The Rise and Fall of a Tech Company

A pilot turned entrepreneur founded RendeRex in Dubai during the COVID-19 pandemic, initially planning to sell 3D rendering workstations but pivoting to AI hardware after adding 'AI' to the website. Despite losing an early bid due to high costs from retail GPU sourcing, the founder secured a contract by offering pre-configured Docker and PyTorch setups, gaining an edge over competitors.

The Rise and Fall of a Tech Company The Beginning It’s 2020, I’m a pilot but its the height of COVID, so not really a lot of jobs. I just turned 30 and have decided I’m going to start a hardware company in Dubai. I call it RendeRex https://renderex.ae . I’m not really sure what it does yet, and I figure I’ll just build and sell workstations for 3D rendering hence the name . Ironically, for the next 4 years of the company’s life, I’ll never sell a single 3D rendering computer. While looking up websites of US companies that provide similar services, I notice that the two largest competitors curiously have an ‘AI’ section. This is miles before ChatGPT, so naturally I just slap ‘AI’ onto RendeRex’s website and call it a day - I figure it will make the site look more ‘professional’. Besides, who’s going to contact me asking about AI anyway? After all, my logo is a T-Rex. Fast forward a couple of months, I’ve gotten a few e-mail inquiries, all of which are for AI workstations from various universities. I should probably learn something about this AI stuff. I start doing some light research. Looks like a lot of math. Pretty cool, but I don’t need to learn how to make AI, I just need to know what it is and how to run it. I learn about the hardware, and I also learn that a lot of this AI stuff runs on Linux. Thanks to an unhealthy obsession with Linux as a teenager for no reason whatsoever , it should be pretty easy to set up some workstations bundled with useful packages and configs, a practice I would later discover is known as DevOps. I start replying to the AI-related inquiries, but no bites yet. An Opportunity I studied Computer Science at university. Never bothered to finish the degree, but got pretty far. One random night, I remember an old friend from university that I haven’t spoken to in years, and decide to give him a call. Turns out he’s working at some big semi-government software company now. I tell him about my newly minted company that also apparently does ‘AI’ now. “Oh? Our company just started an AI department, They want me to write the Bill of Materials for 2 new workstations. Why don’t you write it instead, I’ll submit it, and you can bid on it.” An opportunity. I’m excited and begin frantic research. This was for 2 workstations with 2x 3090’s each, at the height of the crypto mining craze. It would be worth more money than I’ve ever seen a computer be worth. I try to source parts on NewEgg, Amazon, local shops, anywhere. The GPU’s are hard to find but not impossible. I’m sure I have the GPU spacing right for the NVLink, the power supply is the right wattage, and so on. I submit my bid. The procurement department promptly ignores my bid. Over the next few weeks I am the most annoying supplier on their list. The procurement officer politely tells me that my bid is 30% higher than the nearest competing offer. But how? I’ve kept my margins very low to try and win my first customer, but my competitors obviously aren’t buying their parts on NewEgg. Something however, changes. The new head of the AI department is unhappy with the other bidders, and starts calling suppliers directly. He asks me if RendeRex can pre-install Docker with the NVIDIA Container runtime, PyTorch, and a list of other demands. He sounds relieved when I say yes, and asks if I can deliver by Monday. I deliver the workstations, and learn that I have an advantage. He gives me a heads up that in the next few weeks, they’re going to be ordering GPU servers next, and that I should be prepared. He also wants pre-configured virtual machines they can easily spin up with GPU’s. I need to learn some new things, but first, I need to do something about improving my parts sourcing. They want servers with 8x 3090 GPU’s, and there are only 2 companies which make blower-style 3090’s that would fit in a standard server chassis. I begin harassing ASUS customer support, who keep pointing me to their distributor, who then tells me they don’t sell the Turbo RTX 3090 anymore. I try customer support again - relentlessly. They eventually cave and forward my inquiry to the business department, who invite me to coffee. The business exec begins the meeting by informing me that, the only reason they’ve agreed to meet at all is because its COVID, and this is the only way they can get out of the house during business hours. Good start. I explain the situation and drop the name of the big semi-government I delivered the workstations to, Big Company TM . I tell him I built their entire AI hardware infrastructure, which is