2027 Investing Needs Alignment, Not Prediction
A new investing book argues the hard part was never picking the next winner. It's staying aligned with where value and bottlenecks are moving while keeping old human weaknesses in check.
Good investing methods are old. Good investment targets keep changing.
Reading investing books long enough, you run into a strange contradiction. Warren Buffett's principles haven't changed much in decades. Buy good companies, don't overpay, hold for a long time, and change your mind when you're wrong.
But look at the market, and it gets harder every year to invest with a decade-old map of the industry. Smartphones, cloud computing, GPUs and generative AI have rewired entire corporate value chains in just a few years.
That's the strength of this framing. It isn't "throw out every principle because it's a new era," and it isn't "just stick to the old ways." It comes closer to: keep the unchanging weaknesses of human nature under control, while aligning quickly with the direction value is moving.
The market doesn't know your cost basis
When a stock falls from 100 to 50, people wait to break even. When it rises from 100 to 200, they want to lock in gains and sell. So in a portfolio, good stocks shrink and bad stocks grow.
One line from this framework sticks: "Return rate is ego. Return in dollars is your account balance." If an investor won't buy more of a good company because a rising average cost basis would lower the displayed return, they're managing their own story about investing more than they're managing money.
Even in 2027, the most dangerous variable isn't AI. It's old human habits: break-even anchoring, ego, overconfidence, FOMO. Technology changes. Loss aversion doesn't get an update.
The 2027 investing process fits in five words
01. Alignment — Watch where capital, demand and productivity are moving.
02. Bottleneck — Find what the whole industry is short of.
03. Moat — Figure out whether that shortage is temporary or structurally defensible.
04. Price — Calculate how much of the good future is already in the current stock price.
05. Survival — Keep position sizes small enough that being wrong doesn't stop your compounding clock.
This flow fits AI investing in 2027 particularly well. Guessing which model is smartest matters less than tracking how fast the cost of producing intelligence is falling, and how much usage grows as a result.
The core of the AI revolution isn't "smarter." It's "cheaper intelligence."
In July 2026, OpenAI said that price per million tokens fell 97% from GPT-4 to GPT-5.4. Today, GPT-5.6 Luna's API pricing runs $0.20 per million input tokens and $1.20 per million output tokens.
What matters here isn't the token price itself. As intelligence gets cheaper, small cognitive tasks that used to be too costly to hand off to people start entering the market. Meeting-note summaries, customer inquiry triage, document comparisons, test-code writing, first-pass contract review, all of these can turn into billions of individual AI tasks.
- -97%: The price decline per million tokens OpenAI cited from GPT-4 to GPT-5.4
- $89 billion: NVIDIA's data center revenue in fiscal Q2 2027, up 117% year over year
- 10x: NVIDIA's stated target for cutting inference cost per token on Rubin versus Blackwell
Costs are falling while infrastructure revenue keeps growing. It looks like a contradiction, but it's actually the core of the AI investment thesis. If usage grows faster than prices fall, the total market expands. In 2027, the question isn't "tokens got cheaper." It's how much more useful work can now be done for the same dollar.
Bottlenecks make you money. Moats keep it.
Early in the AI buildout, GPUs were the bottleneck. As GPU supply grew, HBM, networking, power, cooling, transformers and data centers each became critical in turn. SK hynix said it began mass shipments of HBM4 in the second quarter of 2026 and signed long-term supply agreements with about 10 core customers. In August, the company decided to invest roughly 54 trillion won in its Yongin Y2 and Cheongju M17 facilities.
Still, a bottleneck and a moat shouldn't be confused. Prices rising because of a supply shortage is a bottleneck. Technology that rivals can't easily replicate even after they expand capacity, customer lock-in, ecosystems, software and economies of scale, those are moats.
Treating NVIDIA as just a GPU company misses this distinction. CUDA, networking, system design, developer tools and cloud partnerships are all bundled together. Competition itself is widening from the chip level to the rack, to the data center, and ultimately to how many tokens can be produced per megawatt of power.
Between a good company and a good stock sits "expectation"
The claim that AI is changing the world and the claim that AI stocks are cheap right now are entirely different claims. The internet revolution was real, but that didn't mean every internet stock price in 2000 was right.
That's why reading a P/E ratio simply as "years of earnings" falls short. A multiple is closer to a timetable the market has attached to a company's future. A high multiple demands long growth and a strong moat. A low multiple can mean the market doubts that growth will last.
| Question | The surface way to look | The question that matters more in 2027 |
|---|---|---|
| AI models | Who's the smartest | Who processes the most useful work at the same cost |
| Semiconductors | Who makes the fastest chip | Who lowers the system-wide cost per token the most |
| Memory | Is HBM in short supply | Does that shortage translate into a moat through long-term contracts, technology and customer relationships |
| Stock price | Is it a good company | How much of that good future is already priced in |
