We have been proudly investing the Canada Pension Plan Fund for 25 years, helping ensure that more than 22 million Canadians have financial security in retirement.
Artificial intelligence (AI) has been advancing rapidly for more than 60 years, though for most of that time, it’s largely operated “behind the scenes.” Unbeknownst to many, over the last decade it has increasingly powered the search, e-commerce, social media, navigation and online video applications that billions of users engage with every day.
All that changed in November 2022 with the launch of OpenAI’s ChatGPT and later, Google Bard and Microsoft Copilot. Together, these tools helped thrust AI into the mainstream as hundreds of millions of users, entrepreneurs and enterprises began to interact with it in natural human language. What followed was an explosion of both investment activity and human curiosity, as the world contemplated AI’s potential as an added driver of innovation and productivity.
What lies ahead? The underlying technical building blocks that enable Generative AI (Gen AI) will likely continue to advance at an accelerated rate. These advancements will be driven by ongoing innovation in computing systems and large language foundation models. And they will, in turn, broaden the spectrum of models and Gen AI applications available to consumers, developers and enterprises. Computing technology and computing infrastructure will continue to scale along with the most powerful cutting-edge AI models. And experimentation will push on as adoption accelerates and users and enterprises continue to imagine the art of possible.
Along the way, key decisions will have to be made about what data can and should be securely provided to AI models and applications, how AI can be deployed responsibly and how we can ensure all stakeholders, including investors, prepare for the rapidly shifting landscape that awaits.
Here are five key AI trends to watch over the next 12 to 18 months:
1. Gen AI will continue to grow more powerful and cheaper to deploy
Over the past 15 years, the cost of training AI models has been on the downswing. The training cost per FLOP (a pivotal metric of computational capacity and efficiency) declined by more than 95% over the period—as observed through Epoch AI data1 —and continues to be on an exponential cost decline trajectory. This has made it easier to train even more powerful and larger AI models, setting the stage for a new phase of scaling that won’t be limited to deep learning models. It will also expand the spectrum of more efficient models fine-tuned to perform specific tasks, like summarizing, coding, and enabling customer service Chatbots, Copilots and AI agents. Looking ahead, we estimate the scale of investment in AI computing infrastructure is set to continue surging from approximately US$52 billion in 2023 to over US$100 billion in 2024. This will be driven by cloud service providers, enterprises, and sovereign nations.
As investment rises and costs decline, expect a proliferation of new applications and AI agents powered by a variety of AI models. These will pave the way for Gen AI adoption across the consumer and enterprise landscape. And they’ll further stimulate investment in AI compute infrastructure, including the data centres and cloud computing infrastructure that enable the training and deployment of Gen AI applications and AI agents.
GPU computing power of AI frontier models is growing on an exponential trajectory
The cost to train AI models is declining at an exponential rate
GPU computing power of AI frontier models is growing on an exponential trajectory
The cost to train AI models is declining at an exponential rate
2. Organizations will need to ensure their data is ready for AI
As the AI ecosystem continues to evolve, generalized models trained solely on web-based data should become more widely available. But while these models serve as useful foundations, they don’t unlock the full benefit of Gen AI. Enterprises, governments, and other organizations have an opportunity to leverage their own domain-specific and proprietary data to customize models in a way that generates more relevant, productive and differentiated insights for their specific needs.
To put these models to work, enterprises will first need to take a hard look at their data footprints. That means building data pipelines that provide structured data sets to feed into Gen AI models. As natural language interactions help to inform future training sets, the role of unstructured data (like text files, websites, audio, and images) will also become more important to training.
To make it all happen, enterprises may need to reevaluate their data vendors, re-examine their data governance frameworks to ensure proper access rights, and establish protective mechanisms to prevent leaks of proprietary data. In addition, enterprises and model providers must work together to mitigate potential biases, and ensure the responses are accurate and relevant. For organizations, meeting all of these needs may seem like a daunting task—but it’s a pivotal one.
