Steve Forbes discusses the big-spending Los Angeles Dodgers and whether they’ve replaced the New York Yankees as Major League Baseball’s “Evil Empire.”
Years ago, when the New York Yankees with their huge payroll dominated baseball, a Boston Red Sox executive dubbed the Bronx Bombers “the Evil Empire.”
A growing number of baseball aficionados are applying this epithet to the Los Angeles Dodgers, who humiliated the Yankees in the World Series last fall. Since then, the Dodgers’ already-large payroll didn’t stop the team from splurging hundreds of millions of dollars on several new players.
“Foul!” cry the critics.
“Boo-hoo!” responds this episode of What’s Ahead, which makes the case that upon examination the concerns are strikeouts.
In the last two years, most new developments in generative AI have resulted in applause and admiration from the tech-interested community in the U.S. Industry leaders have noted the achievements of others—sometimes rushing to copy them—while investors and pundits generally celebrate the growing ability of the technology as a whole.
When Chinese AI chatbot DeepSeek appeared last week, the polite admiration from tech leaders was there, but nothing else that was positive. The apparently self-funded platform that reportedly cost just $5.6 million to train, and which was likely developed with less powerful chips than those behind U.S.-based models, made many U.S. tech investors pause and pull back, weighing down the stock market as a whole. DeepSeek, which outperformed some OpenAI models, showed the U.S. that a Chinese tech company could become a frontrunner in the race to develop generative AI platforms. Investor and billionaire Marc Andreessen wrote on X that DeepSeek’s model is “AI’s Sputnik moment,” Forbes’ Rashi Shrivastava and Richard Nieva wrote. And the speedy, explosive success of DeepSeek quickly catapulted founder Liang Wenfeng to billionaire status, writes Forbes’ Giocomo Tognini and Phoebe Liu.
Beyond the financial panic, there are deeper issues for AI technology. How did a relatively unknown Chinese company get to this point so quickly? Several analysts and experts have pointed out that DeepSeek didn’t exactly start from nothing, and probably spent a lot more on equipment than they are claiming. Prior to DeepSeek, Liang founded and led quantitative trading hedge fund High-Flyer Capital Management, through which he spent millions amassing Nvidia GPUs to improve that business. President Donald Trump said Monday that DeepSeek was a “positive development” for technology, but is a “wake up call” for the U.S., and he is considering additional restrictions on Nvidia products that can be sold in China, Bloomberg reported. OpenAI and Microsoft are reportedly investigating whether DeepSeek’s strong start came from tapping into OpenAI’s data through a process called “distillation”—using outputs from a larger, more experienced AI model to train a smaller one. OpenAI has an API to allow other businesses and platforms to tap into its model, but using it for distillation is prohibited.
But not only is there the possibility that DeepSeek is potentially improperly using U.S. technology, there are also national security issues. After all, the U.S. has been clamping down on high-tech exports to China in a move to retain national dominance in AI. There’s also the issue of data. Forbes’ Thomas Brewster writes that DeepSeek is based in the Chinese city of Hangzhou, and its privacy policy details that user data—including keystroke patterns and IP addresses—is stored on servers there. The problem of Americans’ data stored on Chinese servers—and potentially readily available to the Chinese government—was the impetus behind the TikTok ban-or-sale bill, which is awaiting implementation. Brewster also wrote in this week’s issue of The Wiretap newsletter that researchers from cyber intelligence firm Kela were able to make DeepSeek “turn evil,” easily bypassing its safeguards to make it create malicious code that steals credit card numbers. DeepSeek will also suggest that users buy stolen data and provide tips on money laundering.
So another gauntlet in the AI space has been thrown by DeepSeek, but many are treating this one like a grenade. While it does provide something to learn about the cost to develop an AI system, it’s left an unspoken challenge to the U.S. to do better and for less. There’s no doubt Silicon Valley is already working on it.
Boston Consulting Group took a look at how companies across the globe are using AI, and found that one in three plans to invest at least $25 million in upgrading technology this year. However, only about 25% are already seeing ROI. I talked to BCG Global Leader of Tech and Digital Advantage Vladimir Lukic about the reasons behind that. An excerpt from our conversation is later in this newsletter.
