Companies Are Pouring Billions Into A.I. It Has Yet to Pay Off.
Corporate spending on artificial intelligence is surging as executives bank on major efficiency gains. So far, they report little effect to the bottom line.
According to recent research from McKinsey & Company, nearly eight in 10 companies have reported using generative A.I., but just as many have reported “no significant bottom-line impact.”…
Investments in generative A.I. by businesses are expected to increase 94 percent this year to $61.9 billion, according to IDC, a technology research firm.
But the percentage of companies abandoning most of their A.I. pilot projects soared to 42 percent by the end of 2024…
The winners so far have been the suppliers of A.I. technology and advice. They include Microsoft, Amazon and Google, which offer A.I. software, while Nvidia is the runaway leader in A.I. chips…. [end quote]
The investment in hardware, software and providing electric power to AI centers is huge. The companies mentioned in the article already have thriving businesses so they can afford to pour money into the technology even if it doesn’t pay off short-term.
Other companies (such as Perplexity buying Chrome)…not so much.
I expect AI will be neither the short term huge economic success many (based on the stock market) expect, nor the bust others (based mostly on quasi-religious moralistic reasoning), and so I am NOT investing in AI companies, but I am carefully pondering hedging against ascendant AI power.
I was very knowledgable and well compensated for knowledge on this subject in the past, but I now find myself adrift as the interaction betwixt the underlying hard maths and algorithms, the economics, and the political/religious reasoning puzzles me utterly.
The bigger budgets would be spent on coding either way. Think Google, Meta, Apple, Nvda, Amazon, and Microsoft. The small players who are not ubiquitous will never wake up from this waste of resources.
Tesla is the divide in the marketplace, it has failed.
Using DSI I extracted the most significant content from the first three posts of this thread:
McKinsey & Company report.
Have you ever thought about how your brain thinks? I was puzzled by how I came up with ideas while sleeping. Ever heard the expression, “There is nothing new under the Sun?” If true, how does progress happen? The answer is evolution, not just what Darwin said but at a cosmic level. An infinite number of particles interacting with each other during endless time. What works becomes reality from Big Bang to AI.
Early in this though journey I figured the brain was a pattern matching machine that stored every input it ever received and somehow matched new input against stored input. One similarity between brain and the universe is the huge number of neurons and an even larger number of synapses the brain is made of. Complexity at work.
The latest version of AI, neural network based AI, is an attempt to mimic the brain. The difficulty is reaching the brain’s scale, billions of neurons and trillions of synapses. Where, on the biological scale, from life-capable molecules to human brains, are today’s largest data centers? Cells? Microbes? Insects?
If tiny brains can solve simple problems and large ones complex ones, the question becomes, what problems can our current sized neural networks and data centers handle successfully?
Jumping to investing, which companies can monetize AI successfully? Failures exceed successes but it’s not AI failing, it’s tasking AI with jobs above its current pay grade.
Tesla has not failed. Tesla is in the process of setting up the AI Monetizing Machines, Optimus robots and Robotaxies, whose intelligence requirements match AI’s current capabilities. Palantir also seems to be on the right path.
A good video about NVIDIA made me revisit the topic. I understand enough (at a macro level) about the training but not about inferencing. This video was very helpful.
AI Inference: The Secret to AI’s Superpowers
AI, the brain, the Universe, are all about being massively parallel.
The above video disclosed an important difference between Tesla and other AI users – massively parallel inferencing. While providers of LLMs need Giga DataCenters Tesla puts an inference chip in every EV and in every Optimus robot. Just the right size massive parallelism at the inference level and the customer pays for the chip up front. LLM providers have to amortize the inference hardware, a less efficient use of capital (cash flow).
The good video about NVIDIA
First critical moment at time 7:15
In the world of AI a chip has to do two very different jobs. Understanding this difference is the key to the entire competitive landscape. The two jobs are train and inference.