The elements are in place for growth in AI adoption
Investments in accelerated computing data centre infrastructure are expected to grow
Software infused with Gen AI features and capabilities is likely to become a bigger part of software spending
As data continues to grow, led by unstructured data, AI will unlock the ability to extract more insights
Investments in accelerated computing data centre infrastructure are expected to grow
Software infused with Gen AI features and capabilities is likely to become a bigger part of software spending
As data continues to grow, led by unstructured data, AI will unlock the ability to extract more insights
3. Gen AI will be deployed in more industries and business processes
Early experiments with Gen AI have shown promising signs of enhancing productivity and efficiency. A joint study by BCG and Harvard found that Gen AI tools empowered consultants to complete 12% more creative tasks, clock 25% faster turnaround times, and boost quality of output by 40%. Initial ROIs are also compelling, as a joint study by Microsoft and IDC showed a return of $3.50 for every $1 invested in Gen AI, and an average payback on investment of just 14 months.
As AI continues to permeate applications and shape the way enterprises use their data, we expect companies to embrace one or both of the following approaches:
The AI-Enabled Approach: Some will pursue a stepwise method, using Gen AI-enabled applications from existing vendors to augment existing processes. These tools can be made more powerful by infusing proprietary and/or domain-specific data to generate more insightful feedback and cut the time needed to complete tasks.
The AI-First Approach: This approach is more transformative, with workflows entirely reimagined or built from scratch to incorporate AI tools. This may involve training bespoke models for specific tasks, such as creating copilots for doctors that assist with routine documentation.
These pathways could create a “Virtuous Cycle” or self-reinforcing system for AI deployment. Workflows will first be improved by data generated from Human-AI interactions. But over time, AI agents (apps that can make choices and perform tasks on their own after user instructions) could interact directly with the models, and even with each other, automating more parts of processes. Data generated from interactions with both humans and AI agents can then be fed back to further improve the models.
The paradigm shift in how we interact with AI may result in potential disruptions to existing value chains. As a result, organizations need to evaluate appropriate guidelines for these interactions, for which vendors to use, and for when to include humans (or insert “humans-in-the-loop.”)
A virtuous cycle is emerging across an evolving AI value chain
4. Key stakeholders will need to ensure AI is deployed responsibly
Looking forward, we anticipate the continued evolution of AI regulation worldwide—though countries will likely move at different speeds. Data sovereignty could become a bigger concern as nations look to keep sensitive proprietary data onshore. Countries may also collaborate with their national ecosystem of AI startups and companies to boost the supply of critical resources like computational power. Enterprises must determine how best to protect their proprietary data from leaks and how to manage cybersecurity risks. They also must evaluate how to responsibly deploy AI that is honest, helpful, and harmless to users.
Governments will need to balance this goal of fostering innovation with a desire to mitigate risks through regulation. We’ll be watching how the broader ecosystem—including governments, technology providers, non-technology companies and consumers—collaborates to ensure risks and responsibilities are balanced across stakeholders.
We may also see a level of self-policing in the ecosystem. Model providers such as OpenAI, Anthropic, Microsoft, and Google each have outlined frameworks to prevent models from producing undesired outcomes. Organizations outside of the technology industry are also developing their own internal processes to ensure safe outcomes for employees and customers as models are put into real-world use cases.
5. For all players, a mindset of constant learning will be more important than ever
Amid rapid advances in Gen AI, certain skills may diminish in value or even be replaced over time by AI assistants. By contrast, the ability to augment knowledge work with Gen AI and amplify human ingenuity will become more critical in the future. As such, a continuous and virtuous cycle of upskilling and evolving workflow will only grow in importance. By extension, attracting and retaining employees with a continuous learning mindset and the ability to adapt and evolve will be essential.
We expect enterprises and employees to take on different responsibilities as workflows and roles evolve. Enterprises will need to invest in AI technologies. And they’ll need to empower their employees to experiment with ways to integrate AI into their work and to strive to increase innovation. They can do this through tutorials and forums that share learnings.
Users, knowledge workers and business leaders will need to find ways to embrace these tools. Their abilities to learn, adapt, and think outside the box will be tested as online consumer applications and knowledge work increasingly incorporate Gen AI and AI agents.
