Howard Marks: Investing in AI Stocks Is ‘Closer to Speculating’ Than Analysis

The post Howard Marks: Investing in AI Stocks Is ‘Closer to Speculating’ Than Analysis appeared first on 24/7 Wall St..

Although the AI buildout has minted trillions in fresh market value across Wall Street, Howard Marks thinks buyers of these stocks are kidding themselves about what they actually own. On a recent Prof G Markets appearance, Oaktree Capital co-founder Marks laid out a spectrum running from “analytical investing in prosaic, understandable companies” to “speculative investing in futuristic companies that can’t be described at all.” Most of today’s AI darlings, he argued, sit much closer to the speculative end than buyers want to admit.

Speculation, in his telling, is forecasting without honestly accounting for the probability that your forecast is wrong. Analysis is grounded in cash flow you can actually model. For more context on how this cycle compares with prior buildouts, see our earlier piece on the AI capex boom and its historical parallels.

What’s particularly notable is the valuation backdrop he is working against. Marks pointed out that the Shiller CAPE ratio is near 42, close to its dot-com peak of 44, while the standard S&P 500 PE sits around 23 versus an 80-year average of 16. The benchmark S&P 500 is not priced for disappointment, and the technology-packed Nasdaq Composite is leaning harder on a single thesis than at any moment since 1999.

Marks’s Risk Ladder, Applied

Marks named names. The lower-risk way to own the AI theme runs through the hyperscalers. Amazon, Google, Meta, and Microsoft have “established businesses with moats, enormous operating cash flow” that fund the buildout without betting the company. One layer up the risk ladder sit names like Anthropic and Nvidia, which Marks believes have “a high probability of still being successful 5 or 10 years from now.” At the top sit private AI startups, which Marks compared bluntly to lottery tickets: “most people who buy lottery tickets lose all their money. A few people become incredibly rich.”

The host’s pushback was fair. Companies like OpenAI and Anthropic burn cash and still command enormous valuations because revenue is compounding. Does profitability even matter? Marks’s answer was a thought experiment: ask anyone to name Anthropic’s net earnings in 2036, and “I’ll bet them that they’re not within 50% of the truth.” If you cannot model the cash flows within a country mile, you are guessing.

That distinction matters because the numbers funding this thesis are real. NVIDIA (NASDAQ:NVDA) just printed $81.6 billion in Q1 FY27 revenue, up 85% year over year, with data center networking alone growing 199%. CEO Jensen Huang called it “the largest infrastructure expansion in human history.” Nvidia carries $119 billion in supply-related commitments against that thesis, which the company puts at a 31x trailing PE and 23x forward. The shares trade at $205.19, up 42% over the past year.

The Hyperscaler Bill Is Roughly $600 Billion

The capex Marks is implicitly skeptical of has reached eye-watering scale. Amazon (NASDAQ:AMZN) plans roughly $200 billion in 2026 capex, which has compressed trailing free cash flow to a thin trickle and pushed the stock to a forward PE near 31. Alphabet (NASDAQ:GOOGL) guided $175 billion to $185 billion in 2026 capex, and Meta Platforms (NASDAQ:META) raised its own range to $125 billion to $145 billion while absorbing a $4.03 billion Reality Labs operating loss in a single quarter. Microsoft (NASDAQ:MSFT) is running an AI business at a $37 billion annual run rate, up 123% year over year, with a commercial RPO of $627 billion backing the thesis.

Combined hyperscaler 2026 AI spending sits north of $600 billion. The numbers are real. So is the moat. Disclosure: I own Alphabet, Meta, and Nvidia. I read Marks’s most recent filings with the same care I give his memos, and Google’s 16x earnings still looks like the cheapest seat in the cohort. Our prior coverage of Alphabet’s valuation gap walks through why that discount has persisted.

The Mirror From 1999

History tells us what happens when the cash-flow gap between leaders and aspirants gets ignored. Cisco peaked at roughly 150 times earnings in March 2000, and the Nasdaq Composite lost about 78% peak to trough by late 2002. The infrastructure thesis was correct. Fiber did transform the economy. The stocks still got cut in half, and then in half again, before the survivors compounded for the next twenty years.

The market is not asleep to this. The VIX has climbed to 19.44, the 75th percentile of its 12-month range, and the 10-year minus 2-year Treasury spread has compressed from 0.74% in February to 0.39%. Meta and Microsoft are down 18% over the past year, even as Alphabet has doubled. The herd is no longer moving as one.

The Takeaway

Marks’s larger point is that uncertainty is not a reason to avoid investing. Long term, Wall Street still heads higher in the decades to come, and the hyperscalers may well earn back every dollar of this capex with interest. But buying Anthropic on a tweet, or sizing Nvidia like it is a Treasury bond, is a different activity than the one Benjamin Graham described. It is closer to speculating. Call it what it is, size it accordingly, and you can still play.

The post Howard Marks: Investing in AI Stocks Is ‘Closer to Speculating’ Than Analysis appeared first on 24/7 Wall St..