저자: AdamFeather · 외국인
요약 보기
결론: 영상은 마이클 버리의 소프트웨어 기업 50곳 AI 위협 등급 분석을 소개하며, 대부분의 기업이 AI 위협에 노출되어 있지만 일부는 저평가 상태일 수 있다고 본다. 다만 명확한 가격 방향 예측은 거의 없다. 배경: 2026년 2월 5일 AI 기업의 비즈니스 소프트웨어 연동 기능 출시 이후 미국 소프트웨어 주가가 6거래일간 약 17% 하락했다. 원인은 AI 자체뿐 아니라 사모신용 대출 부실과 헤지펀드 모멘텀 청산이 겹친 결과로 제시된다. 위협 구조: AI 위협은 두 가지 방식으로 나타난다. 첫째는 소프트웨어 자체의 대체, 둘째는 좌석당(per-seat) 과금 구조에서 인력 감소에 따른 라이선스 수 감소다. 기업들은 과금 방식을 AI 사용량 기반으로 바꾸려 하지만, 버리는 이것이 인력 감소분을 가리는 수단일 수 있다고 보고 수익률(마진)을 보라고 권한다. AI 에이전트 운영 비용은 여전히 높고 투자금으로 보조되는 면이 있어 실제 대체 비용은 불투명하다. 경영진 발언 해석: 버리는 경영진 발언의 패턴을 지적한다. '아직 좌석은 성장 중이다'는 '아직'이라는 단서가 붙은 말로, 향후 감소를 암시한다고 해석한다. 세일즈포스 CFO의 마진 질문 회피는 마진 위험을 시사한다고 본다. 워크데이의 주식 보상 비중, 허브스팟의 14% 목표는 주주에게 부정적으로 평가된다. 페이컴은 자사주 약 15%를 차입으로 매입한 점이 긍정적으로 평가된다. 5단계 분류: 요새(Fortress, AI 공격 또는 위협 없음), 성(Castle, 강한 해자와 자체 AI로 반격), 예배당(Chapel, 위협이 크지만 자체 AI와 고객 이탈 비용 보유), 돌(Stone, 위협 대비 적응력 부족 또는 경쟁 치열), 나무(Wood, 차용 AI로 멸종 위협). 버리는 50개 기업 중 요새와 나무 등급에는 아무도 해당하지 않는다고 본다. 성 분류 예: 인튜이트(턴택스), 팔로알토 네트웍스, 유니티, 알파벳(조건부). 예배당 분류: 어도비, 세일즈포스, 서비스나우, 오토데스크. 돌 분류: 워크데이, 허브스팟, 먼데이닷컴, 도큐사인. 페이컴은 정부 세금·급여 체계 연동으로 돌이지만 성에 가깝다고 평가. 가격 관점: 크라우드스트라이크는 성 등급이지만 주가가 15% 연수익 기대치 대비 과도하며, 주식 보상 희석이 연 매출을 초과하는 비용을 유발한다고 평가된다. 어도비는 예배당이지만 주가가 이미 약 2/3 하락해 저가 매수 대상이며 버리가 보유 중이라고 한다. 소프트웨어 섹터는 '가장 깨끗한 돼지도 우리 안에서는 지저분하다'는 비유로 요약된다. 보안 기업의 고객 데이터 집중은 장기 리스크로 지적된다. 투자자 체크리스트: AI를 자체 보유하는지 임대하는지, 공식 기록 시스템인지, 좌석당 과금 구조인지 확인하라고 권한다. 결론적으로 시장은 성 기업을 과대평가하고 예배당 기업을 과도하게 할인했을 수 있다고 본다. 출연자 전망 요약: 진행자는 버리의 분석을 소개하며 어도비 저가 매수 논리에 공감하나, 명시적 1~3개월 가격 방향 예측은 제시하지 않았다. 장기적으로는 소프트웨어 기업 중 일부가 저평가 상태일 수 있다는 중립적 견해를 보인다.
다듬은 글
On the 5th of February this year, an artificial intelligence firm released add-ons that let its chatbot log into business software and do the work itself. Six trading days later, American software shares had lost another 17%. Companies that had been stock market royalty for 20 years were priced like video rental shops.
Wall Street had reached a verdict. AI eats software, all of it. Michael Bur, the investor who saw the 2008 crash coming, thought that was lazy. So, he did the dull, slow thing that almost nobody does in a panic. He went through 50 software companies one at a time and sorted them by how well each would survive.
He even gave the categories names like a medieval estate agent. Fortress, castle, chapel, stone, wood. What he found does not fit the headlines at all. Nobody so far makes it into his top category. The company he ranks dead last is one of the safest and the one he likes second best enough to own it is supposed to be finished.
ot advice. I should also confess that I edit this channel with one of these AI tools. So, I'm not a neutral party in the war on software. I'm a small enthusiastic part of the invading army. We will start with what actually caused the crash because it was not only AI. Then, how AI really threatens these businesses and a short guide to what their bosses mean when they speak.
After that, Bur's five categories, who is in each, and why the safest company on the list is the worst possible investment. For 20 years, the most admired business model on the stock market was software you rent. You pay every month forever for the program your company runs on. Investors adored it. At the peak, some of these businesses were valued at 30 times a year's sales, which is roughly like paying 30 years rent up front for a flat because the landlord has a nice website.
Bur's first point is that the rot started somewhere much less glamorous than AI. Late last year, two American companies most people had never heard of went bust. First Brand, which makes car parts, and Triricolor, a lender for secondhand cars. Neither had anything to do with software, but both had borrowed from private credit funds.
The same lenders that had also lent heavily to private software firms. Private credit defaults hit a record last year. Private loans are hard to sell in a hurry. So, the lenders protected themselves the only way that they could. They sold shares in public software companies that looked a bit like their loans.
The software crash began as somebody else's fire drill. Then came the 5th of February and the selling finally had a story. The big hedge funds that trade on momentum all charged for the exit together. Bur calls that the third dagger, this one in the back. By the spring, Adobe had lost about 2/3 of its value from its peak.
into it about 60%, docuign about 85, and unity almost 90. His description of the mood is my favorite line in the whole series. Investors, he wrote, had concluded that the LLM ate my baby. So, what is the actual threat? This is part two. How AI eats a software company. Well, it comes in two flavors. The obvious one is replacement.
If an AI agent can fill in the forms, chase the invoices, and update the records, why pay for the software at all? The quieter one is probably bigger. Most of these companies charge per seat, meaning per person who logs in. If AI lets a firm do the same work with half the staff, it needs half the login.
Nobody cancels anything. The bill just quietly halves. Picture owning a gym that charges per member and discovering that your biggest corporate clients have replaced half their staff with machines. none of whom are interested in spin classes. I see a small version of this in my own work. A few years ago, editing videos like this one properly meant paying people or paying for a more license of an expensive software product.
Now, a lot of it gets done by an AI tool on my laptop. Multiply me by every small business on Earth, and you can see why software investors are not sleeping very well at all. The company's answer is to change how they charge from a fee per person to a meter that ticks over whenever an AI agent does some work.
That's not a silly idea. It might even grow their businesses. But Michael Bur points out that it also makes it much harder for an outsider to tell whether the meter is making up for the people who have gone. When that happens, he says, "Ignore the story and watch the profit margins." To be fair to the software companies, there is a case they do not make very well themselves.
Running AI agents is still expensive and a lot of it is being subsidized by investors pouring money into AI firms at a loss. Michael Bur calls it modern-day predatory pricing. If that money ever dries up, the true cost of replacing software with AI may look a lot less attractive. And the first real casualties are already visible.
Cler, the Swedish buy now pay later firm ripped out both Salesforce and Workday and rebuilt with AI. Michael Bur knows its boss personally and describes his enthusiasm for cutting costs this way as epic. Which brings us to part three, the executive phrase book. Michael Bur listened to these companies results calls and one pattern made him laugh.
Almost every boss boasts that AI has made their own staff wildly more productive. Almost every boss also insists that their customers are not using AI to need fewer staff. Both cannot be true for very long at the same time. So, here is a small phrase book for travelers in a land where nobody quite says what they actually mean.
When a finance director tells investors that seats are still growing, Michael Bur's ear catches the word still. He calls it the more subtle cousin of if. If these executives were completely honest, he says, every one of those sentences would end with one more word, yet. When Salesforce's finance director is asked whether AI will hurt profit margins and answers with 90 seconds and 175 words about tokens and efficiencies, that in Michael Bur's translation means margins are definitely in trouble.
And when the boss of service now announces that the company does not live in the software as a service neighborhood anymore, I think it means that they moved out just before the house prices collapsed. That's what Bur says. When Workday's returning co-founder called his comeback chapter 4, Michael Bur notes that for shareholders, the first three chapters were grotesque.
And when HubSpot proudly promises to cut the share of its revenue paid to staff in shares down to 14%, he replies that even that target is egregious. One boss does get a good review. Paycom, the payroll company, bought back around 15% of its own shares in a single quarter, borrowing in order to do it because it thought that those shares were too cheap.
If you want your team to make a stand, Bur wrote, Paycom just did. Which brings us to part four, the five buildings. Now, the categories. Michael Bur calls the system AI competitive threat tiering, which is the sort of name only a hedge fund manager could love. The buildings are much easier to remember.
A fortress either uses AI to attack or faces no real threat from AI at all. Its software is woven so deeply into regulation and compliance that switching it off would mean a very awkward conversation with a regulator. A castle has a deep moat and serious AI of its own at real scale. It is not just defending, it is firing back.
A chapel faces a sharp threat but owns real AI and a proper research team and its customers would find it painful to leave. Solid walls, loyal congregation not designed for a siege. Stone is a real threat with little ability to adapt or a lesser threat in a brutally competitive market with no mode to speak of and wood products that may be everywhere but with AI that is borrowed not owned and almost no research of its own.
Michael Bur's assessment is that a wood company is threatened with extinction. Wood, for the record, is also the house the wolf blows down in the three little pigs story. I don't know whether Michael Bur meant that. I would like to think though that he probably did. Which brings us to part five. Who lives where before the answers a small game?
Crowd Strike is the cyber security company whose faulty update crashed millions of computers in the summer of 2024, grounding flights and closing hospital systems. Which building do you think it lives in? Hold that thought. Here is the strangest result first. Across the 20 odd companies Michael Bur has published so far, Bur has not put a single one in the fortress category.
Nobody in wood either. The penthouse and the basement are both empty and everybody is crammed into the middle floors. In the castles category, he puts into it the maker of Turboax, which is using AI aggressively. Its AI tax agent finds small businesses an average of over $1,000 in extra deductions. As someone who reads tax law for fun, I find that both impressive and slightly threatening.
Paulo Alto Networks is a castle as well. So is Unity, whose software builds about 7 in 10 mobile games. And so is Alphabet, the owner of Google, though only just because nobody knows yet what chatbots will do to search. The chapels include Adobe, Salesforce, Service Now, and Autodesk, whose design software is used by architects and engineers worldwide.
Autodesk shows the dilemma neatly. AI helps its customers design faster, but faster design means fewer engineers, and fewer engineers means fewer Autodesk licenses. It may have to charge more per person just in order to stand still. The stones category include workday, hubspot, monday.com, and docyign.
Workday is the odd one. Its sales have grown enormously for a decade, yet its shares were back at 2018 levels. Michael Bur's explanation is that so much value went to staff in shares that owners over 14 years ended up with less than nothing. Docuine has a different problem. Its signature business has stalled and most business contracts start life in Microsoft Word where Microsoft's own AI assistant now lives.
Michael Bur's description of Microsoft is a Star Trek reference. The Borg inevitable. And Paycom is a stone that is almost a castle because its software is plugged into a tangle of government tax and benefit systems that no chatbot can simply replace. Government, in Michael Bur's words, is notoriously bureaucratic and old-fashioned, which for once works in somebody's favor.
Bureaucracy is finally good for something. Which brings us to part six. Safe is not cheap. Back to your guess. Crowd Strike is a castle. Even after the 2024 outage, it is one of the strongest businesses on the entire list. Cyber security is one of the few corners of software where AI may create more demand because every new AI tool is also a new weapon for hackers.
And out of 50 companies, Bur ranks it 50th. The problem is the price. On his numbers, the shares were trading at more than 20 times the level at which he would expect a 15% annual return at an all-time high. And the company has been paying its staff so generously in new shares that on his sums, the dilution costs owners more each year than its entire annual revenue produces.
A superb business at a price that leaves nothing on the table for whoever buys next. Then look at Adobe, the maker of Photoshop. Only a chapel in terms of the category he assigns to it. The case against it writes itself. AI can now make an image from a sentence. He puts it second out of 50. the only big name in his bargain zone and owns a full position in Adobe.
His reasoning here is simple. The shares had already fallen by 2/3 while its customers kept on paying. It was buying back its own shares at a cheap price. And it was building AI that learns each customer's own brand, which a general chatbot cannot very easily copy. In other words, a threatened business already priced for the funeral.
Michael Bur's summary of the whole sector is one of his best lines. The cleanest pig in the pig pen is still a gritty mess. That's what he said in May 2026. In plain English, even the best software company can be a bad buy at the wrong price. In it, a castle on the attack, still cost nearly three times his target price after falling 60%.
He didn't touch it, and two companies came out so badly that on his numbers, there is no price at all, not even 1 cent a share, at which you could expect a 15% return. Michael Bur has one more worry about the castles in cyber security. The big security firms gather enormous amounts of customer data in one place.
In a world of powerful AI hacking tools, he says that pile of data becomes the fatted calf, valuable and vulnerable. He's not even fully reassured by Anthropic's glasswing project, which gives security firms early access to its most powerful model so that they can find their own weak spots first. A model that knows everyone's weak spots, he points out, is exactly the model an attacker would love to copy.
Which brings us to the seventh part of this video. What it means for you. You probably don't own Docuine or Monday.com directly. But if you have a pension or a global index fund, you own a slice of almost every company in this video and of the AI giants trying to eat them. You don't need Michael Bur's spreadsheets in order to borrow his thinking.
Three questions get you most of the way. Does the company own its AI or rent it from somebody else? Renting your future from a competitor is how you end up made of wood. Is it the official record that everything else depends upon? Payroll, tax, and contract systems are hard to replace because they hold the version of the truth that the government and the auditors rely upon.
Michael Bowie thinks that this is the strongest defense in the whole industry. And how does the company charge if it bills per person? Ask what happens when its customers employ fewer people. If you want a bonus question, listen to the boss on the results calls. If every question about profit margins gets an answer about growth and every answer about customers sounds like it is missing the word yet, you've probably learned more than the company intended for you to learn.
And to be fair, there is a version of the future in which Michael Bur is too gloomy about software. These companies are using AI to cut their own costs as well. Freshworks says over half of its code is now written with AI. If that saving flows through to profits faster than seats disappear, some of today's stones could turn out to be perfectly comfortable houses.
And Michael Bur admits it is very early. Almost nothing has actually happened yet except the share prices have moved. Then whatever the answers, ask the question that actually decides whether or not you make any money. What are you paying for the shares? Michael Bur's conclusion after 50 companies is that the market has got it wrong in two directions at the same time.
It has priced the castles as if nothing could ever touch them and written off the chapels far too early. So which companies survive AI? On Michael Bur's list, nobody is safe for certain and nobody is finished either. The penthouse is empty, the basement is empty, and everybody else is reinforcing the walls and hoping the siege engines are slower than advertised.
But the best investment he found was not a castle. It was a chapel bought cheaply while everybody else was busy deciding the LLM had eaten their baby. In this example, he bought Adobe as we've discussed. If you want to read a company the way Michael Bur does from his own account rather than from . It also gives you free access to eight other books I've written and a stack sovereignty quiz at the website which gives you more detailed personalized feedback about your position. Thank you very much for watching. I hope that this video
Comments 0