Indeed, the impact of Gen AI will be felt not just by technology companies, but by a range of industries and sectors. As such, investors will also need to understand how Gen AI disrupts existing value chains. They’ll need to be on the lookout for the emergence of new business models and investment opportunities—as well as for potential threats to existing business models. In addition to the key enablers of Gen AI, new cohorts of companies may emerge, including those that generate new revenue streams by infusing products and services with Gen AI, those that manage to extend the durability of existing revenue streams and those that see existing revenue streams erode.
Investors should try to understand the key inputs and drivers of Gen AI adoption. They should understand the investments in technology and talent required for companies to integrate and embed Gen AI into operations at scale. And they should attempt to measure the long-term ROI to justify the level and pace of ongoing investments.
The potential productivity that companies may achieve by adopting Gen AI will continue to be a focus for investors. However, they should also try to assess the long-term sustainability of benefits from Gen AI adoption, identifying when the benefit is likely to be competed away versus when productivity growth could improve competitive advantage over the long-term.
For all the promise that artificial intelligence and other rapidly-evolving technologies offer the world, the downside risks can't be
Article
March 3, 2023
Artificial intelligence (AI) has been advancing rapidly for more than 60 years, though for most of that time, it’s largely operated “behind the scenes.” Unbeknownst to many, over the last decade it has increasingly powered the search, e-commerce, social media, navigation and online video applications that billions of users engage with every day. All that changed in November 2022 with the launch of OpenAI’s ChatGPT and later, Google Bard and Microsoft Copilot. Together, these tools helped thrust AI into the mainstream as hundreds of millions of users, entrepreneurs and enterprises began to interact with it in natural human language. What followed was an explosion of both investment activity and human curiosity, as the world contemplated AI’s potential as an added driver of innovation and productivity. What lies ahead? The underlying technical building blocks that enable Generative AI (Gen AI) will likely continue to advance at an accelerated rate. These advancements will be driven by ongoing innovation in computing systems and large language foundation models. And they will, in turn, broaden the spectrum of models and Gen AI applications available to consumers, developers and enterprises. Computing technology and computing infrastructure will continue to scale along with the most powerful cutting-edge AI models. And experimentation will push on as adoption accelerates and users and enterprises continue to imagine the art of possible. Along the way, key decisions will have to be made about what data can and should be securely provided to AI models and applications, how AI can be deployed responsibly and how we can ensure all stakeholders, including investors, prepare for the rapidly shifting landscape that awaits. Here are five key AI trends to watch over the next 12 to 18 months: 1. Gen AI will continue to grow more powerful and cheaper to deploy Over the past 15 years, the cost of training AI models has been on the downswing. The training cost per FLOP (a pivotal metric of computational capacity and efficiency) declined by more than 95% over the period—as observed through Epoch AI data1 —and continues to be on an exponential cost decline trajectory. This has made it easier to train even more powerful and larger AI models, setting the stage for a new phase of scaling that won’t be limited to deep learning models. It will also expand the spectrum of more efficient models fine-tuned to perform specific tasks, like summarizing, coding, and enabling customer service Chatbots, Copilots and AI agents. Looking ahead, we estimate the scale of investment in AI computing infrastructure is set to continue surging from approximately US$52 billion in 2023 to over US$100 billion in 2024. This will be driven by cloud service providers, enterprises, and sovereign nations. As investment rises and costs decline, expect a proliferation of new applications and AI agents powered by a variety of AI models. These will pave the way for Gen AI adoption across the consumer and enterprise landscape. And they’ll further stimulate investment in AI compute infrastructure, including the data centres and cloud computing infrastructure that enable the training and deployment of Gen AI applications and AI agents. GPU computing power of AI frontier models is growing on an exponential trajectory The cost to train AI models is declining at an exponential rate GPU computing power of AI frontier models is growing on an exponential trajectory The cost to train AI models is declining at an exponential rate 2. Organizations will need to ensure their data is ready for AI As the AI ecosystem continues to evolve, generalized models trained solely on web-based data