- CROWDSCALE
- Posts
- THE ATOMBEAM REPORT
THE ATOMBEAM REPORT
the biggest thing in crowdfunding
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/30ed08d2-9b5b-4746-8abc-31287e5fac7b/CROWDSCALE.jpg?t=1712547056)
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/4d7851bb-eb18-4aa8-a5d7-70c11dee849b/CROWDSCALE_Divider.png?t=1712547056)
THE STORY STARTS 1,146 DAYS AGO
For 3+ years I have been closely tracking this startup.
When it first came across my radar, I invested $4,000 and made a 7-minute video detailing their potential.
Back then I didn’t even own a tripod, my camera setup looked like this….
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/3c20a9ad-93e4-4736-8aac-416dbc7b5f34/Video_Setup.jpg?t=1721515168)
a folding table, empty box, couple of boardgames, and a cheap ring light were holding my ‘recording studio’ together!
Despite my lackluster setup, that video caught the attention of their founder, who reached out to me.
He liked my video style and asked me to make paid content to help promote their fundraising efforts.
I could have taken cash as payment - my wedding was coming up and I certainly could have used the money.
But this startup was special - and my mind was made up.
I opted to receive equity in exchange for my services - a decision I would not replicate for the other 12 startups I’ve worked with.
The potential for this startup was far too great to sell out for a quick payday.
To this point, the decision to invest has been one of the greatest in my entire portfolio, at least on paper.
In just 2-3 years, the company has rocketed into a valuation 9.7x higher.
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/660e8fe3-a4c5-43f3-8930-c2d8207ade09/Screenshot_2024-07-22_at_10.01.30_AM.png?t=1721656934)
Atombeam’s valuation has rapidly increased within the past 3 years alone
And in what might be their final fundraise as a private company, this might be investors’ last chance to own equity in this startup.
I decided to go back and fully immerse myself in all elements of the company.
I re-watched hours of webinars, months of progress updates, and interviewed Charles Yeomans, the founder & CEO.
As a result, my belief in this company has only grown stronger.
ONE CHART IS ALL IT TAKES
Every so often, there are enormous shifts in consumer behavior, technology, or business models that create generational opportunities.
These opportunities pave the way for ‘one-chart businesses’.
This is a term that refers to businesses where someone only needs to look at one chart to know it will be a success.
Jeff Bezos, the founder of Amazon, is believed to have seen a chart early on that showed the internet growing by 1,000% a year.
He thought to himself - nothing grows at 1,000%... I need to build in this space.
That one chart led him to create Amazon - a multi-trillion dollar business today.
The startup I’m going to introduce shortly has one chart that blew me away.
Below is a chart that tallies the global number of data being produced by iOT devices alone.
By 2025, we are going to be producing 90 zettabytes of data just from iOT devices. I will be shocked if any of you know how much a zettabyte is - hint, it’s a lot.
We’re talking TWENTY-TWO zeroes at the end of this number.
90,000,000,000,000,000,000,000.
If each byte was a grain of rice, 90 zettabytes would cover the entire United States 100x over.
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/012f77a5-f3d9-4d37-a87c-57dbe44127f0/grains_of_rice.jpg?t=1721409096)
I got carried away with AI imaging on this one…
Charles Yeomans - a seasoned investment banker and prior CEO - saw this trend and set out to build his one-chart business.
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/8b740990-fc6f-4e2f-9ec2-c288aeef9e51/Charles_Yeomans.jpeg?t=1721409464)
Charles Yeomans, CEO of AtomBeam
His co-founder, Asghar Riahi, came up with the idea of data codebooks as a way to solve the problem. But they needed a way to make it real.
Now, Charles had an ace-in-the-hole. An unfair advantage.
Charles had Joshua Cooper.
Cooper, an MIT-educated PhD mathematician who teaches at the University of South Carolina, has published over 60 papers in mathematics and advanced computing.
There are two ways to handle the deluge of data seen by the world today:
Pay for more hardware to handle the increased load
Use software to reduce the size of the data
Charles, Asghar, and Joshua decided there was a massive opportunity in the latter method.
The three were all-in, and created Atombeam in 2017.
Josh took Asghar’s idea and made it into a working software - a new way of compacting data that worked far more efficiently than current solutions.
