Every time you open ChatGPT, you are using a supercomputer.
You just do not see it.
That was one of the first things that hit me when I sat down with Nithin Mohan, AI and Supercomputing Leader at HPE, for the latest episode of AI Insights San Francisco. Nithin flew in from Denver specifically for this conversation, and within the first few minutes, he completely reframed the way I think about AI infrastructure.
This one is for anyone who has ever wondered what is actually running underneath the tools we use every day.
Think of It as Thousands of MacBooks Working Together
I asked Nithin to explain supercomputing in plain language, and his answer was clean:
“Think of supercomputing as a bunch of computers connected together at massive scale. What that enables is a simulation you can run on a laptop. There are different kinds of simulation in the world which are very compute intensive. Think drug discovery. Think COVID research.”
I pushed back with a practical example. My MacBook M3 is pretty fast, but a supercomputer would be like thousands of those machines all focused on one task at the same time. Nithin confirmed it immediately.
That is the core idea. Supercomputing is not some distant, futuristic concept. It is the engine running behind the AI tools you already use daily.
The Aviation Metaphor That Changed How I See This
This is where the conversation really opened up for me.
Nithin described supercomputing the way aviation changed the world. Before commercial flights, only a small number of people could travel long distances. Aviation democratized access to the world. Now almost a billion people fly every year.
AI and supercomputing are doing the same thing for creativity.
“What took a movie studio thousands of people to create, can now be done by the combination of generative AI and access to massive scale computing by 4 or 5 people.”
He went even further:
“I think it’s very possible that a team of just 2 or 3 people, generative AI, and access to large-scale computing will develop content that’s at the level of our next Harry Potter or Lord of the Rings or Game of Thrones.”
That quote stopped me. It is bold. But when you think about the trajectory of the tools available today, it is not unrealistic.
Startup vs. Enterprise: A Line That No Longer Exists
Nithin has worked in both worlds. He helped scale a startup called SolidFire to a billion-dollar company, worked at NetApp, and now leads a team inside a 60,000-person enterprise.
His take on the startup vs. enterprise debate was honest. Startups move faster because fewer people need to align before a decision gets made. Enterprises have scale and can reach billions of users but coordination slows things down.
What is interesting is that the line between the two is dissolving.
“Would you call OpenAI a startup? It’s a half-trillion dollar market valuation.”
He made a good point. A decade ago, Facebook going public at $50 billion was a record. Now we are talking about companies with a fraction of the headcount already surpassing that. AI has flipped the script on what scale even means.
You Are Already Using Supercomputing Without Knowing It
One of the most interesting moments in the conversation was when I asked about accessibility. Running supercomputing infrastructure requires massive energy, serious money, and deep technical expertise. So who actually has access?
Nithin’s answer was simple: you do.
If you use ChatGPT, Gemini, Grok, or any AI tool, you are already accessing supercomputing through the cloud layer. The infrastructure has been abstracted away. You do not need to know how the airplane is built. You just buy a ticket and get on.
The bigger shift is that supercomputing as a service is still in early days. AWS is primarily traditional compute. But Nithin sees a near future, maybe months to a couple of years away, where you can buy credits and access high-performance supercomputing infrastructure the same way you spin up an EC2 instance today.
Ethics and Bias Are Not Afterthoughts. They Should Not Be.
We covered a topic that I think gets glossed over too often: the ethics and bias problem inside AI systems.
Nithin was clear that these are not soft, philosophical problems. They are deeply technical.
“The higher the quality of the data is, the better the output is. If you are feeding data to a model that’s heavily biased towards a certain population, that’s all the model is going to know.”
His concern is that as investments pour into compute and foundational models, ethics and bias keep getting treated as something to fix later. He believes that if this continues, we are heading toward heavy inequity within one to two decades.
Proactive regulation. Active dialogue. These are his words, and I think he is right.
Small Language Models Are Coming, and They Change Everything
One of the most technically interesting parts of the conversation was about the future architecture of AI.
Right now, the dominant model is one massive foundation model that tries to do everything. Nithin sees that shifting toward small language models, specialized models that each handle specific tasks and communicate with each other.
The analogy he used was computing power itself. In 2008, one petaflops was the record for the world’s most powerful supercomputer. Nvidia recently launched a personal device with the same processing power. In 17 years, we went from world record to something that fits on a desk.
Small language models will follow a similar curve. What runs today on massive shared infrastructure will eventually run on a personal device, customized to your needs.
“In a few years, you are going to have your own creative content model that you can run on a personal supercomputer and achieve your task. It’s the equivalent of everyone having their own private jet.”
The Next 5 Years: 30 to 50 Years of Progress, Compressed
I asked Nithin to look ahead. Where does all of this go over the next five years?
His answer was honest about both the opportunity and the weight of it.
“We have seen in two and a half years what took the internet ten years to reach in terms of impact. If every year is worth seven years in internet age, you’re going to see over the next five years 30 to 50 years worth of progress.”
That is not hype. That is a real extrapolation based on the trajectory we are already on.
The exciting part is what that means for creators, builders, and small teams. The concerning part is what it means for jobs, skills, and who gets left behind if we do not actively address access and equity.
Start With the Problem. Not the Model.
Before we wrapped up, I asked Nithin what someone new to AI at work should know first.
His answer cut straight to it:
“It’s always important to start with the problem. There’s eagerness to jump to the latest and coolest model. But that’s like going to the destination and then figuring out what to do. Start with the problem. Then the data. Then the model.”
That is genuinely the best framing I have heard for how to approach AI adoption. Not which tool is trending. What problem needs solving.
