Hi mcm1636,
I've previously had a thought along similar lines but it seemed a bit too speculative for my liking. However, given some recent information coming to light I think I might share it.
In a Brainchip video from late 2019 it showed a clip of what looked like an NVIDIA simulator (this has been mentioned by someone in the past on here, sorry I can't remember who but see the link at bottom). This indicates Brainchip may have approached NVIDIA to discuss licensing the Akida technology. This was a few years ago, around the time Brainchip started developing the Akida chip IIRC. This was back when no none would buy an IP license because customers wanted physical proof it worked.
If NVIDIA saw the true potential of Brainchip, they would have come to the conclusion that may become a GPU killer (this has been verified by the Ford engineer Chapman spoke to
https://hotcopper.com.au/threads/ford-utilising-neuromorphic.6535212/page-205?post_id=58993104).
If NVIDIA could buy out Brainchip they'd have a chance of not only selling the most advanced neuromorphic processor, but they could continue to sell GPU's on the side (they'd have no direct competition in the neuromorphic space as Intel and IBM were still at least 5-10 years behind according to Peter at that point in time). If they controlled Brainchip they could keep gradually start winding down GPU sales once the neuromorphic chips hit commercialisation and still retain a leading position in the market.
In contrast, if they couldn't buy out Brainchip there would be the potential for Brainchip to decimate their industry. While Brainchip only currently have 4-bit processors (which some might argue aren't accurate enough for top-of-the-range machine learning models, and which would normally be run in the likes of data-centres, where GPU's are currently sold in huge volumes. At least for the short term this might matter, until Brainchip finishes developing a cortical column Akida and it radically changes the way things are done).
Now assume NVIDIA approached Brainchip for an informal chat to see what it would cost to buy them out. Peter said nope, this is my life's dream and I'm never selling. Since Peter and his friends owned at least 25% of the company, there was no way to force a takeover either. So NVIDIA gave up on that idea.
So with that thought in mind, NVIDIA hear Brainchip are developing an Akida chip and they hear it's CPU agnostic. So they look to identify the best, most popular CPU out there and decide to buy them out instead. ARM.
By taking over ARM it would ensure they still retain a solid size in the hardware industry. ARM will probably be used in most AKD1000 chips due to it being so popular (and this is now more likely given so many ex-ARM employees are working at high level positions in Brainchip). ARM are also used in high volumes in datacentres. By gaining control of ARM, NVIDIA may have had a chance of retaining a foothold in the data centre, which would give them more leverage to try keep GPU's getting sold to datacentres, at least for the short term. If ARM was used in most Akida chips, it would also mean they'll end up most of the new revolutionary applications Brainchip opens up like portable COVID breathalysers. In addition, it would show NVIDIA as having more foresight than most other leaders.
Unfortunately the NVIDIA ARM deal is looking like it may fail. If it fails NVIDIA will have $40 billion dollars cash spare sitting around and may be looking for another way to stay alive. The stakes are higher now that Brainchip are just beginning commercialisation. Other large players are also aware of the advantages Akida would bring, and the news from Mercedes ensures every company in the automotive industry will have have heard of Akida. If they're not careful, NVIDIA risk losing a huge part of their company over the next few years - GPU sales.
The most obvious solution would be to integrate Akida IP into their GPU's to make them more efficient. But this approach would only but them limited time (until which more advanced versions of Akida potentially make GPU's obsolete). Buying a stake in Brainchip won't ensure work for the large number of NVIDIA employees, plus I have a suspicion NDA's may prevent NVIDIA's may be preventing big companies from buying a stake in Brainchip - otherwise I think you would have seen other big companies like Samsung or DELL trying to do so (especially other US companies given NASA's interest)
@Fact Finder may be able to provide more insight from his lawyer point of view here. I think there may be a common clause for start-ups to prevent big companies buying them out and changing the company strategy. On top of this it's not a good look if the only way NVIDIA can survive is by buying out a small, smarter company and relying purely on their innovations to survive. If this is the case, most of the billions in GPU research will be wasted, indicating poor use of shareholder money and it will show a lack of foresight by the leader. NVIDIA will also struggle to repurpose their GPU's for other applications given how power hungry they are and their greater impact on climate change - something a lot of leaders will have at the top of their minds at the moment with renewed global interest. Instead you may find NVIDIA pivoting strongly into other new areas - one of the more obvious areas could be developing games and applications for the AR / VR space which Brainchip should help properly open up. NVIDIA do have a huge amount of patents which reference mention neuromorphic chips. I initially thought this was just them covering themselves but it may be more about trying to survive that bit longer. It will be interesting to watch NVIDIA's next mores as their future may depend on it.
Pure speculation, DYOR
Brainchip video with what appears to be an NVIDIA driving simulation:NVIDIA driving simulation video
(20min delay)