The initial research quickly watershed into 3 key areas of focus for Atombeam:
Neurpac
Neurcom
Large Codeword Models (LCMs)
I’m going to dive into the technical elements of these breakthroughs, but for this investment opportunity, you really just need to know that Atombeam makes big data → small.
If that’s enough for you, you can scroll past the technical details in this next section (or click here if you are on a web browser to skip ahead).
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/4d7851bb-eb18-4aa8-a5d7-70c11dee849b/CROWDSCALE_Divider.png?t=1712547056)
(1/3) NEURPAC
Since the inception of modern computing, the basic concept of how data files are structured has been based on how humans organize their thoughts.
For example, files generally have a date-time stamp, and the rest of the information in the file is set forth in human-readable terms translated into binary (1’s and 0’s), such as expressing a designating a geolocation, or whether or not to use a capital or lower case letter in a person’s name.
For a binary computer, much of this data is useless and repetitive. The essential information is surrounded by a cloud of (from a computer’s perspective) fluff – the fluff being a combination of human contextualization and unneeded repetition.
This translates to an enormous degree of inefficiency.
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/c805d8e2-c435-4f45-a6c2-247aaeb61693/AtomBeam_Data_Compaction.gif?t=1721407813)
Just how inefficient this approach is can be seen in the amount of reduction compression algorithms can achieve on a typical file; 75% or even greater reduction is common.
The problem is, though, that compressing files with standard methods results in making them unreadable, and so it is simply accepted that usable data = uncompressed data.
Atombeam’s breakthrough is an entirely different way of thinking about data compaction.
It revolves around one central concept: data-as-codewords.
Within the ‘fluff’ data (but also the valuable data), predictable patterns emerge.
Neurpac identifies patterns in the data and assigns shorthand ‘codewords’ to these patterns.
These codewords are merely shorter pieces of code, it’d be like seeing ‘would not’ as a highly repeated phrase and replacing it with “won’t”.
Long strings of binary data can be compressed into shorter ‘codewords’
Let me give you a real-world (but simplified) example of Neurpac in action. Let’s say your smart thermostat is constantly communicating to the app on your phone.
‘Set the living room thermometer to 75 degrees’
‘Set the living room thermometer to 81 degrees’
‘Set the living room thermometer to 67 degrees’
In all these examples, there is repetition in the beginning part of the communication. Atombeam might create a smaller codeword for the phrase ‘set the living room thermometer to’.
Whereas that long phrase may equate to 20 bytes worth of 1’s and 0’s, Atombeam could replace it with a codeword that is only 5 bytes.
Both the thermostat and app have an identical codebook, which allows the codewords to be translated back to their full message once received.
This method of compaction comes with some amazing qualities:
Powerful - Atombeam’s Neurpac can reduce the size of data by 75%
Lossless - Throughout this advanced process, there is no loss in the data or its quality, as the data is fully returned to its original state
Minimal Latency - One would think that translating the data codewords, sending them, then decoding them would take a decent amount of time. But that’s not the case - Neurpac encodes in mere microseconds, it’s unnoticeable to the human senses
Fully Searchable - Generally, compressed data is unreadable. It needs to be uncompressed to do anything with it. Neurpac’s compaction leaves the data searchable and randomly accessible, making it capable of being as useful to a computer as the original data, despite its reduced size like compressed data
It also naturally applies an ultra lightweight encryption to the data being transmitted, since one would need the codebook to make any sense of the commands being sent.
This may not sound important now, but this could be mandatory as we evolve our technology…
People aren’t rushing to encrypt their toaster’s iOT data, even if they’re burning their poor bagels to a crisp.
But what about when driverless cars take the road? Or humanoid robots entering the home?
These machines will constantly emit data, and it could be a major safety liability if their data is compromised.
One of the largest IoT security companies, Palo Alto Networks, says 98% of iOT data currently has no security measures applied - current solutions add too much ‘weight’ to the data for it to remain efficient.
Atombeam does the opposite, adding a “deep obfuscation” security, while actually making the data ‘lighter’.
Security is not the main focus of Atombeam’s value proposition, but it’s a nice added benefit of the technology. And the company is planning a future product called Neurpac+, which would use encryption & obfuscation, but not add any additional computing steps… reducing the data 75%, adding encryption that could be stronger than standard encryption, AND still searchable as though it was in the original form.