technically true, ignoring the fact that it was only 2 baby workstations. It works, and I see his eyes light up with dreams of a promotion. ASUS had only recently gotten into the enterprise server game and weren’t doing well on sales, at least definitely not in our region. Closing Big Company TM would be a big win, but the problem was that NVIDIA didn’t want anyone selling 3090’s in servers, and had apparently warned ASUS to stop selling the Turbo RTX 3090 under threat of sanctions. So, there’s a GPU shortage in the market, and ASUS has a warehouse full of these 3090’s they can’t sell. A few days later I’m on an e-mail chain with 40+ other people, none of whose names I can read, let alone pronounce. E-mails are bouncing back and forth at 3AM, by the end of the e-mail dance, I’m told that HQ has decided that RendeRex can, exclusively, buy as many GPU’s as it wants, at distributor prices. The Rise We deliver server after server. New customers keep coming and, pretty soon my T-Rex logo is a familiar sight in data centers across the country. The head of AI at Big Company TM asks if I can fit an AI inference computer into a car. With the help of a friend I quickly mangle together a horrible prototype of a computer with a DC power supply and an NVIDIA T4 with a 3D printed blower mount. I guess he likes the general direction, because he’s set up a meeting with Big Company ’s CEO. I’m completely unprepared. He asks where my presentation is, “What presentation?”, I ask. He throws around terms like soft and hard deliverables, I have no idea what he’s talking about. Somehow, I walk out of there with an offer for an engineering contract, where Big Company would pay RendeRex a monthly fee to build them custom hardware. It’s enough for me to hire some decent engineers. I learn a lot about corporate structure, proposals and business terms. Fast forward a year, and everything is pretty good. I have a genuinely incredible team of engineers, and we’re building things for Big Company TM . Although the work is demanding and somewhat boring, it’s decent and steady. By this point the RTX 4090 has released, which ASUS didn’t have a blower-style GPU for that could fit into servers, so we have to compete fair-and-square with everyone else on standard enterprise GPU’s. Obviously, this is a lot harder than exclusively having access to blower-style 3090’s, so we don’t really make much on server sales. In the summer, one of my friends is getting married in Italy. I fly over to attend, and figure while I’m in Europe, I might as well visit some companies. One of these companies is a water-cooling outfit in Slovenia called EKWB. The team that greets me are very pleasant, show me around their facilities, and end the tour by showing me their 7x GPU water-cooled workstation. There’s a few problems with it, and they tell me their sales are lacking. One of the problems is that their custom fan controller is horrendous and doesn’t really work. By this point we’ve already delivered a bunch of electronics to Big Company . In a couple of weeks, we have a working, sleek controller whose design is based on the popular AquaComputer Octo. We agree with EKWB to buy their chassis and water cooling components, white-labelled. We make custom cables, and small tweaks, we put 7x RTX6000 PRO GPU’s in it and start advertising it as a new offering. The orders won’t stop coming. We can’t fulfill them fast enough. This is a $100,000 machine, and customers are ordering them in bulk. It’s so popular, and we have no competitors, that the suppliers we previously competed with for server sales are constantly trying to buy out our inventory, so they can at least bid on government contracts asking for our workstations. The Fall It’s 2023. We need to start working on the next generation. EKWB seems to be falling apart, they have excellent engineers, but aren’t very well managed and orders for parts are being delayed. We start buying in bulk, we also design our own chassis, and start working on our own water blocks. We begin looking at water-cooled servers. It’s obvious that this is the next step. We’ve made quite a lot from the workstation sales, and the engineers are excited to work on something a little more challenging, but we can’t juggle the work for both Big Company TM and for ourselves. So we decide to end the contract with Big Company . We start designing our own daughter boards to fit our GPU’s, and we’re reading the PCIe spec as if it were a religious document. The orders keep coming, it’s overwhelming and we need more people. We decide to sign a lease on much larger office space, we’re going to need it. We start hiring sales, pre-sales, technicians, etc., and we buy a lot of additional engineering equipment. There’s a ceremony coming up, hosted by NVIDIA’s main distributor, PNY. They’re giving us an award for our successes. We’re already pretty popular with NVIDIA, we’re a partner, and I’ve met Jensen Huang at a private event for partners. There’s