Earlier I said that we are trying to mimic how the brain works but there is one huge difference between humans and technology. A human brain has to do both jobs train and inference while technology can specialize separating training from inferencing. Human communications are very slow via language, and every brain has to do both jobs but technology can communicate over long distance much faster.The result is that data centers need not do both jobs, one can specialize on training and others at inferencing. This has economic and investing consequences, monetizing AI.
At Tesla, the very nature of the products creates the separation, a few data centers to train on data collected by the products and an inference chip in every product. Most AI providers don’t have this divide. BTW, this justifies Tesla dropping Dojo. NVIDIA and others can supply the training hardware while Tesla sources the inference chips via ARM Holdings and chip fabs.
This is where my interest in the video ends but there is more interesting stuff’
Time 17:20 - Geopolitical lockout - China’s AI architecture
Time 20:22 - Broadcom - covered call candidate?
Time 27:15 - Monetizing AI
Here it comes. Remember TickleMe Elmo and Furby? Mattel (and others) are trying to put AI into new toys so parents don’t have to pay attention to the kids (again.) Because it worked so well the last time.
So sure, rev up those data centers and fire up the coal mines again, because what we really really need is less time with the kids and more time with, uh, whatever.
Yeah, it seems like a lot of AI is aimed at bringing out the very worst in us. I already have an issue with parents who take their young kids out to eat and have them on a screen the whole time because they don’t know how to parent. Now we have to worry about Tickle Me Elmo giggling to our kids - “Lower!, Lower! Lower!”
Apart from me just being a curmudgeon, there are real world consequences. For example, watch the fertility rate continue to plummet as more and more people choose to have a virtual partner, instead of a real one. Sounds crazy, but it’s already happening.
Most companies are still trying to figure out how to monetize AI in the long term. They’re spending like crazy, betting that revenues will follow. What if they don’t?
Interesting take on the possibility of an AI Kaboom -
“How accounting drives huge AI spending
"What these things do — and they certainly did with fracking, but more ominously with the telecom and dotcom era — is drive capital spending and earnings for a very interesting reason. It’s not just that animal spirits are stirred; there’s also an accounting angle that everyone forgets. When you get a major capex boom focused on physical tech assets, like the fiber and Internet buildout in the late '90s, it significantly boosts reported earnings, even if the underlying returns aren’t there.
What we’re seeing now with AI chips is similar to what happened during the telecom boom. The big tech companies — the so-called Mag 7 — are buying chips as fast as they can. The companies selling the chips book that spending as revenue and profit, just like Lucent and Nortel did back then. But the companies doing the spending — like MCI or AT&T in the past — are the ones taking on the cost. They capitalize those expenses and spread the write-off over several years. So, a chip that might become obsolete in just two years is being depreciated by some of the big hyperscalers over five, six, or even seven years. That creates a huge boost to corporate earnings during tech buildout booms like the one we’re seeing now. That’s number one."
What happens if the AI spending boom stops?
"Number two, the spending — and thus the earnings — can collapse. Everyone kind of remembers the global financial crisis, but they kind of forget that from 2000 to 2002, corporate earnings dropped 40%. The same as the global financial crisis. Because capex spending stopped in 2001, and literally there was no lag time. Earnings just completely collapsed when the spending did.
I’m starting to worry there’s so much spending right now on the AI physical boom — the buildout of data centers, chips, and so on — that if anyone decides to pause and ask, ‘What’s our real economic return here?’ it could be a big problem. It’s one thing when we’re spending $50 billion a year; it’s another when that changes. As a group, we’re spending $500 billion a year, and if there’s no real return, can we really spend a trillion dollars a year on this without seeing results? It’ll be interesting to watch. Right now, spending is growing faster than operating income or revenues."
Go broke? Global Crossing went broke because the price of fiber dropped so drastically that they could not pay off the investment loans. This question is why I’m bullish on Tesla, they have a clear path to monetize AI: FSD, RoboTaxies, and Optimus robots, their in-house products.
It’s the AI suppliers who lose business, not the users who can monetize their AI investment, data centers, etc.
The dot com bust bankrupted websites and optic fiber providers.
The financial crisis bankrupted housing speculators and financial institutions.
Who will go bankrupt in an AI crash? Buy stock in companies that bounce back.