But there’s more…
(2/3) NEURCOM
Neurcom resembles Neurpac, but is designed for rich media formats like images, video, and audio.
Neurcom utilizes a neural net to compare different streams of data flowing into an advanced video codec. It uses AI to compare the different streams of data flowing into the codec, such as R/G/B video image streams.
The AI finds correlations in the streams that it uses to reduce the amount of data that requires processing by the codec.
Dense stuff, but it often results in a 50% reduction of data that’s transmitted- smaller in, smaller out.
And, all the good stuff - no loss of quality from the original. About 70% of the internet is video and images - a solution that can cut down the amount of data these require while keeping the same quality is a grossly valuable in my opinion.
Now, Neurcom is a more complicated technology to build and is not yet ready as a commercial product just yet.
However, its applications are real and being tested. Atombeam was awarded a $1.2M contract by the U.S. Air Force to apply the technology to synthetic aperture radar images produced by satellites.
The Air Force is using these images now, but they want more of them and they want them faster.
Neurcom is hard at work bringing this vision to life.
(3/3) LARGE CODEWORD MODELS
This one is exciting to talk about but I should caution that it’s very early stages.
Large Language Models (LLMs) are what AI assistants are built on, they’re data sets with hundreds of billions of parameters.
To train a model, it can take several months, and a lot of money & power, due to the colossal scale of data that needs to be processed.
A Large Codeword Model (LCM) is the concept of feeding the model data that consists entirely of codewords. The idea is that the training can be done up to 50% faster, with an end product far lighter than the bulky data sets we have today.
One of the benefits of the LCM is that it is truly multi-modal, meaning it can ingest various formats of data (video, text, audio, IoT data, etc).
This is because it uses codewords, and not text like LLMs. Codewords can represent anything, including text, but also numerical image data, data that is generated by machines like an airplane or a Mars rover.
LLMs can ONLY understand text, including understanding pictures as they are described as text, which means that they are limited to a static, single moment in time.
LCMs may be able to take in and understand data with date-time stamps, and so understand how things change over time. The LCM may be able to make a prediction, understand from the flow of new data with date-time stamps how well its prediction turned out, and then make a better prediction next time.
Another interesting point on the LCM is that codewords are smaller than the data it represents - again, the data-as-codewords idea - and so the training process for the LCM, taking in oceans of data, may turn out to be much faster than for an LLM.
Maybe twice as fast, consuming much less time, computing resources and power, which is a big issue for generative AI right now.
Atombeam tested LCM as a proof of concept, creating an exceptionally small LCM using the book War and Peace as the data set. Charles has indicated that the test was a success, thereby clearing the way for potential innovation in the AI space and more advanced prototypes.
This is not the core product (yet) and it won’t be available anytime soon, but fascinating to see the exceptionally broad bounds for this technology.
I do not think it is crazy to say that this could become a standard software for every single generative A.I. model.
And if Neurpac/Neurcom take off, Atombeam will have the credibility and size to effectively sell this into the big players in the A.I. space.
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/4d7851bb-eb18-4aa8-a5d7-70c11dee849b/CROWDSCALE_Divider.png?t=1712547056)
SO HOW DOES ANY OF THIS COME TO LIFE?
First and foremost, Atombeam has the full might of the US Patent and Trademark Office at its back. At the time of this writing, the office has granted Atombeam 50 patents, with another 63 pending patents on the way.
Atombeam has been extremely aggressive in protecting its landmark technology, ensuring that they alone have the ability to bring this to market.
Secondly, Atombeam has made huge strides in two areas of business: government contracts and commercial partnerships.
On the gov’t contracts side, Atombeam has been selected for numerous contracts by several branches of the military. Most notable of these, are:
$1.2M Phase 2 Contract with the U.S. Air Force
$1.2M Phase 2 Contract with the U.S. Space Force
This has brought meaningful revenue into the company coffers and sets up Atombeam for more growth. The way DoD contracts work is they take a phased approach to test & ramp up deployment.
Phase 1: Test the technology conceptually at a small scale
Phase 2: Confirm the legitimacy with an increased rollout
Phase 3: Deploy at a large scale
If Atombeam carries out these initial phases effectively, they’ll be prime candidates to receive Phase 3 contracts, which can easily run into the tens of millions of dollars.
