Startups bag billions to fill gaps left by chip world giants
(2022/03/15)
- Reference: 1647349031
- News link: https://www.theregister.co.uk/2022/03/15/chip_startups_funding/
- Source link:
Venture capitalists funneled billions into semiconductor startups in 2021, we're told, targeting designers of machine-learning technologies that fulfill specific or niche needs.
PitchBook, a research firm that covers private equity and the financial markets, said this month it [1]tracked 170 funding deals last year, and reckoned that all added up to $9.9bn of investment. The cash injections benefited startups working on AI accelerators, intelligent sensors and devices, and optimizations for deploying machine-learning models, according to the analysts.
In February, S&P Global Market Intelligence told us it had [2]tracked 475 funding rounds for chip startups in 2021, and said those added up to $20bn.
[3]
The biggest funding rounds listed by PitchBook were in AI chip designers on consecutive days in April, when SambaNova scored a $676m investment and Groq landed $300m. Cerebras Systems also received a $250m injection that month. Who says April's the cruelest month?
[4]
[5]
The cash comes as applications move to probabilistic computing models, in which processing and decision making is based on inferences drawn from large piles of data, hence the need for machine-learning acceleration in devices and equipment to perform this all efficiently.
Bigger or better?
The consolidation of the chip industry over the past decade concentrated power in the hands of a few large chip makers including Intel, Nvidia, and others, which have horizontal business models with a large set of chip offerings. Now semiconductor startups are filling the gaps in between the larger chip makers.
Meanwhile, all kinds of businesses are now working on chip development in-house to add automation and smarts to their products, presenting opportunities for specialist processor design firms to license their tech and know-how. All of this has attracted funding.
"There's a lot of work going on with automating everything – whether it's a car or a factory or whatever," said Ruben Roy, senior equity analyst at WestPark Capital. "The more automation you do, the more processing power you need, the more memory and even the more chips you need. That's been a big driver as well."
[6]Canada invests in chip and photonics future
[7]PsiQuantum envisions a datacenter-sized quantum computer
[8]Russian conflict unlikely to harm global ICT spending, yet
[9]Plans for UK rival to Silicon Valley ditched
There's also a renewed interest in analog chips, which are in short supply. That demand opened up opportunities for companies like Mythic, which mixes analog processing units with flash memory in dense, low-power AI chips. Mythic, which was founded in 2012, received a $70m round of funding in May.
"The first four and a half years of the company were heads down, focused on building the technology. We knew the market was going to show up. Fast forward to 2017, when we raised our first venture capital round – that gave us a bunch of energy because now we had capital to build a product," Mythic CEO Mike Henry told The Register .
[10]
The chip shortage is working in Mythic's favor, as the company has seen customers moving away from state-of-the-art 5nm and 7nm process nodes that are hard to source. These clients are turning to mature nodes that are easier to source, according to Henry. Mythic's analog AI chips are made using these older process nodes, which means customers can more easily get hold of them.
While Intel and TSMC focus their investments on cutting-edge nodes, companies like Texas Instruments, NXP and ST Microelectronics are investing in mature nodes based on older processes.
"You have much more variety to pick from, so you have a better supply chain. But now you need to augment the artificial intelligence, and that's where we come in. That really plays to our advantage at the end of the day," Henry said.
[11]
Software startups have been the darling of VCs, and a lot more money and product planning is needed to put chips into customers' hands, Cerebras Systems CEO Andrew Feldman told The Register .
"It usually costs more than $50m to see the first product in a semiconductor startup," Feldman said. Cerebras has so far raised more than $730m, and was last year valued at more than $4.1 billion.
"From start to first product in a semiconductor company will take an order of three years. As a result, the customer to semiconductor product feedback loop is much longer. VCs don't get feedback for years after they invest capital," Feldman said.
His company was able to raise money because it solved problems relating to building a [12]wafer-scale chip , which hadn't been done previously, according to Feldman. Its CS-2 silicon is a wafer-sized mega-chip with 850,000 cores for large-scale AI computing. The biz is now [13]rolling out its offerings to customers in sectors that include pharmaceuticals and oil and gas exploration.
