When ChatGPT was unveiled to the public late last year, it stunned users with its humanlike conversations. The souped-up chatbot — a “large language model,” in the parlance of the field — can banter about restaurants, assist with searches and even plumb philosophical questions, all in naturalistic language that easily passes for a real person’s.

More than a million users flocked to the system within a matter of days, overwhelming its servers and creating digital queues. Google rushed to unveil its own long-simmering artificial intelligence project, Bard — seemingly the start of an AI technology arms race.

Perhaps surprisingly given the hoopla, ChatGPT isn’t based on new technology. Many big players in the AI space have been refining machine learning for years and even decades; companies such as Google and Meta were jogging in this race long before the recent starter pistol fired. IBM’s Deep Blue system famously defeated Garry Kasparov in a chess match in 1997, for example, and AI is already widely used to streamline industrial logistics, create innovative medications and design new products.

But there is a major difference catapulting ChatGPT and systems like it: Computing power has taken a startling leap forward, increasing by a millionfold since 2017. That jump is transformative, with advances coming from more-powerful chips and new types of computing accelerators; data centers of a scale that’s never been seen before; and machine learning breakthroughs with more-efficient systems and larger models. That magnitude of improvement could usher in a golden age for AI development. 

Of course, these technologies are still rapidly evolving and maturing. There are going to be unexpected hurdles and false starts. For example, large language models sometimes “hallucinate” and gin up fake sources or confidently respond with nonsense. Over the long run, I believe AI will profoundly change the world, but — like many revolutions — progress will move in fits and starts. Some companies will overestimate AI and overpromise as they develop the technology in coming years.

Similarly, today’s technology has its limits. Researchers haven’t yet cracked the problem of “artificial general intelligence,” or truly adaptive machines. Current technology is commonly called “artificial narrow intelligence,” or ANI, reflecting that each application has to be trained for a specific use. ANI in the form of autonomous driving technology, for example, might be very good at recognizing traffic signs and hazards, but it can’t catalog and detail the works of Matisse the way ChatGPT can. And because these systems are very expensive and time-consuming to train, there are natural limits on who can develop them and what tasks they can be set to, though that might ease in coming years.

Despite those issues, I fully anticipate that companies from every industry, not just cutting-edge technology firms, will benefit from AI. I also think there are going to be waves of startups deploying the tech in novel ways to tackle seemingly intractable problems.

I envision wide applications across a range of fields, from marketing and sales to agriculture and manufacturing. One can imagine dramatic changes in how we handle all manner of repetitive tasks, freeing us to focus on what matters most in our personal and professional lives. In short, I think AI is going to become as integral to everyday life as smartphones, computers or the internet. Given the explosion in uses that this technology is enabling, I believe the AI revolution will boost many companies and bring compelling investment opportunities.

The investment implications are broad — and already having an effect.

I think AI will disrupt a number of industries, often in unintuitive ways, while creating opportunities for forward-thinking companies and investors. I envision that food production, with its thousands of tiny variables, such as fertilizer usage, planting cycles and animal feed, is ripe for an AI-powered system that can cut costs while potentially increasing yields. Legal systems will likely undergo big changes, as their endless codified laws and tiers of precedents are the perfect target for AI search tools. In the medical field, AI will impact and augment everything from radiology scans to drug discovery to robotic-assisted surgery. 

Hundreds of entrepreneurs also see the possibilities and are building startups with ChatGPT as a foundation. New platforms and businesses will emerge from this technology. While history suggests that many of these nascent companies are unlikely to succeed, some could find unique and compelling uses they can monetize. I think software creation could be automated to some degree. OpenAI, ChatGPT’s creator, has reported that more than 40% of its code is written by machine. And customer service could become far simpler and more effective: Many estimates suggest that nearly 90% of all customer questions and problems have a recorded solution somewhere, but finding that answer in a timely fashion can be challenging. AI could help quickly cut through the clutter. 

Looking upstream, I think there could be some clear beneficiaries of this technology. Semiconductor companies are the backbone of any digital device, and they’ll likely see a boost as wider AI adoption strengthens demand for microchips. Many parts of the semiconductor ecosystem may benefit, from compute services to networking to memory companies. Similarly, cloud hyperscale services stand to benefit, as much of the processing power used by AI takes place in centralized data centers, far from consumer devices. As AI usage expands, such computing power should become far more important.

Further advances in computing power will likely create new uses for AI.

The explosion in AI applications was enabled by faster and more powerful chips, as well as expanding server farms, but advancement has not plateaued on either front. I think chips and server farms will continue to improve, joined with advances in efficiency — a somewhat neglected lane of development until recently, which I think will be instrumental in bringing down the cost of AI development. 

As an example of a potential efficiency gain, Google and Meta have said that many large language models are likely overtrained, meaning they could use far fewer sources of information — called parameters by developers — without significantly harming accuracy or usability. Fewer inputs would reduce the time and cost of training and, more important, make inferencing, the actual process of responding to questions, dramatically cheaper. Semiconductors should also continue to see innovations. More general purpose and traditional processors, such as GPUs and CPUs from Nvidia, AMD and Intel, are the de facto standard for AI and will likely continue to see demand grow. However, they weren’t necessarily designed for it, and some larger companies are developing and building application-specific processors. For example, Google’s tensor processing units, or TPUs, co-developed with Broadcom, are specially designed for machine learning.

Taken altogether, the barriers for AI models are likely going to fall further. I think it’s plausible that there will be another millionfold increase in computing power by the end of the decade, and that could bring more astounding technologies and compelling investment opportunities.