On the Commercial Partnerships side, Atombeam has etched out a few key alliances.
REDACTED ($20B+ company) - There is a major deal that is nearly finalized with a huge player in the telecommunications space. I really wish I could say more here, but until the company publicly announces it I can’t divulge more. The revenue potential from this easily reaches the tens of millions of $$$ range.
Viasat ($1.9B company) - Atombeam is working closely as a key technology partner to promote its technology to Viasat’s customers
Intel ($140B company) - Atombeam partnered with the semiconductor giant as a part of its work to integrate its technologies into chips, which it expects will bear fruit in early 2025.
Hewlett Packard Enterprise ($26B company) - Atombeam was accepted into HPE’s highly selective partnering program as a key potential technology provider for HPE’s networking hardware.
Nvidia ($2.9T company) - Atombeam’s Neurpac was identified by Nvidia as having strong potential for AI chip integration in this ongoing partnership.
Atombeam has developed the Neurpac tech, and it has now been productized in the AWS cloud so that it can be applied seamlessly at an enterprise scale. Neurpac now has an easy-to-use interface and has been rigorously tested.
Now that they are ready to roll, I imagine the floodgates holding back demand will be opening up in the fall of this year.
Atombeam has not announced pricing for their technology, but given conversations with the founder I am expecting this to be a high-margin business model (95%+).
Given the end-goal is to make this a plug-and-play software application, copies of software are pretty close to free for Atombeam.
One of the reasons I am particularly bullish on its adoption is that In many cases, this can be pitched as a cost-savings measure. But Neurpac’s biggest appeal is in cases where network congestion is holding up revenue opportunities - which is more common than you would think.
If Atombeam can increase data bandwidth by 4x for a company that’s using satellites, that company will need to launch fewer satellites to handle the capacity.
Geosynchronous satellites can cost $500 million - they’re expensive pieces of equipment. Once they are there, you can’t upgrade the hardware, so if you are a “geo” operator (like Viasat) you need to figure out how to get more out of the satellites you have.
This enables Atombeam to position itself as a lower-cost software option in contrast to an expensive hardware solution like launching a whole new constellation of satellites.
The decision becomes quite clear - increase current equipment by ~4x with a lightweight software solution, or spend a few billion dollars to launch a dozen new ones.
Just one example, but it’s applicable across multiple types of hardware and software costs that companies incur.
FOUNDING TEAM
I’ve gotten to know Charles over the past few years, and here’s what I can tell you.
He is uniquely positioned to take this thing to the next level.
Charles has the energy & charisma of a young Stanford grad, but the judgment of a wise business executive with decades of experience.
He’s able to rally people around his grand vision - the man has gotten 3,600 people to care enough about a boring (sorry Charles) data business, so much so that they invested their own money into it.
While painting broad strokes of the grand plan, Charles does not shy away from the technical aspects of the products. He knows his product and can speak to it thoroughly without having to pass off to his engineering team.
As a previous intelligence officer in the U.S. Navy, Charles is able to speak with the Department of Defense in their language, knowing full well what their priorities are.
This has led to tremendous success in scoring $1M+ contracts that are bringing in valuable revenue to fuel Atombeam’s R&D while also lending major credibility to the startup.
I’ve observed Charles from near and from afar, and I can tell you that he has the strategic leadership that the company needs to blaze a trail forward.
Through directly investing and equity compensation for promotional work, I’ve invested approximately $9,200 into Atombeam, which has a book value of $66,000 at the current valuation. That’s more than any of the other 39 startups in my investment portfolio.
Now you don’t need $9,200 to invest in Atombeam.
Their round is open and accepting investments at a minimum of $648 ($8.28 per share).
Check out their investment page below for more details, and be advised that this fundraise is only open for a limited time.
DISCLOSURE: I am not being compensated by Atombeam in any way to write this review. This is not investment advice, only the author’s own thoughts and opinions. Please conduct your own due diligence before deciding to invest and only consider investing that which you can afford to lose.
Did you like this article? |
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/4d7851bb-eb18-4aa8-a5d7-70c11dee849b/CROWDSCALE_Divider.png?t=1712547056)
Reply