GSIT - GSI Technology: Maybe A Push Is All The Stock Needed

2023-05-15 14:41:29 ET

Summary

• GSIT just popped more than 200% last Friday, holding its levels with a further climb of over 20% as of this moment.
• The catalysts seem to be a mention of the company on Fox Business, where GSIT was proposed as a superior bet to NVDA in terms of AI-pureplay technology.
• Really, the catalyst should have been the fact that they have a deal to provide their chip for the Israeli defense industry, working with radar company Elta Systems.
• It seems we were right in our last article that ChatGPT and the AI mania would reach GSIT eventually. GSIT's APU is probably revolutionary, and only now might people have heard of it.
• Priced around normalised historical levels, that didn't reflect much publicity or interest, GSIT may well have a way to move higher from here as new investors digest the case.

Introduction

GSI Technologies ( GSIT ) is a company we've been following for some years, whose major optionality comes from its associative processing unit, or APU, which can do in-place convolutions, which was extremely advantaged for sparse data like in recommendation algorithms and in search applications.

Until recently, the proof of concept for investors has been the competitions. While legitimate, there were still no commercial agreements for the chip, and the fact that the only time anyone's heard of APUs is around AMD ( AMD ) (entirely unrelated product) was a little concerning, even when asking people on elite teams focused on AI in big tech who would have presumably been in the loop. Things took time, but eventually there was an initial commercial agreement with a very legitimate player in the Israeli defense industry, but the stock was completely unreactive, probably because the company had done nothing for a while and was thrown into the forget bin .

Now, randomly, the stock was mentioned on Fox Business and that seems to have been the catalyst for the recent growth of more than 200%. Because we believe the technology is legitimate, but just unknown and pre-commercial since they're still shopping around their prototype chip, this isn't a sell on the news situation as far as we're concerned. Dissemination of information about GSIT should have a lasting impact, especially given that the valuation is far below the potential addressable market, and initial commercial runs are already in a desirable space of the defense industry.

We were ready to wait for a commercial launch of their next-generation chip, more suited for foundry production, as a catalyst for the stock. We think that those returns may become front-loaded as markets begin to get informed on GSIT and fully appraise the opportunity. Acquisition could be in the cards at some point, especially as big tech companies are at war to be ahead of the pack in the AI opportunity, which by its networked nature is going to be characterised by winner-take-all dynamics.

The stock is likely going to be extremely volatile, it may even become a meme stock, although that would require shorts to enter the picture, and they might not take issue with the story here. We would say this remains a highly speculative opportunity, but is still very much a buy, since some years ago we were even buyers more or less at these levels.

A Summary for those who are new

GSIT is a company that used to just produce SRAM chips, nothing exciting, that had some resilient applications in defense and for satellites, and had features that distinguished them like radiation resistance and so forth. Then they bought a company called MikaMonu, which contained a patent for a chip architecture called an APU that could do in-place matrix operations, meaning it could run matrix multiplications without wasting time fetching the data from the memory - it would do it directly within the memory.

This is especially efficient for search applications and also for recommendation algorithms. Recommendation algorithms are the easiest to visualise as using sparse data, because most 'profiles' on Netflix ( NFLX ) for example would not have been viewers of the majority of series on offer. Most Netflix users see only a very small subset of the catalogue, meaning there are going to be a lot of zeroes for each show that the user hasn't viewed. These mostly empty but also massive matrices are very inefficient to move for computation.

But the reality is that such a computational tool could be useful for a lot of machine learning models and neural networks.

So GSIT has been spending a lot on R&D for bringing MikaMonu's technology to fruition, subsidised by their ongoing SRAM business. In that respect, it's been a highly advantaged VC-like exposure, because cash burn has been structurally limited.

Currently, the company has produced a prototype APU, the Gemini-I, and is in the process of getting its design done of the Gemini-II and getting it ready for foundry production.

But even before the Gemini-II, the company has made its Gemini-I available for some researchers through the cloud, and most importantly they have a deal signed with Elta Systems , which is part of the Israel Aerospace Industries, who are going to try to use the Gemini-I for an application called SAR, which is basically an imaging algorithm that attempts to impute high resolution images from low resolution radar input - that's the idea in a nutshell. This happened some time ago, and we can expect updates in one of the next two quarters, the next one coming tomorrow.

The update will concern whether the APU can beat a GPU in computational efficiency - there is every reason to believe it will, based on related imaging competitions the APU has won in recent times, including some sponsored by Israeli defense. But who knows what will happen here.

The Pop

The pop didn't happen because of any business development, it happened because a guest called Lou Basenese on Fox Business went into some depth on the GSIT case, pitting it directly against Nvidia ( NVDA ) as a superior pure-play AI idea - very topical. While it's not exactly a pure-play AI option, at its current valuation it's fair to say that it is, since by its valuation the low-growth SRAM business meant nothing to markets, let alone the business option they have in the APU.

The stock is up more than 200% since last Friday.

Prices are still around the historical levels it was fluctuating around for the past five years. The stock only really cratered as money flowed out of SPACs and speculative tech; admittedly GSIT was also burning some cash, although not enough to sound the dilution alarm, which now looks even less concerning with the stock price packing a much bigger punch in terms of fundraising than it would last week.

Selling the news can be a good strategy when target returns for more traditional, blue-chip style holdings are around 20-30%. It can really give that extra edge. But for stocks of GSIT's profile, that were obviously unknown before this, and that actually depend on tech companies acknowledging them if acquisition is to be an eventual catalyst, seeing news around GSIT is of tangible business value. The stock was unreactive to any important business updates prior to this, now with some visibility there is actually some transmission from the markets.

This push in visibility is also likely to spread much further. A relatively unknown guest making one point about GSIT on Fox Business is a shock that may need some time to echo out, but it is a very pronounced start. Moreover, the retail crowd could pick up on the stock, and it could quickly turn into a contagion phenomenon like meme stocks. Who knows what can happen as far as the technicals and the trading aspects of this stock go.

Bottom Line

While it is still quite early days for the GSIT commercial strategy, there are plenty of companies that have garnered massive values with an even more distant pie-in-the-sky. The addressable market still makes a joke of the $167 million valuation of GSIT in market cap, especially where the SRAM business has been subsidizing R&D, and some of the traditional VC risks aren't nearly as present like dilution. Moreover, the Gemini-II might start seeing orders towards the end of the year. At the very least, they already have one probationary commercial agreement with a customer. While it's tough to place any sort of target price on GSIT, just look at NVDA's $713 billion valuation, and consider how much of their value comes from expectations around their role in AI (probably a meaningful portion of it).

The risks are that the Elta deal falls through. That would be really bad, because it might mean that the technology isn't as great as investors hoped. Other risks are major delays that start risking dilution, but the severity of that problem is also a function of the stock price, which has now recovered. The stock could also become extremely volatile, especially if the retail crowd enters heavily. New buyers should beware of these risks, but we still believe, especially when looking at historical prices, that GSIT could break upwards a lot further, with its historical prices not being reflective of any publicity on the stock that may now develop.

For further details see: