Semiconductor companies have been exceptional investments. But will they still deliver the best shareholder returns over the next five years?
In AI Capex Is Not AI Profit, I looked at the gap between AI spending and durable profits. This time, I want to compare the businesses supplying AI with those selling it to customers.
The last twenty years
I started with three ETFs that were already trading in 2006: SOXX (semiconductors), FDN (internet businesses) and IGV (software). SPY is the S&P 500 benchmark. [1, 2]

Figure 1. Annualized USD total returns, with dividends reinvested. [1]
Over twenty years, SOXX returned about 17.6% annually, compared with 13.2% for FDN, 12.9% for IGV and 10.8% for SPY. FDN also led in the first decade, at about 13.2% annually.

Figure 2. Ten-year returns from a consistent November 13, 2026 CAGR snapshot; the earlier decade is derived from the ten- and twenty-year figures. [1]
Internet businesses led in the first decade. Semiconductors returned about 29% annually in the second.
These are broad proxies, not today’s AI categories. FDN is not a pure cloud index; IGV does not represent proprietary-data companies such as RELX and Experian. Choosing historical ETFs reduces single-stock survivorship bias, but does not eliminate hindsight in the ETF selection. The figures show a change in leadership, not which AI business model will win next.
Twelve companies, four business models
I’m selecting three established businesses in each category, rather than trying to replicate entire industries.

Infrastructure. Nvidia supplies accelerators and software, TSMC manufactures advanced chips, and Broadcom supplies custom silicon and networking. Their advantage is difficult to reproduce. Their returns still depend on how long customers keep spending.
Cloud. Microsoft, Alphabet and Amazon accounted for roughly 61% of worldwide enterprise spending on cloud infrastructure services in Q3 2026. [3] Their scale is hard to match. The question is how much profit that new capacity eventually earns.
Proprietary data. Even if AI changes how a credit decision is made, a lender still needs reliable credit history. That’s the opportunity for Experian, and a similar question applies to RELX and S&P Global: does AI increase the value of their information or erode what customers pay for it? This is also the question I explored in What Still Gets Paid If AI Changes the Interface?, using credit history as an example.
Workflows. Constellation Software, Intuit and ServiceNow are embedded in customer operations. AI could improve their products, but agents could also reduce paid seats or take over the customer interface.
These are selected leaders, not representative industry portfolios. Some, particularly RELX, Experian and Intuit, also reflect valuation opportunities I’ve researched already. A win would test my stock selection as much as the category.
Subscribe for free to receive new posts
Valuations: what is already priced in?
I compared forward P/E, trailing price-to-free-cash-flow and consensus estimates of free cash flow in FY2027 and FY2028. The valuation baseline is November 13, 2026; the forecast snapshot is November 15. [4, 5, 7]

Figure 3. Forward P/E uses provider-reported earnings estimates, not a standardized earnings year. Forecast P/FCF divides the November 13 market value by each company’s estimated fiscal-year FCF; fiscal year-ends differ. N/M indicates negative FCF. The underlying market caps, currencies and calculations are in companion workbook. [4, 5, 7]
Nvidia and Broadcom trade around 12 times estimated FY2028 FCF. Microsoft and Amazon are above 80 times. For Alphabet, analysts expect a $43 billion FCF outflow in FY2027, narrowing to roughly $0.9 billion in FY2028. [5]
The three data companies trade at roughly 16–20 times estimated FY2028 FCF, and the three workflow companies at around 10–13 times. [5]
Microsoft paid $39.4 billion in cash for property and equipment last quarter, reducing reported FCF. Constellation paid $484 million for acquisitions, a major source of growth that sits outside the usual FCF calculation. [6]
The forecasts are broadly plausible as scenarios, but depend heavily on assumptions about AI demand, capital spending and margins. The multiples are not perfectly comparable: Intuit’s forecast FCF even exceeds operating cash flow in the provider’s tables. [5]
Cloud providers may earn good returns on today’s capacity once utilization improves. If that happens, the forward cash-flow multiples could fall quickly. If capex stays high, they may not. Semiconductor suppliers face the reverse risk if orders slow.
A five-year test
My initial expectation was that the three hyperscalers would outperform. I’m less certain after looking at the cash-flow estimates.
My prediction: an equal investment across the six proprietary-data and workflow companies will outperform an equal investment across Microsoft, Alphabet and Amazon by November 2031. Infrastructure remains a separate contender.
I’ll also report the four original three-stock baskets and SPY. Each starts with equal-dollar positions, receives reinvested dividends, and is not rebalanced. Prices and returns are converted to USD. TSMC is tracked through its NYSE:TSM ADS, not the Taiwan ordinary share. [7]
The paper return baseline is the November 13, 2026 close; this predates publication and wasn’t executable after the fact. I’ll use the first trading-day close on or after November 13 for each annual update. Takeover cash will move into SPY at the next trading close, while distributed spin-off shares remain in the basket. No company will be replaced for poor performance.
Subscribe for free to receive new posts
Sources and methodology
[1] ETF total returns, through November 13, 2026: FinanceCharts trailing CAGR data for SOXX, IGV and SPY. FDN comes from ChartRow / Twelve Data: +1,083.9% cumulative total return over twenty years and +242.3% over ten. The earlier decade is derived from the ratio of those two growth factors. The November 2016 midpoint was not a trading day; displayed figures are approximate. FinanceCharts’ FDN series differs slightly.
[2] ETF mandates: iShares SOXX, First Trust FDN, iShares IGV.
[3] Cloud concentration: Synergy Research Group, Q3 2026, share of worldwide enterprise spending on cloud infrastructure services.
[4] Forward P/E and trailing P/FCF: Stock Analysis company statistics, November 13 closing-value snapshot, accessed November 15: Nvidia, TSMC, Broadcom, Microsoft, Alphabet, Amazon, RELX, Experian, S&P Global, Constellation Software, Intuit, ServiceNow.
[5] Forecast FCF: Simply Wall St, consensus forecasts attributed to S&P Global Market Intelligence, accessed November 15: Nvidia, TSMC, Broadcom, Microsoft, Alphabet, Amazon, RELX, Experian, S&P Global, Constellation Software, Intuit, ServiceNow.
[6] Company disclosures: Microsoft FY2027 Q1 earnings call; Constellation Software Q3 2026; Amazon Q3 2026 earnings.
[7] FX, ADS and calculation basis: ECB reference rates, Taiwan central bank, TSMC ADS ratio (five ordinary shares per ADS). Market caps are rounded, so the last 0.1× of a multiple is not precise. Companion workbook: Download the Excel calculations here below