In context of greater future uncertainty, Shaposhnik prudently reduced software risk many months ago, which has proven to be a reasonably good decision given what has occurred in software.

Software company selling to municipal government market only began investing in AI last year, causing concern about management's forward-thinking capability.

Microsoft is heavily tied to the success of a single LLM, so if that LLM doesn't prove to be the winner of the AI race, they will have a hard time justifying massive CapEx investment supporting a single customer.

There is significant uncertainty about whether software will continue to be one of the highest-quality business models in the world, as it has been thought to be for the longest time.

Great businesses can power through macro shocks and geopolitical dynamics, but it is important to think about what is going on in the world as you manage a portfolio because events could impact specific businesses and opportunities.

Defense spending by NATO countries could increase by a trillion dollars over the next 10 years as those countries get up to NATO targets, creating a super cycle tailwind for aerospace and defense businesses.

There are massive beneficiaries of AI, companies in the middle experiencing muddled impact, and companies that will be clearly negatively impacted by the AI cycle.

The Magnificent Seven experienced extraordinary earnings and cash flow growth as a percentage of S&P 500 that went disproportionately to those seven stocks, while valuation multiples of the rest of the market (Mag 493) diverged massively relative to the Mag 7.

The rotation toward the Mag 493 reminds Shaposhnik of the year 2000 rotation after the tech bubble, when the rest of the market was left behind for several years and then benefited greatly from 2000-2002 as there was a significant rotation back to traditional businesses performing well.

Asset-light, capital-light businesses with high recurring compounding cash flow are becoming asset-heavy, capital-intensive businesses as hyperscalers invest massively in AI infrastructure, which is a transformation that makes them less attractive and more difficult investments.

When Microsoft was asked whether it could return to being a capital-light cash flow compounding machine, management answered that CapEx growth for their business will grow more slowly than the cloud business growth rate, which means Microsoft is unlikely to get back to being capital-light for the foreseeable future.

The better route for capturing AI upside with less risk is investing in semiconductor suppliers of picks and shovels who serve multiple LLMs and hyperscalers rather than betting on specific LLM winners, because suppliers have more diversified customer bases and healthier access to capital.

On the magnitude vs. timing of AI investment payoff: It's producing significant growth for OpenAI and other LLMs right now; they're monetizing IP and growing revenue very rapidly. The question is whether they will generate significant free cash flow at some point, and whether there will be multiple LLMs at scale generating high returns, which technology history suggests is unlikely given winner-take-all markets.

Technology sector history indicates that it's closer to a winner-take-all market than a four or five horse race, which is why those investing massive money into the future are betting on specific customer duration and specific customers who will win.

The fund reduced software exposure dramatically in mid-2025 due to two factors: (1) realization that AI would disrupt many software businesses at an accelerating pace as LLMs improve rapidly, and (2) the illness and departure of Mark Leonard as CEO of Constellation Software, which triggered a disciplined reevaluation.

The Buffett 1987 letter quote: 'Experience indicates that the best business returns are usually achieved by companies that are doing something quite similar today to what they were doing 5 or 10 years ago...a business that constantly encounters major change also encounters many chances for major error...Such a franchise is usually the key to sustained high returns.' This principle drove the software exit.

In an era of AI and AI disruption, there is no substitute for organizations that are decentralized, have appropriate incentives put in place for decentralized businesses, and have a track record of being learning cultures.

Job opening data for programmers and engineers have been growing very rapidly over the last 6 months, suggesting that AI may actually increase the need for software engineers and developers rather than displace them.

Amazon has experienced significant disruption and bugs from employment of AI in their development processes and has had to throw a lot of people at fixing those bugs created by AI, suggesting AI complexity requires more skilled human resources.

The fund's focus on recurring revenue businesses that are needed by everybody or by everybody in a particular market provides some protection against K-shaped economy outcomes because if the high end experiences pressure but the low/middle end picks up, it should be a wash.