Entering the Exaflop Era
Just a few days ago, NVIDIA announced their new AI chip, Blackwell. This is the largest graphical processing unit (GPU) chip to date (see photo below, source: NVIDIA).
Leaders from Alphabet (parent of Google), Amazon, Dell, Meta, Microsoft, OpenAI, Oracle, and Tesla all spoke highly of the work NVIDIA is doing for AI and accelerated computing, with most of them specifically citing the new Blackwell chip and systems. These companies, including NVIDIA, represent roughly $11.5 trillion in market value. When leaders representing this much in market value speak up, I think it’s wise for us to pay attention to what they’re saying.
And right now, they seem to be shouting from the rooftops that AI and accelerated computing has arrived and isn’t going anywhere.
A SuperPOD of eight Blackwell systems can output 11.5 exaflops of AI performance. NVIDIA is currently working on building a cloud AI supercomputer with Amazon Web Services (AWS) called Project Ceiba that will combine the processing power of 20,000 GPUs with more than 10,000 special CPUs to produce 414 exaflops.
One exaflop is 1,000,000,000,000,000,000 (one quintillion) floating point operations per second (FLOPS). An average computer from around 2005 could process 1 gigaflop (1 billion FLOPS). By around 2010, an average computer could process 1 teraflop (1 trillion FLOPS).
To put 414 exaflops into context, it would take a 2005 PC around 1.1 billion years to do what the Project Ceiba supercomputer will be able to do in one day. It would take 1.1 million years for a 2010 PC to equal the Project Ceiba performance.
Of course, this is somewhat theoretical. There could be structural things that could impact the math on this.
Still, to think that one day with a supercomputer would have taken more than 1 billion or 1 million years for an older PC is hard to wrap one’s mind around. And these chips, computers, and supercomputers keep getting faster and more efficient. Ten years from now we might see a chip that is 1,000 times faster.
How might this translate into our day-to-day lives? What might the use cases be of being able to crunch that many numbers?
We probably won’t see robots taking over the streets any time soon, but we might see massive improvements in testing, manufacturing design, and medical discovery and advancement. We’re already seeing computer programs writing other computer programs. We’re also number crunching millions of miles driven from autonomous vehicles.
Imagine for a moment all the books that were ever written, laws that were ever created, court findings, genealogical records, scientific journals, or academic research posts that were ever written. What about all of the documentaries, instructional videos or recorded lectures that have ever been given?
Systems can already read, understand, and make sense of much of this information and, due to being trained in multimodal (e.g. multiple forms of media – text, image, audio, video) large language models (LLMs) they might someday be able to provide responses to queries or prompts considering brand new information and do so in virtual real-time. That this information could be factual and without bias would be even more helpful.
This could be useful from everything from fact-checking what a politician says in real-time to analyzing security risks, risks of a cyber-attack, an invasion of a country, anticipating weather events or the probability of a natural disaster, or identifying a malfunction of an electrical grid before it happens. There is likely a plethora of uses for this type of big data analysis that simply wasn’t possible just a few years ago.
Imagine a government that used this data to monitor economic trends, demographic changes, productivity changes, inflation pressures, or what is happening in the labor market and with small businesses and real estate in different regions. Rather than waiting for reports about what happened one to three months ago, what if we had this data in real-time? Would a system like this be useful to warn smaller banks about buying treasury bonds at all-time low yields? Or to advise the Treasury department to issue more longer-dated debt at such low yields to help refinance the national debt while rates were lower and not higher?
Imagine how schools and universities might be able to teach students better based on the student’s different learning styles and based off their individual strengths.
What if there were better ways to detect and limit the flow of harmful drugs? Or to identify human trafficking more easily? Or to identify groups that were vulnerable in other ways? What if we could prevent government waste, fraud, or corruption better?
How about benefits regarding investing?
What if we could calculate the risk, diversification, and potential upside performance of a portfolio with more precision and do it in real-time based on current data around the world. What if we could detect speculation, asset bubbles and other market anomalies with more confidence? Or to identify an investment scam more easily?
What if we could have a fully adaptive financial plan that automatically adjust to changes in wages, costs of living changes, changes in tax rates, social security adjustments, new laws passed regarding retirement benefits and programs, or changes in risk capacity or risk willingness and have it update us in real-time?
We are not there yet but we’re probably getting closer.
I know some of this might sound scary and I certainly have some trepidation about how all of this will work and how the data will be used, who profits from it, and how will we prevent bad actors. But I am also hopeful that these tools will help us to make things more secure, not less secure. They could probably help us discover more truth, not less of it. We might be able to use AI to identify the fake stuff that was also maliciously made by AI.
Despite all the drawbacks, I am far more excited about all the possibilities than I am worried about the potential issues.
It will probably be messy in the beginning. Just like the first light bulb, the first automobile, the first airplane, and yes, the first days of the internet. But we humans are very resilient and very adaptable. We will collectively “work the problem” and refine the system in positive ways to make our lives better.
The future of investment management and financial planning may in the not-too-distant future have far more tools to make life better. While we will likely have massive increases in productivity and some automation, we continue to believe that humans will always be involved in relationships pertaining to our financial lives.
Securities offered through LPL Financial, Member FINRA/SIPC. Investment advice offered through WCG Wealth Advisors, a registered investment advisor. WCG Wealth Advisors and The Wealth Consulting Group are separate entities from LPL Financial.
Jim Worden offers investment advice through WCG Wealth Advisors, LLC a Registered Investment Advisor doing business as The Wealth Consulting Group. Jim is not affiliated with LPL Financial.
The opinions voiced in this material are for general information only and are not intended to provide specific advice or recommendations for any individual.
All performance referenced is historical and is no guarantee for future results. All indices are unmanaged and may not be invested into directly.