Several Big Tech companies are delivering their first earnings reports of the year, and unsurprisingly, much of their reception with investors is depending on their performance in the AI space. Microsoft, which reported its earnings after markets closed Wednesday, saw growth across the board, with total revenues of $69.6 billion, a year-over-year increase of 12%. Microsoft’s cloud business pulled in $40.9 billion in revenue, up 21% from last year, but investors balked because that was on the low side of expectations. As trading opened Thursday, Microsoft stock was down more than 5%. CFO Amy Hood and CEO Satya Nadella said on the earnings call that Microsoft’s growth in this area is capacity constrained. Microsoft has committed to spending $80 billion in infrastructure improvements to ease those constraints in 2025 alone, and Hood and Nadella said more growth can be expected in future quarters.
Meta, which also reported earnings after markets closed on Wednesday, posted a 21% year-over-year bump in revenue, and 49% increase in net income. The company reported increases in traffic, advertising revenue and e-commerce throughout its social apps, as well as hitting the milestone of 700 million monthly users of its Meta AI platform. Investors reacted positively, as Meta’s stock went up nearly 4% on Thursday morning. Meta CEO Mark Zuckerberg had already announced more developments on the AI front last week, saying the company plans to invest between $60 million and $65 million on AI. Reports on Wednesday also indicated another large forthcoming payment from Meta: A $25 million settlement to end President Donald Trump’s 2021 lawsuit accusing the social media company of wrongfully censoring him after it suspended his Facebook and Instagram accounts in the wake of the January 6 Capitol insurrection.
Meta, which also reported earnings after markets closed on Wednesday, posted a 21% year-over-year bump in revenue, and 49% increase in net income. The company reported increases in traffic, advertising revenue and e-commerce throughout its social apps, as well as hitting the milestone of 700 million monthly users of its Meta AI platform. Investors reacted positively, as Meta’s stock went up nearly 4% on Thursday morning. Meta CEO Mark Zuckerberg had already announced more developments on the AI front last week, saying the company plans to invest between $60 million and $65 million on AI. Reports on Wednesday also indicated another large forthcoming payment from Meta: A $25 million settlement to end President Donald Trump’s 2021 lawsuit accusing the social media company of wrongfully censoring him after it suspended his Facebook and Instagram accounts in the wake of the January 6 Capitol insurrection.
The “Doomsday Clock,” a curated measurement of how close human civilization may be to total collapse, ticked one second closer to midnight, the Bulletin of the Atomic Scientists announced this week. According to this measure, it’s 89 seconds to “midnight,” the closest the world has ever been midnight. One of the new risks that moved the minute hand: AI. The technology, the scientists write, can be used for harm, both in weapons of war and “processes of information corruption. AI-enabled distortion of the information environment may be an important factor in preventing the world from dealing effectively with urgent major threats like nuclear war, pandemics, and climate change.” The scientists call out former President Joe Biden’s AI executive order as a positive initial step to “seize the promise and manage the risks of AI,” but the order was rescinded on Trump’s first day in office.
BITS + BYTES
A recent study from Boston Consulting Group examined how companies were integrating AI into their workplaces, and found that 75% of executives rank AI as a top-three priority, and one in three is planning to spend more than $25 million on the technology. However, only about a quarter of companies are seeing ROI from the investments they’ve made so far. I talked to Vladimir Lukic, BCG’s global leader of tech and digital advantage, about why more companies aren’t seeing return just yet. This conversation has been edited for length, clarity and continuity.
In the study, you found that only about 25% of companies are seeing ROI right now with their AI. Of the remaining 75%, you mentioned that about 25% don’t have a business case for moving quickly to implement AI. How do the others feel?
Lukic: They are [saying]: Listen, I have a data science team, and I have a data engineering team, and I have a center of excellence for automation, and I have the IT team—and yet they’re not unlocking the value. They’re extremely frustrated. And there are three reasons for why they can’t break through it.
One is many executives, the last time they were in school was 30 years ago. A lot of these technologies did not exist. There was no iPhone at the time. There was for sure not AI at scale deployed at people’s fingertips. They don’t know how to anticipate these things. It’s not in the DNA of how they deploy this into the workflow. Two, as they were middle managers and going through, they didn’t need to rethink the processes around these types of technologies, so the muscle was really never built. That’s one set of reasons why they can’t do it.
