In a new interview with Y Combinator, Sam Altman, the mastermind behind OpenAI, spilled some wisdom about business and startups. The biggest talking point is that Altman believes this is the best time to start a company. 

He talked at length about how important it is for a business to build defensibility and how small teams can cut coordination costs and gain a first mover advantage. If you’re an AI enthusiast or an aspiring entrepreneur, this article, where we’ll dive into all this and much more, including the future of AI, is just what you need.

This is the best time in the history of technology, ever, to start a company. – Sam Altman

If you’re an entrepreneur working on a startup, Altman opines that this is the best time to do so simply because the ‘ground is shaking’. While AI has admittedly been eating up a lot of jobs and replacing redundant tasks, Sam chooses to look at the brighter side of things. 

First things first, AI-driven changes are happening for everyone. No one is immune. This is where the opportunity for startups lies.

Startups win because they can iterate faster than bigger companies and can do so more economically. Large corporations have a lot of moving parts, which makes them slow to adapt to changes. When the cost advantage fades away, so does their advantage of economies of scale.

Take Tesla for example. When auto giants like Ford, GM, and Toyota were shying away from exploring Electric Vehicles (EVs), Tesla went all in and focused on building EVs without any Internal Combustion Engine (ICE) backup.

They were quick to assess the future and took bold risks. Tesla built their own battery infrastructure, software, and autonomous driving tech. Cut to the present, almost every auto manufacturer wants to add EVs to their range of offerings.

Defensibility

A key aspect of building a successful startup today is identifying a solid defensibility strategy. Defensibility is a mechanism through which companies protect themselves from competitors. Simply put, this is a company‘s USP, its differential, which is incredibly difficult for competitors to copy. 

• There can be various types of defence strategies, such as technology or intellectual property rights, as is the case with OpenAI’s GPT models. 
• Or if you have access to a large amount of data, that can also be considered a solid, defensible strategy. Google is a good example of this. 
• Then there are large companies like Amazon that pursue economies of scale as their differentiation leverage. 
• Other brands like Apple and Nike build defensibility on their brand moat and trust.

Altman believes that surviving in today’s startup arena requires some sort of defensibility, which could possibly come from innovation. When you do something that nobody else is doing, you unlock the unique advantage of worrying less about the competition and focusing more on building your own products. 

Building something defensible is tough when everyone’s chasing the same dream. The most enduring companies usually aren’t doing what everyone else is doing. 

Coordination Costs

Another thing Altman touched upon was the increase in coordination cost across companies. When businesses work in large teams, employees end up spending around 30% to 50% of the time coordinating instead of actually doing anything productive. This is where small teams have a huge leverage. 

Interestingly, during Amazon’s early days, Jeff Bezos introduced a ‘two-pizza rule’. This means that if two pizzas could not fully feed a team, it was too big. Bezos believed smaller teams find it easier to communicate and get things done because they don’t have to deal with the mumbo-jumbo around corporate hierarchies.

Altman linked coordination costs with advancements in AI. He says that thanks to new-gen AI tools, smaller teams can get many things done in a very short amount of time, giving them a nifty edge over large corporate teams. 

He said that when people are empowered with better tools and resources, they don’t just get a marginal increase in output. The change is rather huge, and the quality of work is simply remarkable.

OpenAI is a fitting example of this. It started as a small team of researchers with just a couple of founders calling the shots, which allowed it to move quickly and eclipse other, more resource-rich firms.

Runway ML is also worth mentioning. It’s a small team of AI researchers who built AI tools for video generation, editing, and special effects. The company went on to win an Oscar for the movie Everything Everywhere All At Once.

Where traditional VFX studios require a team of around 50-200 people, a team powered by Runway ML requires less than 10 engineers and designers, who can do weeks of VFX work in just a few hours. This underlines Altman’s statement that small teams can outperform larger teams when given the right tools.

OpenAI and Energy

Altman spoke at length about energy costs during the interview. He said that AI and energy were his two obsessions, but he considered them orthogonal vectors before 2015. 

Only after he built OpenAI did he realize the two were deeply integrated. He stressed the need to maximise energy generation to power running GPUs. 

Interestingly, the conversation pointed to a chart that shows the correlation between per capita GDP and energy consumption. Notably, countries with higher per capita energy consumption were higher in the list of GDP per capita. This means the more energy an economy consumes, the more it thrives.

300M+ users on ChatGPT punch in a total of around 1B queries every day. OpenAI consumes 0.0029 kilowatt-hours of electricity per query, or around 2.9 million kilowatt-hours per day.

This is a hundred thousand times more power than a typical US household consumes daily. The annual consumption comes to 1,058.5 GWh, costing $139.72M annually, equal to the consumption of a small country like Barbados.

Naturally, energy cost will become a bottleneck for OpenAI in the long run. Altman is obviously aware of this, which is why he has been investing heavily to make the company self-reliant – just stop saying please and thank you.

For example, the $500B Stargate project is an essential cog in the wheel. The project explores various energy sources and technologies, including solar power integration and battery storage systems. 

This is also where Small Modular Reactors (SMRs) come into the picture. SMRs are compact nuclear fusion reactors with the capacity of producing around 300 MW(e) per unit, one-third of a traditional nuclear power plant. 

Because SMRs are compact, they can not only sit within the perimeter of OpenAI’s data centres but can also be shipped and installed on-site.

The Future of AI

Altman was pretty hush-hush when asked about upcoming OpenAI projects. However, he did murmur about a multi-model AI assistant, an all-in-one version of various products, including image and video generators, Deep Research, and more.

Expectedly, Altman admitted that Artificial General Intelligence (AGI) is the next big thing in the AI domain. And he also hinted at the possibility of human robots with AI capabilities. 

So far, robotics and AI have been growing in parallel without much convergence. However, Altman believes this will change in the next five years, as we could see AI-driven robots. This would essentially wipe away the concept of interface, meaning we’d no longer need to sit behind a screen to harness the power of AI.

Open AI also announced a collaboration with Figure AI, a start-up building human robots for logistics, manufacturing, warehouse, and retail jobs. However, the deal didn’t go through. Nonetheless, the OpenAI founder has been vocal about how humanoids will change the world.

I don’t think the world has really had the humanoid robots moment yet. – Sam Altman

The Product Overhang

In the interview, Altman highlighted that whenever a discovery is made, society and the economy as a whole figure out how to create value from it. 

For example, he says that while ChatGPT is currently only being used as a chatbot, as years pass by, more advanced applications of LLM models will emerge. Five years ago, ChatGPT-3 was launched as an amateur product that was barely good. Cut to the present, it boasts almost a PhD-level intelligence.

Altman believes that we’re currently in ‘product overhang,’ where there’s a huge gap between what the models are capable of and what people can extract from them. Even if the models get no better from here (which they undoubtedly will), there’s still a considerable amount of new stuff to be made.

This drives home the point that there couldn’t be a better time to start a company. You have the best AI tools helping small teams, a gaping product overhang gap, and an ‘unfathomable’ ongoing tech shift. Moments like these hardly come by in decades, so make the most of them. Carpe diem!