going to be some high-up execs from NVIDIA at this event too. We’re feeling pretty good about ourselves. We have a meeting with NVIDIA and PNY after the event. At the meeting, something’s off. We’ve just put in an order for 35x H100 GPU’s for 5 workstations, and they tell us it’s going to be at least 6 months before they can get us the GPU’s - maybe longer. A few days later, the United States announces it is imposing GPU restrictions on many parts of the world, including our region. At first, they tell us we just have to fill out some forms. “We’re friends with the United States, they just need to make sure the GPU’s aren’t going to the ‘bad guys’.” We fill out the forms, there’s an online portal where we can track the status. Months go by, our request bounces from department to department, they ask us to fill out more forms. It takes 9 months to get an approval, and the customer has long since aborted. We have a lot of inventory: chassis, controllers, water blocks, all of which can’t be sold without GPU’s. We have a lease on substantial office space, new hires and existing staff, new equipment. No sales. Hope We’re in 2024. We keep the engineering staff, and no-one else. I know a lot more about DevOps now, and I provided a lot of free DevOps services to previous customers as a major selling point, but I’m stuck handling warranties, support, etc., and it’s overwhelming doing it alone. I go to an out-of-office meeting with our Hardware Design Engineer, M . I forget to pay for parking, as usual, which is a bit of a nightmare in Dubai. The fines are steep and automated unlike payments . I know this all too well, because the in-vehicle inference computer we had built for The Big Company TM , was for an in-vehicle LPR + automatic parking fines system. M is one of the most brilliant engineers I’ve ever met, and he tells me he has an idea for a product he’s been wanting to build for a while. I listen - it’s beautiful. He describes a small, USB you plug into your car and forget about. When you switch off your car, it loses power and wakes up, sending its GPS co-ordinates to a server, which checks if these co-ordinates are a paid parking zone, automatically pays on your behalf, and continues renewing until the device regains power, i.e., you switch your car back on. I would never have to worry about paying for parking again. I would never have to worry about renewing parking fees again. I wouldn’t even have to think about parking. We rush back to the office and start brainstorming, the ideas echo against the walls of our massive, empty office space. It doesn’t matter, we have a good idea , and we have the means to turn it into reality. We spend the next few months designing, prototyping, experimenting. We try out different kinds of technologies with varying degrees of precision, we’re worried about getting it right . Ultra-Wide Band? How many anchors do we need? What signal refraction can we tolerate? Are there any areas that would require Real-Time Kinematic positioning? The office, big or small, suddenly didn’t matter. We spend most of our time outside anyway with our dev kits, testing, documenting. There’s one little catch. In order for this to work, we need an API endpoint from the government to be able to pay for parking on the user’s behalf. It already exists, and several agencies have access to this endpoint. The telecom provider for example, allows you to pay for parking via SMS, and takes a chunky $0.10 fee per transaction. We’re confident we can convince at least one agency to provide us with an API endpoint. We’ll sweeten the deal, share profits from transaction fees, which will be lower because we’re the good guys. We’ll even onboard an investor and offer to pay in advance, I’m sure we’ll figure it out. We start building. The board, firmware, backend, and a fully functional web frontend complete with an admin panel. We want to impress. Our first devices arrive. We make presentations and business proposals. I’ve made plenty for The Big Company TM . Bouncing around from one agency to the next, it’s difficult to get meetings. If we meet with someone who has access but isn’t the decision maker, they ask for a bribe. If we meet with someone at the top, they smile, thank us for coming and tell us they’ll seriously consider our proposal. I’m confused. This is a ‘you win or we lose’ type of situation. There’s no risk to anyone but us, the solution costs nothing, and if users like the product then surely it reflects positively on everyone? Over the following months we keep trying. Meanwhile one announcement after the next is made by various agencies. AI cameras start to get installed. It doesn’t matter what users want. Parking prices go up, ‘VIP’ spots for 10x the cost of a normal parking spot, ‘surge pricing’. Contracts have already been signed, this is the direction it was always going, we just didn’t see it. We probably contributed quite a bit by selling and setting up all that compute.