The other one is: A lot of the companies have been used to 2%, 3% continuous improvement on an annual basis on efficiency and productivity, et cetera. We’re talking now 20%, 30%, 40%, 50% changes in productivity and efficiency. It’s an order-of-magnitude change. It does require a very different setup and governance and ability to move on these things.
This is where they break the teeth, and this is the 10-20-70 [approach to AI deployment] that we talk about. Algorithms and all that is 10%. Twenty percent of it is changing the piping and the plumbing and making sure that the right data is in the right place at the right time. Seventy percent is changing the workflows. That muscle is gone.
The last thing [that makes] it hard for them and why it’s frustrating: I have a data science team. I’ve given them these new tools with gen AI, and a mandate to drive it into the organization. But some of these new tools don’t require a data scientist. They require a business owner to deploy. The data scientist puts a threshold on some of these new tools, [requiring they must be at] 99% accuracy when the tools are at 70%, 80% accuracy. And they’re like, we cannot deploy it. You ask them: What is the human accuracy? Well, 40%, 50%. They don’t have the mandate from the organization to drive the right human conversations on the ground. It’s frustrating as hell.
What are the 25% who are seeing ROI doing right?
There is the base knowledge of the technologies at a higher level. When I sit with the executives in that bucket, they don’t talk about technology at all. The conversations we have are: What are the frictions in the value chain? Where can we add value to the customers? If we did something in a different way on any of those points of friction, how would that give us a competitive edge, et cetera. When I talk to the companies that are in the 50% bucket, it is: Is this real? How does it really work? How fast is it going to come? Do we really need to do something? What’s the payback on it? Et cetera. They’re trying to wrap their head around the technicalities of it, where the folks in [the] 25% know it, understand what it is, and therefore they can focus really on the business questions.
Two, they focus on fewer use cases where they deploy, but they go deeper and they think about it more as a workflow. When they invest money in it, they stick with it for a longer period of time, and they dedicate business people to drive those changes on the ground. The ones that are struggling go for many more use cases. They run them in shorter pilots and they never actually get to make it to stick, and they don’t force themselves to prioritize.
The third one is they focus on the core processes. They’re very ruthless around where the value is. Usually when you do the math, value is in your core workflow. So for insurance companies, that’s the underwriting and claims processing. For software companies, it’s software engineering. They don’t shy away from going after the guts of the business.
And then the last point: They are really fast at rethinking the workflow end-to-end, and then changing the incentives of everyone to make sure they do it the right way.
The study also shows 60% aren’t tracking the right metrics to determine ROI. How are they missing this?
These are smart people that are running successful companies that have good intent, so it’s not incompetence and it’s not people just being ignorant of it. Many times, companies ask the wrong people to own some of these initiatives and they sit in a silo in the organization without the position to actually influence the outcomes.
Data scientists are asked to deploy gen AI. They usually report four or five layers into the CIO organization. They build a tool, [and] it takes them longer to build it because they want to get it to a level of precision that might not be needed. Once they get it, they say, ‘IT organization, take it.’
I am now a salesperson in a call center and I have a tool that can help me do things faster. I’m not using it. Why? Because my quota is to do different actions. To start [getting the AI tool used], I need to change the quota. Well, the data science team is not going to go talk to the head of sales and say, ‘Change the quota for your salespeople.’ They’ll say, ‘I don’t talk to you.’ So the data science team needs to work through their chain of command to get to the CIO, to then get to CFO to engage CEO and chief sales officer to influence that outcome. And then, the chief sales officer needs to work with individual regional chairs who say, ‘This is a great idea, but my bonus is tied to different outcomes for the whole year. So we can do it next year. Let’s put it in the planning process.’
The companies that manage to break through this are the ones that are delivering value. And it’s easy to break. You take that business leader and tag them together with the data science team and the tech team. They’re in a squad, and they’re given permission to change the workflow and the incentives. You create a forum where they regularly discuss these things and act on them, and you make it visible to the CEO. It’s manageable, but you need to manage it that way. Currently, people are in their silos and they have their objective.
