One of the more intriguing conversations as we head into the back half of the year was a piece published earlier this week by Sequoia Capital, which posited the $600 billion AI challenge.
In short, it advocates the sell-in of infrastructure to lead to economic output far beyond where we are now. In other words, we’re on a long and winding road in the AI narrative that goes far beyond the all-at-once payoff storyline involving NVIDIA Corp. and no one else.
As we trace the rollicking ride of AI from chips to consumer experiences, nearly every category in the tech industry’s ecosystem will be touched and radically reshaped.
The golden road to riches started with NVIDIA, which has accrued a staggering market value in a short period of time. But it hardly ends there. A wholesale digital transformation is taking place in what I like to divide into three major stages. This is what I call the multi-tier networks effects of AI beyond the GPU and chip manufacturing (i.e., Taiwan Semiconductor Manufacturing Company), extending from inward consumption of infrastructure to sell-out access to compute and software for AI.
The first network effect entails servers, memory, network, and storage from the likes of Dell Inc., Micron, Marvell Technology Inc., Broadcom Inc., and others. These companies are providing infrastructure for the enablers. (For devices, these include OEMs, but a different set of chips provided to device makers and will add Qualcomm Inc. and Intel Corp. to Advanced Micro Devices Inc. and NVIDIA.)
The second network effect shifts to platform providers and independent software vendors that encompass hyperscalers and SaaS such as Amazon.com Inc., Alphabet Inc.’s Google, Microsoft Corp., Salesforce Inc., ServiceNow Inc., Oracle Corp., OpenAI, SAP, and more. These players are building infrastructure for industries.
The third network effect entails industries that are applying AI to create new and improved experiences from retail to financial services. This is where the heaviest consumption of the infrastructure will take place, and where the most economic value will occur when we discuss the multitrillion-dollar impact of AI by the end of the decade.
Indeed, concerns over the sell-off of AI and the consumption that needs to take place across industries isn’t hard to see. Where are AI and Generative AI being used by banks, hotels, restaurants, manufacturing and other industries—and which companies are making money from these AI powered advancements?
This is what Sequoia meant when it called out the $600 billion AI conundrum. Then again, most of this alarmist content may have more to do with AI’s time horizon than it does with the long-term viability of AI’s buildout.
Amid that recent wave of VC and equity research notes from Goldman Sachs, Sequoia, a16z, one specific example from Goldman stood out: The ability of AI to outperform humans in certain equities research and model building. But Jim Covello, head of Goldman Sachs Equity Research, noted the application had a cost of about six times human capital.
However, most of AI’s buildout equating to more than 50% of the capex investment comes from a small number of hyperscale cloud providers and massive enterprises that are using AI to power their businesses. And the GPUs aren’t exclusively for genAI but for model building, training, and deployment for recommendation engines, video rendering, content filtering, and more.
These are many of the “old AI” applications and they will continue to scale. These same companies, which have benefitted from AI for a long time, continue to pour investment into leverage AI and democratize access to expensive AI infrastructure that will be required to innovate across industries. Still, these applications remain more nascent and will likely take a few years to become widely deployed and fully measurable.
In the meantime, this waterfall of AI spend that starts with power, thermals, design, silicon and manufacturing flows to systems including compute, storage and networking. Following that, it flows to OEMs/ODMs, and then to ISVs and consultants that support implementation.
The speed of AI’s proliferation will take months and years, not days and weeks. But the claim in the Goldman note that suggests there are zero killer genAI applications is flat-out wrong. Use cases for advertising, search, content creation, video development, rendering, and even private models that can streamline financial services or personalize health care are moving quickly and – unlike the dot-com boom – these apps have legs and will spur market growth.
To be clear, I’m not saying there isn’t reasonable consternation on the time horizons and the needed returns from hundreds of billions of dollars in capex.
But this technology isn’t cyclical. It is transformative that will change every industry. Some stocks may have run ahead of the market and others may lag in real AI-driven value.
In the long run, companies and investors that see the impact of AI over the long horizon know that this is, at worst, a short-term, front-loaded capex spend for what is all but certain to demarcate the winners and losing businesses of an AI-powered future.