The semiconductor market is [14]expected to reach $680.6bn in sales this year, driven by electrification of cars and growing demand for computers and electronics as people rely more on the technology. The growth in cloud services is also putting upward pressure on datacenter demand for semiconductors, PitchBook argued.
The numbers from PitchBook are focused mostly on startups and private equity investors, and don't list funding mega-deals that took place in China. For instance, Chinese company GTA Semiconductor last year scored a $1.2bn investment from institutions including government-run agencies. ®
Get our [15]Tech Resources
[1] https://pitchbook.com/news/articles/vc-startups-semiconductor-ai-ml-innovation
[2] https://www.theregister.com/2022/02/02/chip_startup_funding/
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2YjDGPFItC6kgWsDp2tfLSQAAAAA&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YjDGPFItC6kgWsDp2tfLSQAAAAA&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33YjDGPFItC6kgWsDp2tfLSQAAAAA&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://www.theregister.com/2022/03/09/canadian_chip_investment/
[7] https://www.theregister.com/2022/03/08/psiquantum_data_center_quantum/
[8] https://www.theregister.com/2022/03/07/russia-conflict-idc-spend/
[9] https://www.theregister.com/2022/02/28/oxford_cambridge_arc_stalled/
[10] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YjDGPFItC6kgWsDp2tfLSQAAAAA&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[11] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33YjDGPFItC6kgWsDp2tfLSQAAAAA&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[12] https://www.theregister.com/2022/03/14/cerebras_ai_chips/
[13] https://www.nextplatform.com/2022/03/02/cerebras-shows-off-scale-up-ai-performance-for-big-pharma-and-big-oil/
[14] https://www.theregister.com/2022/01/10/chip_growth_this_year/
[15] https://whitepapers.theregister.com/
PitchBook, a research firm that covers private equity and the financial markets, said this month it [1]tracked 170 funding deals last year, and reckoned that all added up to $9.9bn of investment. The cash injections benefited startups working on AI accelerators, intelligent sensors and devices, and optimizations for deploying machine-learning models, according to the analysts.
In February, S&P Global Market Intelligence told us it had [2]tracked 475 funding rounds for chip startups in 2021, and said those added up to $20bn.
[3]
The biggest funding rounds listed by PitchBook were in AI chip designers on consecutive days in April, when SambaNova scored a $676m investment and Groq landed $300m. Cerebras Systems also received a $250m injection that month. Who says April's the cruelest month?
[4]
[5]
The cash comes as applications move to probabilistic computing models, in which processing and decision making is based on inferences drawn from large piles of data, hence the need for machine-learning acceleration in devices and equipment to perform this all efficiently.
Bigger or better?
The consolidation of the chip industry over the past decade concentrated power in the hands of a few large chip makers including Intel, Nvidia, and others, which have horizontal business models with a large set of chip offerings. Now semiconductor startups are filling the gaps in between the larger chip makers.
Meanwhile, all kinds of businesses are now working on chip development in-house to add automation and smarts to their products, presenting opportunities for specialist processor design firms to license their tech and know-how. All of this has attracted funding.
"There's a lot of work going on with automating everything – whether it's a car or a factory or whatever," said Ruben Roy, senior equity analyst at WestPark Capital. "The more automation you do, the more processing power you need, the more memory and even the more chips you need. That's been a big driver as well."
[6]Canada invests in chip and photonics future
[7]PsiQuantum envisions a datacenter-sized quantum computer
[8]Russian conflict unlikely to harm global ICT spending, yet
[9]Plans for UK rival to Silicon Valley ditched
There's also a renewed interest in analog chips, which are in short supply. That demand opened up opportunities for companies like Mythic, which mixes analog processing units with flash memory in dense, low-power AI chips. Mythic, which was founded in 2012, received a $70m round of funding in May.
"The first four and a half years of the company were heads down, focused on building the technology. We knew the market was going to show up. Fast forward to 2017, when we raised our first venture capital round – that gave us a bunch of energy because now we had capital to build a product," Mythic CEO Mike Henry told The Register .
[10]
The chip shortage is working in Mythic's favor, as the company has seen customers moving away from state-of-the-art 5nm and 7nm process nodes that are hard to source. These clients are turning to mature nodes that are easier to source, according to Henry. Mythic's analog AI chips are made using these older process nodes, which means customers can more easily get hold of them.