“(Musk) told investors in July that Tesla Robotaxis would be available to “half the U.S. population” by the end of this year. Austin’s population is about 0.3% of the U.S. population, and the robotaxi service isn’t even available to everyone in the city, only to carefully selected Tesla fans.”
Did you read the article? Who are these experts? The only experts I found were unnamed analysts and their opinion was positive, not negative.
With Alphabet’s war chest, Waymo can continue to expand and improve despite losing billions of dollars annually—though analysts think it will eventually become profitable.
The author is thinking back in time not ahead into the future.
Tesla doesn’t have that luxury. The stakes are much higher for Musk. Unless it pivots back to making sensible EVs, robotaxis will be the make-or-break test for the brand.
The whole point is that individual car ownership is becoming way too expensive and with TaaS it will become even less competitive vs. ride hailing.
Business strategy and technical challenges aren’t the only obstacles for Waymo and Tesla. A new report shows robotaxis are already having a tangible impact on human cab drivers in some of America’s biggest cities.
When Henry Ford got going horses lost their jobs, that didn’t stop cars from going forth and multiplying.
As a management consulting company, McKinsey wants its slice of AI Pie. So far they are a spectator and not a player. Typically the management consulting gets their share of every technology transition and this transition from what I could gather, management consulting is greatly absent.
So come up with a report to scare the C-suite and board. Then sent your “rain-makers, and big-swinging-d..” tell the companies, we know what mistakes companies are making, and how to avoid, how to extract the value, yada, yada, yada… C-suite is going to hire them as CYA.
It may sound cynical, but think about it for a minute…
The timing also aligns with the new Texas Senate Bill 2807, which now governs the deployment of automated driving systems in the state, coming into effect on September 1st.
…
However, SB 2807 introduces stricter oversight for truly driverless operations, including requirements for safety data reporting, first-responder interaction plans, and potential revocation of authorization if safety standards aren’t met.
Companies must also demonstrate that their systems can achieve a “minimal risk condition” (e.g., safely pulling over) in case of failure and need to qualify as a level 4-5 autonomous driving based on the SAE standard.
To achieve level 4-5, there can be no human oversight.
Therefore, by moving the supervisor into the driver’s seat, now serving as a safety driver, the system reverts to level 2, allowing Tesla to continue operating its “robotaxi” without achieving level 4 autonomy.
I believe the safety monitor moves not from back to front but from right to left. How does that help when there are no pedals, switches, or steering wheel?
I hope you are aware that there are in fact, pedals, switches and a steering wheel in all Tesla robotaxis. In fact, they are pictured in the video in the link of someone’s actual experience on 9/2.
Hi Wendy! Your point about the massive AI investments with little payoff so far is really thought-provoking. It’s striking that McKinsey’s research shows nearly 80% of companies using generative AI but seeing no real financial impact yet. The big players like Microsoft and Nvidia can absorb these costs, but smaller companies like Perplexity taking risks on acquisitions sounds like a tougher gamble. It makes me wonder if the payoff is just delayed or if some firms are overestimating AI’s short-term potential. I think focusing on specific, measurable AI applications, like automating customer service, might help companies see returns sooner. What do you think the tipping point will be for AI to start delivering real bottom-line results?
I have little doubt AI will be something, it’s just that - like the internet in 1995 - nobody exactly knows what that might be someday. Still, companies are rushing into it for FOMO, fear of missing out, and investors are right behind. Never mind that billions were wasted on “the internet” in the 90’s, everything from Pets.com to Global Crossing and more.
A recent study from MIT shows that 95% of AI pilot projects have been abandoned, with consequent loss of dollars along the way.
While many companies are investing heavily in AI, a recent MIT study and other reports indicate that a vast majority of AI projects—especially pilot programs—are not delivering a clear return on investment (ROI) and are being dropped
. A viral August 2025 MIT report, “The GenAI Divide,” found that 95% of corporate generative AI pilots failed to deliver measurable financial results.
There will be winners, there will surely be (already are!) losers. Problem is nobody knows which is which yet.
