The Middle East wants chips badly but US restrictions prevent the US most powerful chips entering into certain geographical areas such as China and the Middle East; this entails the UAE. Nevertheless, the UAE wants in on the GPU chips craze and is apparently willing to pay anything to enter into the chip bonanza.

The UAE can't get Nvidia's, AMD's, or Intel's best GPU chips so instead they are investing heavily into alternatives such as Cerebras. There is nothing inherently wrong with this but it's also the biggest red flag as an investment opportunity. If the agreements for Nvidia chips open up in the Middle East and specifically with the UAE that could be a crushing blow to this aspiring startup.

Specifically, Cerebras Systems reported a net loss of $66.6 million for the first six months of 2024, on $136.4 million in revenue. For the same period in 2023 it has a net loss of $77.8 million on just $8.7 million in sales. 87% of this revenue for the first half of 2024 was directly from the UAE G42.

The other red flag from this startup is the way in which they promote their business. It's all seemingly smoke and mirrors and conveniently based on outdated GPU pricing and throughput information; which is very publically available.

For some reason, Groq and Cerebras love to keep using memory to unlock speeds on small/tiny models which is impractical and inefficient for a scaled system; or a system that is a large foundational LLM. As well, they have no clue what models will do next so it's a major after the fact architecture that uses llama because they have access to it. https://cerebras.ai/blog/introducing-cerebras-inference-ai-at-instant-speed. A prime example of this is OpenAI's GPT-o1 model that uses reasoning in coordination with its model capabilities. Because they are not being used on the forefront of this technology they don't know how or when a model's size, function, or needs will evolve into the future.

All of this plus the pricing from OpenAI and Microsoft is coming down exponentially.

For Example:

1. They are referring to a 70b param model that shoves an entire model onto memory.
2. They are going up against h100's which is a very old technology at this point. They make no reference to h200's let alone blackwell
3. Because they are referencing such a small model the pricing model they suggest would be radically different for a 400b param model and forget about trillion param models which are coming next.
4. They're not being truthful about tokens per s. As of today this is Azure GPT 4o and GPT 4 mini tokens per minute

gpt-4o & GPT-4 Turbo global standard

As you can clearly see 30 million tokens per minute is 500k tokens per second and mini is 833,333 tokens per second. So i don't know why they are referring to 20 tokens per second or their 450 tokens per second seems way off. maybe they mean million. Even if that is the case and 70 b would be more like mini it is way higher than their limit.

On pricing which they lay out a 3:1 input versus output is fine the price would be for mini which is a comparable model is roughly .10 cents (input) .20 cents = .30 cents. Per million.

for regular 4o it would be higher and let's face it GPT 4o is a far superior model than llama 3.1

3.33 dollar 5 = $8.33

Source:

https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/

From this information what I can tell you is that what we just went over is pricing. it is not some guarantee for what a model produces per second. That shit is very random. What I can tell you is from GPT 4 to GPT 4 turbo to Gpt 4o the speed is dramatically better. GPT 4o mini is damn near real time. Take that for what it's worth.

I am not saying they're being dishonest here but I am saying they are being very cheeky with how they advertise things.

The company is currently being primarily supported by the UAE, with investments worth roughly $900 million for new AI supercomputers known as the Condor Galaxy series. Any growth here is singular to this source of investment and not driven by organic growth or usage. This is not a competitor to Nvidia but rather a temporary solution in the Middle East until chip embargoes are alleviated.

The media here loves to use sexy headlines on non-technical verifications of what it is they are actually comparing. I.e. H100's are now old, or The fact an entity may need to serve millions of clients... No they are instead reporting self-prompting headlines from Cerebras that say things like our chips are 20x faster than Nvidia.

Live Update:

As I am writing and researching this topic it has just been reported by Reuters that the U.S. is setting a new rule to allow chip shipments to the Middle East including the UAE which is a boon to Nvidia and Microsoft.

US sets new rule that could spur AI chip shipments to the Middle East,centers in the Middle East)

Here are couple excerpts:

WASHINGTON, Sept 30 (Reuters) - The U.S. Commerce Department on Monday unveiled a rule that could ease shipments of artificial intelligence chips like those from Nvidia Corp (NVDA.O), to data centers in the Middle East.

G42, a UAE-based AI company with historic ties to China, has been a focus of those concerns. In April, Microsoft Corp. (MSFT.O), opens new tab announced that it would invest $1.5 billion in the company, with plans to provide G42 with chips and model weights, sophisticated data that improves an AI model's ability to emulate human reasoning.The deal drew scrutiny from China hardliners in Congress, even though G42 said in February that it had divested from China and was accepting constraints imposed on it by the United States to work with American companies.

LOL you can't make this up. Literally this just got reported by Reuters today 9/30/2024. This completely aligns with my argument above regarding the UAE-G42.

With chips now entering into the Middle East from Nvidia and potentially others I don't know how this startup IPO makes it off the ground. I don't mean to be bearish but I don't think this is the time for them to raise an IPO without showing more progress. I could be wrong. As of now, I don't plan on buying any shares.

Instead, I will be adding more shares into Nvidia because now this is bullish news for Nvidia.