Are AI Chips Really Undervalued? Why Market Doubts Persist
Demand is strong and earnings are record-setting, yet shares remain jittery. The question for investors is now the durability of profits and the ability to recoup capital, more than the growth rate.
The 20-second read
1. Why are chip stocks nervous when AI is growing?
The debate over AI semiconductors looks odd. Demand is strong and companies talk about record earnings, yet the market still doubts that high share prices can be justified.
The two facts do not contradict each other.
For an AI chip supplier, the money booked as revenue is capital spending for its customer, the cloud company. The supplier earns a profit when it sells. The buyer has to run data centers and recover the investment over several years.
The problem is that time gap.
Even if AI usage keeps rising, returns can fall short of expectations if service prices drop or if power and depreciation costs climb quickly. Customers could then delay their next chip purchases.
Charles Schwab has made a similar point. AI infrastructure investment is expanding, but monetization is still at an early stage, and financing conditions are emerging as a new constraint.
What the market doubts is less the future of AI than how long the current pace of investment can be sustained.
2. Why a low P/E still leaves investors uneasy
A high-growth chip company with a lower price-to-earnings ratio than a traditional industrial can look attractive.
But the P/E ratio has a trap. Share prices reflect the future, while earnings, the denominator, are just a number for a particular period.
| Measure | What investors should check |
|---|---|
| Trailing 12-month P/E | Based on earnings already reported |
| Forward 12-month P/E | Based on expected earnings over the next year |
| P/E on a specific year's estimate | Based on the forecast for that fiscal year |
Comparing P/E ratios built on different periods can distort the very judgment of whether a stock is cheap.
More important is normalized earnings: the profit a company can keep earning after supply shortages ease and rivals raise output.
Two possibilities coexist in the market.
First, today's high margins may be temporary excess profits created by scarce supply. In that case, a low P/E reflects a future decline in earnings that is already priced in.
Second, structural growth in AI may have set a new baseline for chip demand and profits. In that case, a low P/E could be a substantial investment opportunity.
The key is not the growth rate itself. It is how long current earnings can last.
3. What does SK hynix's "sold out through 2029" prove?
In memory, real evidence of long-term demand is appearing.
On its third-quarter 2026 earnings call, SK hynix said its DRAM and HBM volumes were effectively sold out through 2029. That figure, however, combines long-term supply agreements already signed with purchase demand still under discussion with customers.
Confirmed long-term agreement (LTA) volumes account for more than 50% of total production capacity in 2027 and 2028, with 14 contracted customers.
How long-term demand turns into cash revenue
- Demand talks
- Long-term contract
- Firm order
- Shipment
- Revenue recognition
- Payment collected
The further along the chain, the greater the certainty of revenue and cash flow.
This is evidence that medium- to long-term demand is very strong. It does not mean that output and selling prices through 2029 are both locked in.
What matters is how firmly the contracts bind buyers to purchase, whether prices can be renegotiated, and whether delivery timing can be adjusted.
HBM profitability can also shift with rivals' capacity additions, customer qualification and product generation changes.
The memory cycle is not disappearing. The more useful question is whether the cycle in HBM could diverge from the cycle in commodity memory.
4. Another AI risk: money borrowed against GPUs
Even if AI usage keeps growing, some data center companies could run into financial trouble.
The reason is that a GPU's technical life differs from its economic life.
If the next generation of GPUs improves sharply in performance and power efficiency, older GPUs may keep working fine while their rental rates and resale values fall.
How GPU financing risk could unfold
- Better economics for next-generation GPUs
- Lower rental income and collateral value for existing GPUs
- Weaker cash flow at the borrower
- Pressure to post more collateral or repay debt
- Less capacity to buy new AI chips
The risk matters most for infrastructure operators that rely heavily on outside borrowing. They should not be judged by the same standard as hyperscalers with large internal cash generation.
That a server keeps running does not guarantee the safety of a loan issued against it.
5. Is the Nvidia probe a crisis for the whole AI industry?
Nvidia faces company-specific risks that should be assessed separately from the industry outlook.
They are a US government investigation related to MegaSpeed and allegations of foreign intelligence activity targeting CEO Jensen Huang.
Nvidia's board formally announced an independent investigation on November 8. It said that, for now, it has no basis to conclude that national security or company secrets were compromised.
This is a different kind of event from a decline in AI chip demand. It is closer to export-control, compliance and executive governance risk.
Investors should focus less on the sensational details of the allegations than on the actual findings and the scope of any penalties. If restrictions on export licenses widen or a serious leak is confirmed, company value could be affected.
For now, it is reasonable to judge industry demand separately from the risk premium on Nvidia as an individual company.
- A low P/E is a starting point for analysis, not a reason to buy.
- Compare earnings over the same period, and weigh normalized earnings together with the current share price.
- Long-term contracts can improve demand visibility and soften sharp swings in results.
- However, cycles driven by competitors' capacity expansion, price renegotiations, yields and product transitions still exist.
- Yes. Falling unit prices, operating costs and depreciation can outweigh the growth in usage.
- What matters is not usage but utilization, per-unit profitability and free cash flow after the investment.
- Set an investment horizon of at least five years and prefer dollar-cost averaging (DCA), buying a fixed amount each month.
- Diversify mainly among quality companies and keep out money you may need for near-term living expenses. If the investment thesis breaks down, review your position size.
What to watch
Insight Times Editorial Desk