While Intel and TSMC focus their investments on cutting-edge nodes, companies like Texas Instruments, NXP and ST Microelectronics are investing in mature nodes based on older processes.
"You have much more variety to pick from, so you have a better supply chain. But now you need to augment the artificial intelligence, and that's where we come in. That really plays to our advantage at the end of the day," Henry said.
[11]
Software startups have been the darling of VCs, and a lot more money and product planning is needed to put chips into customers' hands, Cerebras Systems CEO Andrew Feldman told The Register .
"It usually costs more than $50m to see the first product in a semiconductor startup," Feldman said. Cerebras has so far raised more than $730m, and was last year valued at more than $4.1 billion.
"From start to first product in a semiconductor company will take an order of three years. As a result, the customer to semiconductor product feedback loop is much longer. VCs don't get feedback for years after they invest capital," Feldman said.
His company was able to raise money because it solved problems relating to building a [12]wafer-scale chip , which hadn't been done previously, according to Feldman. Its CS-2 silicon is a wafer-sized mega-chip with 850,000 cores for large-scale AI computing. The biz is now [13]rolling out its offerings to customers in sectors that include pharmaceuticals and oil and gas exploration.
The semiconductor market is [14]expected to reach $680.6bn in sales this year, driven by electrification of cars and growing demand for computers and electronics as people rely more on the technology. The growth in cloud services is also putting upward pressure on datacenter demand for semiconductors, PitchBook argued.
The numbers from PitchBook are focused mostly on startups and private equity investors, and don't list funding mega-deals that took place in China. For instance, Chinese company GTA Semiconductor last year scored a $1.2bn investment from institutions including government-run agencies. ®
Get our [15]Tech Resources
[1] https://pitchbook.com/news/articles/vc-startups-semiconductor-ai-ml-innovation
[2] https://www.theregister.com/2022/02/02/chip_startup_funding/
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2YjDGPFItC6kgWsDp2tfLSQAAAAA&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YjDGPFItC6kgWsDp2tfLSQAAAAA&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33YjDGPFItC6kgWsDp2tfLSQAAAAA&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://www.theregister.com/2022/03/09/canadian_chip_investment/
[7] https://www.theregister.com/2022/03/08/psiquantum_data_center_quantum/
[8] https://www.theregister.com/2022/03/07/russia-conflict-idc-spend/
[9] https://www.theregister.com/2022/02/28/oxford_cambridge_arc_stalled/
[10] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YjDGPFItC6kgWsDp2tfLSQAAAAA&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[11] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33YjDGPFItC6kgWsDp2tfLSQAAAAA&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[12] https://www.theregister.com/2022/03/14/cerebras_ai_chips/
[13] https://www.nextplatform.com/2022/03/02/cerebras-shows-off-scale-up-ai-performance-for-big-pharma-and-big-oil/
[14] https://www.theregister.com/2022/01/10/chip_growth_this_year/
[15] https://whitepapers.theregister.com/
The Next Big IPO Splurge
hoola
Forgive me for being cynical, but currently we need chips manufactured out of the Far East to do useful things not develop all sorts of funky stuff that has no real use.
Of course the VC crowd don't see money in that so through billions at this sort of things. At some point the entire VC/IPO thing has to come crashing down as the numbers now are so out of touch with reality. The issue is that when that happens it will make the dotcom bubble appear totally insignificant.
Re: The Next Big IPO Splurge
Charlie Clark
Also, the bottleneck isn't chip design which has become refreshingly open and competitive over the last few years, but manufacturing. But you don't get many UV lithograph machines for the small change the VCs are handing out.
Oh, and while the size of the market sounds impressive, margins are going down across the range.
Doctor Syntax
"applications move to probabilistic computing models, in which processing and decision making is based on inferences drawn from large piles of data"
Back when I was doing these things it never occurred to me to decide where to address an order based on inferences drawn from large piles of data. I took the simple route of dispatching it to the address given with the order. This might be wild supposition on my part but I think that's still how it's done today and that that's how business collects money from customers. That raises the question of what extra value is added to a business by making inference and whether that added value, if it exists, justifies the expense of making the inferences.
Wafer scale chip? Dont think thats new..
Seem to remember it being tried in the 80s.