional IPO process.”
A Metaverse of Potential Opportunities
Another area where the panel had diverging views was on the investment potential in semiconductor startups, which enable the metaverse, blockchain and Web 3.0. Concerns were raised about the level of openness of these technologies and which vendors might end up controlling them. While acknowledging that everything is still evolving, Vaid noted the power of millions of people to drive a new workload at scale, which will require customized silicon and systems to make the workload more efficient, in a manner similar to AI silicon investments first made in 2016 timeframe.
As well as speeding up crypto processing with Bitcoin accelerators, there’s a need for highly accurate physics-based simulation in a fully immersive metaverse. “You want to have people interacting with a physics-accurate world where, if I drop a bottle, it will follow the laws of gravity or when I pour the liquid out, it will follow the laws of fluid dynamics,” Vaid said. “While that level of processing is available on a supercomputer, it’s not yet possible in a data center or on a home PC.”
“I think the metaverse in the future needs to be built on top of the assumption that you have to impersonate the avatar in order to live seamlessly in the virtual world,” said Wu. “In order to do that, you are talking a lot of sensor fusion of all the cognitive intelligence that needs to be integrated and a new architecture.”
Vaid mentioned a blog written by Intel’s Raja Koduri about what’s required to create a metaverse, which could deliver on the promise of readily accessible, persistent and immersive computing at scale: A petaflop of compute and a petabyte of data within a millisecond latency to every single user on the planet. “We are one thousand times away from that capability today,” Vaid said. “At the most, today’s technologies will give you 8x or 10x, not 1,000x, but there are a lot of opportunities here for innovation across the silicon and systems stack.”
At Eclipse Ventures, we seek to partner with founders working on disruptive technologies that can modernize the world’s essential industries and open up new markets. Kushagra in particular would love to hear about your big idea related to next generation Cloud and Edge infrastructure, drop him a note at email@example.com.