Goldman Sachs: AI Risks Fall into Two Categories — Only Aggregate Shocks Hurt the Broad Market
Miles Bennett
Goldman Sachs splits AI investment risk into aggregate shocks and distribution shocks, arguing that confusing the two is the single biggest misjudgment in AI trading today — only aggregate shocks drag down the broad market, while distribution shocks merely reshuffle value within the AI chain.
What are "aggregate shocks" and "distribution shocks"?
Aggregate shock: the market starts doubting how much value AI can create *in total*. This means → the entire AI narrative compresses, and the broad market falls with it.
Distribution shock: no one doubts AI has value, but the market recalculates which companies capture it. In plain terms = the pie stays the same size; only the slicing changes, so indices barely move.
Goldman says confusing these two is the biggest source of error in current AI positioning — and it leads investors to reach for the wrong hedge.
What do these two shocks look like in practice?
Aggregate shock example: on the day DeepSeek R1 launched, the S&P 500 fell 1.5%, Nasdaq 100 fell 3.0%, Goldman's broad AI basket dropped 9.5%, and the VIX surged 20.5% — broad market, AI stocks, and volatility all moved in unison.
When hyperscaler earnings raised doubts about capex sustainability, the pattern was nearly identical: S&P down 1.7%, Nasdaq 100 down 3.1%, AI basket down 6.0%. This reflects the market questioning AI's overall return, not any single company.
Distribution shock example: when Google announced an equity offering to fund AI capex, hyperscaler stocks fell but semis and AI supply-chain names *rose* — the S&P 500 barely moved. This means → value simply relocated within the AI chain.
Why do macro hedges fail against distribution shocks?
Buying Treasuries, the dollar, or index volatility — classic macro hedges — only works when the shock is aggregate, because an aggregate shock pulls the whole market lower.
Distribution shocks are zero-sum inside the chain: if hyperscalers cut capex, their cash flow improves but semiconductor suppliers suffer; if memory prices fall, memory makers lose but cloud and server buyers gain. In plain terms = the seesaw cancels out at the index level.
Goldman's advice: investors concentrated in specific AI sub-sectors need structural pair trades — long suppliers vs. short hyperscalers, or long chips vs. short software — rather than simple index protection.
How optimistic are current AI valuations?
Goldman estimates the present value of future US corporate AI earnings at roughly $9 trillion under its base case.
At a 10% discount rate, that figure rises to $22 trillion; at 20%, it drops to $5 trillion.
Key variables: the pace of productivity gains, the corporate share of AI value, adoption speed, and the non-US share of earnings. This means → a surprise in any single variable could trigger an aggregate shock.
What is the volatility structure telling us?
Over the past year, single-stock implied volatility has risen sharply, yet implied correlation among stocks has fallen to low levels.
In plain terms = the market is not betting on "AI collapses as a whole" — it is fiercely debating who wins and who loses *within* the AI supply chain.
Index-level calm may be masking violent rotation underneath. This reflects a market whose dominant tension is distribution, not aggregate.
What is the first question to ask when new AI news breaks?
Goldman's core takeaway: every time a new piece of AI information lands, the first question an investor must answer is — does this compress AI's total value, or does it only change how that value is divided among companies?
This means → if it is an aggregate shock, index and macro tools can protect you; if it is a distribution shock, you need precise long-short pairing within the AI chain.
That single judgment call determines which hedge to reach for and which way to adjust your book.
Content is for reference only, not financial advice.