Two AI Shocks, Two Market Reactions
What DeepSeek and Claude really did to financial markets — what sold off, why, and why the recovery looked different.
Why these two moments mattered
The last year delivered two very different AI shockwaves. Both were sharp, both triggered “sell first, ask questions later.” But the breadth, depth, and recovery profile told two different stories.
Claude questioned the price of software outcomes (SaaS + data services pricing power).
Translation: DeepSeek shook the “compute complex.” Claude shook the “white-collar workflow stack.”
Part 1 — The DeepSeek shock
When DeepSeek’s reasoning model arrived, the market reaction wasn’t about capability alone — it was about economics. Investors suddenly asked: “If frontier reasoning can be built and run more efficiently, did we overpay for scarcity in AI compute?”
What the market hit first (DeepSeek)
| Segment | Why it mattered | Tickers most associated with the shock |
|---|---|---|
| AI chips | Valuation tied to “compute scarcity” and capex intensity | NVDA AMD AVGO MU |
| Data center infrastructure | Power/cooling/capex sensitivity to AI build-outs | VRT |
| Semiconductor equipment | Long-cycle capex expectations and “build more fabs” narratives | ASML |
Why the recovery was fast
This shock was mostly a discount-rate event on future profitability. Earnings didn’t change overnight — expectations did. Markets often digest those faster because the path to “truth” runs through subsequent spending plans and earnings prints.
Part 2 — The Claude shock
Claude’s recent wave was unsettling for a different reason. It wasn’t framed as “better chat.” It was framed as AI agents executing workflows — legal research, financial analysis, contract review, and synthesis tasks that people and software suites have historically monetized.
What the market hit first (Claude)
| Sector | Why it mattered | Tickers most associated with the shock |
|---|---|---|
| Legal & publishing data | AI threatens value capture in research-heavy workflows and subscription pricing | RELX TRI WKL.AS PSON.L |
| Financial data & analytics | Concern over “cash-cow data” moats and price-per-seat models | FDS MORN SPGI LSEG.L |
| Enterprise software | Seat-based SaaS challenged by task-based agent pricing | CRM ADBE ADSK ORCL |
| Services adjacency | Workflow disintermediation risk in regulated or templated tasks | EXPN.L LZ |
Why the selloff lingered
Claude’s shock maps more directly to business model debates: who owns the workflow, who owns proprietary data, and whether customers continue paying “per seat” when an agent can deliver the outcome “per task.” Because that impacts pricing power and margins, the market typically demands more proof before fully reversing.
Side-by-side comparison
| Dimension | DeepSeek | Claude |
|---|---|---|
| Shock type | Supply-side (compute economics) | Demand-side (workflow monetization) |
| Sector breadth | Narrower (AI infrastructure complex) | Wider (software, data, workflow services) |
| Market psychology | “Reprice the future” | “Reprice the business model” |
| Recovery profile | Often faster snap-back, then differentiation | Often slower, debate-driven repricing |
DeepSeek — impacted ticker cluster
Theme: compute scarcity, capex intensity, infrastructure leverage.
Claude — impacted ticker cluster
Theme: workflow disintermediation, pricing model risk, moat debates.
The “hidden similarity” most people missed
Despite different triggers, both shocks share one truth: AI compresses value chains faster than markets expect.
- DeepSeek compressed the idea of compute scarcity.
- Claude compressed the idea of workflow monetization.
In both cases, the opportunity shows up after the first panic: markets start separating companies that can absorb and integrate AI from companies that sell a layer likely to be commoditized.