Four AI Truths ThaT Will Define The Next Decade
Every fall, leaders from Alpine HQ and our portfolio companies gather for our annual Growth Summit—an opportunity to innovate, connect, and celebrate. The theme of 2025’s Summit was AI Innovators, and the post below is adapted from my keynote speech, where I shared four truths that Alpine is using to guide our decisions as we navigate this tech revolution. I hope they help you, too.
In 1996, when Gary Kasparov had been the world’s number one chess player for more than a decade, he agreed to a six-game match against IBM’s supercomputer, Deep Blue. Chess programs had been around since the 80s, but they weren’t very good. (I was a mediocre chess player, and I could beat the one that came installed on my Mac.) By the time Kasparov was preparing to face Deep Blue, though, he knew the programs had improved dramatically, so he developed a plan to use an unconventional, flamboyant style of play that he thought would confuse the computer.
His plan immediately backfired. Deep Blue shocked the world by winning the first game.
That night, Kasparov devised a new plan: a patient, positional approach in which his advantage wouldn’t emerge for several moves. This time, his plan worked. At the end of the series, IBM’s programmers walked away deflated, having spent years—and millions of dollars—on a four-to-two defeat. The human prevailed, and it wasn’t even close.
Truth #1: With large enough datasets and enough processing power, AI begins to resemble human intelligence
The following year, Kasparov and Deep Blue had a rematch. After the first game, it seemed little had changed; Kasparov used the same strategy that had worked for him in the first series and won. But then, in the 37th move of Game Two, Deep Blue played its bishop to an unprotected square. It was an uncharacteristically “human” maneuver, and it not only set the computer up to win but also reshaped our beliefs about what machines can do.
It was shocking, and theories swirled: Did Deep Blue simply make a mistake that worked in its favor? Had IBM somehow cheated? Kasparov himself said he thought a grandmaster might be behind the scenes feeding the computer moves. But eventually, he realized that wasn’t the case. The truth was, Deep Blue had simply thought ahead of him—far ahead.
Kasparov lost that game and would go on to lose the series. The greatest chess master to have ever played the game lost to a computer.
So, what changed between 1996 and 1997? Did the programmers redesign Deep Blue’s architecture? Did they come up with some new algorithm? No. What happened is the same thing that’s happened every single year since modern computing began: processing power increased. In 1996, Deep Blue could process 100 million moves per minute; by 1997, it was 200 million. That was enough for its tactics to feel like the computer was “thinking”—even to the greatest chess player in the world. In the rematch, Kasparov had come face to face with an undeniable fact: With large enough datasets and enough processing power, AI begins to resemble human intelligence.
“What happened is the same thing that’s happened every single year since modern computing began: processing power increased.”
Truth #2: Processing power will continue to increase exponentially
In 1997, advanced CPUs contained nearly 10 million transistors per microprocessor, or TPM—a good shorthand for processing power. A decade later, 300 million was the norm, and recommendation engines, sentence completion, and machine learning had become commonplace. By 2017, another decade later, transistor counts broke into the billions, powering voice recognition, facial recognition, and more. And that same year, Google’s breakthrough paper, Attention is All You Need, introduced the idea of “transformers,” AI models that assigned weights to the relationships between words and sentences to understand language. It’s the same technology that now powers large language models (LLMs) such as ChatGPT.
Today, GPUs are at 80 billion TPM and greater—8,000 times the power that made Kasparov wonder if Deep Blue could “think.” And according to Moore’s law, which states that transistor counts double about every two years, we’ll be in the 400 billions by 2030, if not sooner. In other words, the next five years of compute power increases will be four times greater than the increases in all of history to date.
“The next five years of compute power increases will be four times greater than the increases in all of history to date.”
Is there an AI tool you’re using or evaluating now that seems a little clunky? Maybe you’re wondering whether AI will ever be able to handle a certain task.
It likely will, and you won’t have to wait long.
Truth #3: You have the right to win.
When people use the term “AI” today, they’re usually talking about LLMs such as ChatGPT, Gemini, or Claude. These are the center of the technology—like the meat of a sandwich. Underneath them sits infrastructure: data centers, compute power, and energy. (This layer, by the way, is where investment dollars are going right now; a projected $7 trillion will go to infrastructure in the next five years.) Then there are the apps, which sit on top of the models—just like the apps that sit on your iPhone. Maybe you’ve used some of them. If you run a call center, for example, you might feed data, such as call logs, into an app that powers your chatbot or helps you book appointments.
But these apps, even the big-name ones, aren’t performing the core AI work themselves. Fundamentally, they’re “prompt-engineering” the LLMs that sit underneath them—using specialized knowledge to improve the output, just like when you tell ChatGPT to use a specific tone in an email, or to limit a paragraph to 150 words. Unlike previous technological revolutions, you don’t need specific knowledge or coding skills to access these tools; you just need 20 bucks per month. In this sense, AI is far more democratized than any other technology revolution we’ve seen.
“Unlike previous technological revolutions, you don’t need specific knowledge or coding skills to access these tools; you just need 20 bucks per month.”
