[Firefly/Adobe]
AI spending is certainly ramping up. Microsoft and OpenAI are reportedly mulling building a $100 billion data center project in the coming years. Amazon and Google are also investing hefty sums in generative AI products and partnerships. Some pundits such as former OpenAI employee Leopold Aschenbrenner foresee the emergence of “trillion-dollar clusters” in computing in the coming years, speculating they would require power on par with smaller U.S. states.
Does AI already have a spending problem?
While it remains to be seen whether such massive investments will pay off, there are growing concerns that AI may indeed have a spending problem already. Covello argues that the astronomical price tag of AI demands groundbreaking solutions, not just incremental improvements. What trillion-dollar problem will AI solve? he asks. “Replacing low-wage jobs with tremendously costly technology is basically the polar opposite of the prior technology transitions I’ve witnessed in my thirty years of closely following the tech industry.” In terms of R&D, for example, he’s skeptical that AI, in its current form, can do much more than accelerate existing processes like data analysis. “We’ve found that AI can update historical data in our company models more quickly than doing so manually, but at six times the cost,” he notes, “the value proposition simply isn’t there yet.”
While Covello sees AI’s current capabilities confined to speeding up existing processes like data analysis, other experts foresee a more radical transformation. Avital Balwit, Chief of Staff to the CEO at Anthropic in a piece for Palladium, argues that much white collar work is vulnerable to automation in the coming years.
“Tasks that involve reading, analyzing, and synthesizing information, and then generating content based on it, seem ripe for replacement by language models,” she noted. Balwit predicted a rapid expansion of AI’s proficiency across an array of cognitive tasks, from copywriting and tax preparation to software development and legal analysis.
While not focusing on R&D roles specifically, she points to areas like software development and contract law, both essential to R&D operations, as early examples of AI’s encroaching influence. She predicts AI will soon excel at a much wider range of cognitive tasks.
Where are we now with AI anyway?
Covello pushes back against the common refrain that AI is simply retracing the path of previous tech revolutions, destined to become more affordable over time. He contends that unlike the internet, which offered a low-cost alternative to existing systems from the outset, “AI technology is exceptionally expensive.” He cautions against complacency about cost declines, pointing to Advanced Semiconductor Materials Lithography (ASML) as a notable example. ASML, which supports lithography tools essential for chip manufacturing, has seen its costs increase over time as a result of its dominant market position. “Technology can be so difficult to replicate,” Covello warns, “that no competitors are able to do so, allowing companies to maintain their monopoly and pricing power.”
Gartner’s “Hype Cycle for Artificial Intelligence, 2024” report acknowledges the current surge in AI investment, particularly in generative AI. Yet the tech analyst firm emphasizes that “in most cases, [generative AI] has yet to deliver its anticipated business value.” The report highlights the need for “standardized processes to aid implementation” and suggests that companies should prioritize exploring a range of AI techniques beyond just generative AI.
Regarding AI’s potential to boost revenues for non-tech companies, Covello expresses doubt. He states, “I place low odds on AI-related revenue expansion because I don’t think the technology is, or will likely be, smart enough to make employees smarter.” He believes that even in improving search functionality, genAI is more likely to help employees find information faster rather than find better information.
Has AI hype peaked already?
Other experts at Goldman Sachs strike a more optimistic tone, even as Gartner’s Hype Cycle report cautions that generative AI has “passed the Peak of Inflated Expectations.” Kash Rangan, Senior Equity Research Analyst covering US software, remains enthusiastic about AI’s long-term potential, particularly as tools and best practices mature. He points to early signs of productivity gains in creative design, code development, and customer support. Eric Sheridan, Senior Equity Research Analyst covering US internet, has a similar take. While acknowledging the current “limited visibility into AI applications and adoption rates,” he shares Rangan’s optimism, citing the demonstrations of AI’s capabilities he’s witnessed Examples include the ability of generative AI to produce new design concepts in minutes, a task that previously would have taken hours. This speed, Sheridan believes, points to AI’s potential to accelerate innovation and reshape industries.
The search for AI’s killer app in R&D
While Covello is skeptical that AI can justify its massive investment in the near term, Gartner points to several emerging AI technologies with significant implications for R&D. In its annual AI hype cycle report, the firm describes, for instance, composite AI, which the firm predicts will become the “standard methodology for developing AI systems” within two years. This approach, which involves combining different AI techniques to enhance learning efficiency and broaden knowledge representation, could be particularly beneficial for R&D in that it could support the creation of more robust and adaptable AI systems capable of tackling complex research problems.
Another notable technology Gartner highlights is the use of knowledge graphs. These machine-readable representations of the physical and digital worlds, capturing entities and their relationships, offer “dependable logic and explainable reasoning,” it notes. This capability would address a key concern about deep learning systems’ lack of transparency. For R&D, knowledge graphs could prove their worth by organizing and analyzing vast amounts of research data, facilitating new discoveries and insights by revealing hidden connections and patterns.
Finally, Gartner emphasizes the growing importance of AI-based simulation, which could lead to AI agents and the simulated environments in which they can be trained, tested, and deployed. In terms of R&D, such technology would enable researchers to experiment with new ideas, test hypotheses in a controlled environment, and accelerate the development of AI-powered technologies for an array of research applications. AI simulation can also sidestep the challenges of data scarcity in R&D, allowing researchers to generate synthetic, yet potentially accurate, data for training AI models.
Balancing the AI ROI teeter totter
Even as AI spending continues to surge, some experts remain skeptical about its near-term revenue impact. Covello remains dubious about AI’s ability to meaningfully boost revenue in the near term. He argues that even if AI enhances efficiency, these gains would likely be “arbitraged away” as competitors adopt similar technologies. “If a company can use a robot to improve efficiency,” he points out, “so can the company’s competitors. So, a company won’t be able to charge more or increase margins.”
Yet other Goldman Sachs experts, such as Kash Rangan and Eric Sheridan, take a longer view. They draw parallels to past tech revolutions, arguing that AI, like the internet and smartphones before it, may initially appear overhyped and expensive but could eventually support new applications and revenue streams. As Rangan puts it, “Nobody today can say what killer applications will emerge from AI technology. But we should be open to the very real possibility that AI’s cost equation will change, leading to the development of applications that we can’t yet imagine.”
This long-term optimism extends beyond purely financial returns. Avital Balwit, in Palladium, envisions a future where AI’s impact on work transcends traditional notions of employment and productivity. She suggests that AI could fundamentally alter our relationship with work, potentially freeing us from the “psychological burdens of shame and duty” linked to traditional jobs.
No matter how things play out, R&D professionals will be instrumental in determining which AI innovations deliver tangible value, separating hype from genuine breakthroughs, and translating research into practical tech.
I support the overarching goals of the AI revolution, but I see a more urgent and pernicious issue at hand. We need AI systems to develop smart countermeasures to combat bad actors—like fake AI videos, autonomous phone-bots, and other malicious applications—that threaten to turn the internet into a polluted, infected, and dying swamp of misinformation and abuse. Without swift action, the web risks becoming a cesspool with no real value beyond email and prurient interests. The time for intelligent AI defenses is now, before it’s too late. But I see the problem, who shall pay to keep us clean!!!
Completely agree that the amount of “noise” on the internet is likely to swell and as you note, misinformation as well. What you describe has already been a theme in the cybersecurity space, too, where you see more ML on both the attackers’ and defenders’ sides.
There is also the question of who would be responsible for the AI systems used to combat bad actors? So far it’s mostly tech and cyber companies focusing on the issue from what I’ve seen.