should become more widely available. But while these models serve as useful foundations, they don’t unlock the full benefit of Gen AI. Enterprises, governments, and other organizations have an opportunity to leverage their own domain-specific and proprietary data to customize models in a way that generates more relevant, productive and differentiated insights for their specific needs. To put these models to work, enterprises will first need to take a hard look at their data footprints. That means building data pipelines that provide structured data sets to feed into Gen AI models. As natural language interactions help to inform future training sets, the role of unstructured data (like text files, websites, audio, and images) will also become more important to training. To make it all happen, enterprises may need to reevaluate their data vendors, re-examine their data governance frameworks to ensure proper access rights, and establish protective mechanisms to prevent leaks of proprietary data. In addition, enterprises and model providers must work together to mitigate potential biases, and ensure the responses are accurate and relevant. For organizations, meeting all of these needs may seem like a daunting task—but it’s a pivotal one. The elements are in place for growth in AI adoption Investments in accelerated computing data centre infrastructure are expected to grow Software infused with Gen AI features and capabilities is likely to become a bigger part of software spending As data continues to grow, led by unstructured data, AI will unlock the ability to extract more insights Investments in accelerated computing data centre infrastructure are expected to grow Software infused with Gen AI features and capabilities is likely to become a bigger part of software spending As data continues to grow, led by unstructured data, AI will unlock the ability to extract more insights 3. Gen AI will be deployed in more industries and business processes Early experiments with Gen AI have shown promising signs of enhancing productivity and efficiency. A joint study by BCG and Harvard found that Gen AI tools empowered consultants to complete 12% more creative tasks, clock 25% faster turnaround times, and boost quality of output by 40%. Initial ROIs are also compelling, as a joint study by Microsoft and IDC showed a return of $3.50 for every $1 invested in Gen AI, and an average payback on investment of just 14 months. As AI continues to permeate applications and shape the way enterprises use their data, we expect companies to embrace one or both of the following approaches: The AI-Enabled Approach: Some will pursue a stepwise method, using Gen AI-enabled applications from existing vendors to augment existing processes. These tools can be made more powerful by infusing proprietary and/or domain-specific data to generate more insightful feedback and cut the time needed to complete tasks. The AI-First Approach: This approach is more transformative, with workflows entirely reimagined or built from scratch to incorporate AI tools. This may involve training bespoke models for specific tasks, such as creating copilots for doctors that assist with routine documentation. These pathways could create a “Virtuous Cycle” or self-reinforcing system for AI deployment. Workflows will first be improved by data generated from Human-AI interactions. But over time, AI agents (apps that can make choices and perform tasks on their own after user instructions) could interact directly with the models, and even with each other, automating more parts of processes. Data generated from interactions with both humans and AI agents can then be fed back to further improve the models. The paradigm shift in how we interact with AI may result in potential disruptions to existing value chains. As a result, organizations need to evaluate appropriate guidelines for these interactions, for which vendors to use, and for when to include humans (or insert “humans-in-the-loop.”) A virtuous cycle is emerging across an evolving AI value chain 4. Key stakeholders will need to ensure AI is deployed responsibly Looking forward, we anticipate the continued evolution of AI regulation worldwide—though countries will likely move at different speeds. Data sovereignty could become a bigger concern as nations look to keep sensitive proprietary data onshore. Countries may also collaborate with their national ecosystem of AI startups and companies to boost the supply of critical resources like computational power. Enterprises must determine how best to protect their proprietary data from leaks and how to manage cybersecurity risks. They also must evaluate how to responsibly deploy AI that is honest, helpful, and harmless to users. Governments will need to balance this goal of fostering innovation with a desire to mitigate risks through regulation. We’ll be watching how the broader ecosystem—including governments, technology providers, non-technology companies and consumers—collaborates to ensure risks and responsibilities are balanced across stakeholders. We may also see a level of self-policing in the ecosystem. Model providers such as OpenAI, Anthropic, Microsoft, and Google each have outlined frameworks to prevent models from producing undesired outcomes. Organizations outside of the technology industry are also developing their own internal processes to ensure safe outcomes for employees and customers as models are put into real-world use cases. 