Remember, too, where the information that’s powering all of this comes from. If you run the call center, it’s not the LLM or the app that owns those hundreds of thousands of call logs: It’s you. That data is yours, and that means you have the right to win. You just need to capture your data, apply AI, refine, and repeat.
We see evidence of this dynamic all over Silicon Valley. There are only a handful of LLMs to invest in, and it’s not clear that they’re even actually going to make money. The big venture firms invest in apps, too, but most of those businesses are built on a relatively thin layer of intellectual property. So now, VCs are also beginning to invest in aggregating services businesses, an approach where Alpine has a long and established track record.
Truth #4: You will win by planting oak trees now.
This is the second major tech revolution I’ve lived through, and, like many of you, I’ve noticed a pattern.
PHASE 1: Magic. You can’t believe what you’re seeing, and it feels like science fiction.
PHASE 2: Over-hype. Everyone clamors to get involved—until the bubble bursts.
PHASE 3: The long run. Eventually, we see how the underlying technology was actually under-hyped.
Truly transformative technologies are rare, but they typically have a greater impact than any of us can imagine at the start. Take the automobile:
PHASE 1: In 1902, every vehicle on Fifth Avenue in Manhattan was a horse-drawn carriage. The first time anyone saw a car—a carriage moving without a horse—it was mind-boggling.
PHASE 2: Over the next decade, thousands of auto companies sprang up, and anything with the word “motors” in it got funded. Most of them flopped.
PHASE 3: By 1913, Fifth Avenue was lined with cars—and today, 92% of U.S. households have at least one.
The internet was the same. When I started my career as an analyst at Morgan Stanley, we routinely shipped financial models to colleagues on floppy disks via FedEx. I’ll never forget the day our IT director plugged our computer into a phone line and sent one to a team across the country in seconds. It was amazing—and the hype ensured that every web business launching around that time got funded. But despite all the excitement and investment, the dot-com bubble eventually burst.
Long term, though, I would argue the internet was actually under-hyped. Look at where we are today: I bet it’s been decades since you went a week without Wi-Fi.
Right now, AI is in the over-hyped phase—the pre-bubble-bursting mayhem. The first time you used ChatGPT, it was probably hard to believe it was real. Now, you may have started to wonder if it’s really better than the alternative. Its clunkiness is becoming apparent. Many of the countless AI companies that got funded won’t weather this storm. But rest assured: There will be a long run this time, too.
Where we are now with AI is where Walmart was with the internet in 1995. Online shopping was all anyone could talk about, but it was still impractical for mass adoption in the short term. Netscape had only recently gone public; meanwhile, Walmart was the largest retailer in the world by revenue, and took in one of every six dollars spent at U.S. general merchandise stores. That year, their CEO said online sales would “always be small,” echoing dismissals that were common at the time from corporate leaders:
“We’ll keep an eye on it.”
“No one is selling much online.”
“Customers won’t spend more than $100 online.”
“We’ll wait until the tech matures.”
“Our CTO is going to build a website.”
In the short term, those Walmart leaders were exactly right, but we all know how the story of Walmart and Amazon ends. In the middle, though, both companies faced all the same problems. Carts were abandoned over shipping fees; pages crashed in the middle of orders. The two companies navigated the “clunky period” very differently, but I would argue they both made optimal decisions. The only difference? Their timeframe.
Walmart was the incumbent. They had the merchandising, distribution centers, and sourcing deals with China. They should have won. But while they were optimizing for the next quarter’s earnings call, Amazon was thinking decades out. And today, Amazon is approaching five times Walmart’s market cap.
There’s a Chinese proverb: “The best time to plant an oak tree was 20 years ago. The second-best time is now.” That’s one of my favorite quotes, and it’s the reason Amazon won. If you experiment, invest, and—most importantly—persist through the “over-hyped” phase, you will win.
“If you experiment, invest, and—most importantly—persist through the ‘over-hyped’ phase, you will win.”
It’s not easy to do, and Alpine is no exception. We’ve spun up and shut down AI projects, and we’ve said things like:
“We tried using AI. It didn’t work.”
“We have some people using ChatGPT.”
“We’re a service business; there are fewer implementation opportunities.”
“We’re waiting for the tech to mature.”
“Our CTO is working on something with AI.”
Sound familiar? I’ve said things like this myself. But as we’ve learned from Walmart’s example, this is the formula for optimizing the quarter, not for being the best in the world.
To be the best, you have to skate to where the puck is going to be in the future, not where it has been. You have to prioritize things that won’t be profitable this quarter but will set you up for the long-term win. As you consider where to place your bets in the decade ahead, keep these four truths close:
1. With large enough datasets and enough processing power, AI begins to resemble human intelligence.
2. Processing power will continue to increase exponentially.
3. You have the right to win.
4. You will win by planting oak trees now.
Alpine is fortunate: we’re Walmart in this scenario—the big, well-resourced incumbent—and our team and companies have our support in making the investments they need. But our approach—developing AI literacy, understanding how to leverage the highest-impact tools, and building an AI vision—is something anyone can emulate. And the even better news for us, you, and everyone else, is that it’s early. In terms of AI, it’s only 1995.
So, as you navigate the next decade of AI’s expansion, remember the four truths above, and ask yourself: What oak trees am I planting today?