5. For all players, a mindset of constant learning will be more important than ever Amid rapid advances in Gen AI, certain skills may diminish in value or even be replaced over time by AI assistants. By contrast, the ability to augment knowledge work with Gen AI and amplify human ingenuity will become more critical in the future. As such, a continuous and virtuous cycle of upskilling and evolving workflow will only grow in importance. By extension, attracting and retaining employees with a continuous learning mindset and the ability to adapt and evolve will be essential. We expect enterprises and employees to take on different responsibilities as workflows and roles evolve. Enterprises will need to invest in AI technologies. And they’ll need to empower their employees to experiment with ways to integrate AI into their work and to strive to increase innovation. They can do this through tutorials and forums that share learnings. Users, knowledge workers and business leaders will need to find ways to embrace these tools. Their abilities to learn, adapt, and think outside the box will be tested as online consumer applications and knowledge work increasingly incorporate Gen AI and AI agents. Indeed, the impact of Gen AI will be felt not just by technology companies, but by a range of industries and sectors. As such, investors will also need to understand how Gen AI disrupts existing value chains. They’ll need to be on the lookout for the emergence of new business models and investment opportunities—as well as for potential threats to existing business models. In addition to the key enablers of Gen AI, new cohorts of companies may emerge, including those that generate new revenue streams by infusing products and services with Gen AI, those that manage to extend the durability of existing revenue streams and those that see existing revenue streams erode. Investors should try to understand the key inputs and drivers of Gen AI adoption. They should understand the investments in technology and talent required for companies to integrate and embed Gen AI into operations at scale. And they should attempt to measure the long-term ROI to justify the level and pace of ongoing investments. The potential productivity that companies may achieve by adopting Gen AI will continue to be a focus for investors. However, they should also try to assess the long-term sustainability of benefits from Gen AI adoption, identifying when the benefit is likely to be competed away versus when productivity growth could improve competitive advantage over the long-term. 1Trends in the Dollar Training Cost of Machine Learning Systems – Epoch (epochai.org) Author Nadeem Janmohamed Managing Director, Active Equities North America Contributors Matt Kleffman Senior Associate, Active Equities North America Tianquan Wang Senior Associate, Active Equities North America Visualizations by Voilà. Thanks for subscribing to CPP Investments Sign up for our latest news, insights, reports and other information about CPP Investments Email address * Please enter valid email id Job title Select Job Title Associate Analyst Consultant advisor Manager/supervisor Government official/regulator General manager/director Board director Chairman/board member VP/SVP/EVP President Partner/Owner/Entrepreneur Parent/guardian C-level other Chief Human Resources Officer Chief Marketing Officer Chief Financial Officer Chief Sustainability Officer Chief Digital Officer Chief Technology Officer Chief Operating Officer Educator/professor Student Editor/reporter Other Organization How did you hear about CPP Investments? Select Source CPP Investments alumni CPP Investments employee or board member CPP Investments portfolio company Online search (e.g. Google) Social media Other What news would you like to receive? * News and updates from CPP Investments Latest news from the Insights Institute Consent * By checking here, you are subscribing to receive our newsletters and other similar types of insights and reports by email, and permit CPP Investments to use cookies and similar technologies to analyze your interactions with our emails. Unsubscribe at any time by clicking the link in the newsletter’s footer. Visit our Privacy Policy for more information. Questions or concerns? Contact us. An error has occurred. Please try again later. Next Investing in the age of artificial intelligence Chris Gillam, Director, Operational Risk and Nadeem Janmohamed, Managing Director, Active Equities, share their views. Article June 9, 2023 Views on Artificial Intelligence (AI) with Jordan Jacobs, Co-Founder & As AI changes the future of work, what’s the most important skill that schools need to teach? Hear Jordan’s take. Video June 9, 2023 Q&A with our Head of Strategy Execution & Relationship Management For all the promise that artificial intelligence and other rapidly-evolving technologies offer the world, the downside risks can't be Article March 3, 2023
