Learn

⌂Dashboard◈Learn

Practice

⌁Charts◷Replay↻Review

My learning

▥Stats☆Bookmarks⌕Search✦AI

Learning principle

Understand risk before practising decisions.

Trade ButyFree · Neutral
👤 Log in
📚Learn📈Markets⏮Replay✎Review🔍Search🤖AI👤 Log in
Trade Buty

A free & neutral trading education platform for Chinese speakers worldwide. Structured courses (learn) × live charts & replay (practice).

⚠️ Risk notice: All content is for study and research only and does not constitute investment advice. Markets are risky.

Navigate

LearnMarketsReplaySearchAIStatsPrivacy PolicyContent from kline-butyFeedback
© 2026 sun1090 · MIT LicenseContent from kline-buty

On this page

  • 1. What Is an Industry Chain
  • The upstream–midstream–downstream value chain structure
  • Profit flows along the value chain
  • 2. Three Classic Industry Chains Dissected
  • The smartphone chain
  • The EV chain
  • The semiconductor chain
  • 3. Profit Distribution Along the Chain: The Smile Curve
  • What the smile curve is
  • Why design, brand, and chips earn more
  • Why assembly earns less
  • 4. Investment Logic for Each Segment
  • Upstream: watch price and supply
  • Midstream: watch capacity and cost
  • Downstream: watch demand and brand
  • 5. The "Sell Shovels" Logic
  • What selling shovels means
  • Typical "shovel" segments
  • 6. Industry Chain Research Methods
  • Method 1: Find the bottleneck
  • Method 2: Track price transmission
  • Method 3: Reverse-engineer the landscape from related-party transactions
  • Method 4: Draw the industry chain map (template)
  • 7. Hands-on Exercise: The AI Compute Industry Chain
  • Step 1: Draw the map
  • Step 2: Find the bottleneck
  • Step 3: Judge transmission and prosperity
  • Step 4: Reach an actionable conclusion (example)
  • ⚠️ Risk Warning

Chapter progress

19 · Industry Research

Before you understand a company, first understand the industry it operates in.

0/5 lessons0%

Next chapter →

20 · Classic Reading List (Reading Guide)→

The first 19 chapters of this knowledge base have already explained the "methods" thoroughly: how to read candlesticks,

Learn/19 · Industry Research
Lesson 02/2 / 5 lessons

02 · Value Chain Analysis

An industry is not a monolith but a value chain with upstream/downstream division of labor: the upstream sells raw materials, the midstream manufactures, and the downstream builds brands and channels

📖 ~12 min read
On this page▾
  • 1. What Is an Industry Chain
  • The upstream–midstream–downstream value chain structure
  • Profit flows along the value chain
  • 2. Three Classic Industry Chains Dissected
  • The smartphone chain
  • The EV chain
  • The semiconductor chain
  • 3. Profit Distribution Along the Chain: The Smile Curve
  • What the smile curve is
  • Why design, brand, and chips earn more
  • Why assembly earns less
  • 4. Investment Logic for Each Segment
  • Upstream: watch price and supply
  • Midstream: watch capacity and cost
  • Downstream: watch demand and brand
  • 5. The "Sell Shovels" Logic
  • What selling shovels means
  • Typical "shovel" segments
  • 6. Industry Chain Research Methods
  • Method 1: Find the bottleneck
  • Method 2: Track price transmission
  • Method 3: Reverse-engineer the landscape from related-party transactions
  • Method 4: Draw the industry chain map (template)
  • 7. Hands-on Exercise: The AI Compute Industry Chain
  • Step 1: Draw the map
  • Step 2: Find the bottleneck
  • Step 3: Judge transmission and prosperity
  • Step 4: Reach an actionable conclusion (example)
  • ⚠️ Risk Warning

An industry is not a monolith but a value chain divided among upstream, midstream, and downstream: raw materials at the top, manufacturing in the middle, brands and channels at the bottom. Profit is not evenly distributed — within the same industry, different segments differ wildly in their ability to earn. This article covers industry chain structure, the smile curve of profit distribution, the investment logic for each segment, the "sell shovels" logic, and industry chain research methods, closing with a full hands-on walkthrough of the AI compute chain.


1. What Is an Industry Chain

The upstream–midstream–downstream value chain structure

SegmentWhat It DoesTypical TraitsSource of Profit
UpstreamResources, raw materials, basic componentsCapital-heavy, strongly cyclical, supply sets priceResource endowment, supply-demand gaps
MidstreamManufacturing, processing, assembly, contract productionCapital-heavy, fierce competition, utilization-dependentScale and cost, process barriers
DownstreamBrands, channels, end products, servicesAsset-light, close to consumersBrand premium, channel capability, demand insight

Three questions to locate a company in the chain:

  • How many steps from your product to the final consumer? (The farther away, the more upstream)
  • Who holds pricing power? (Who decides what the end product sells for and what the raw material sells for)
  • Where does added value occur? (Which segment earns above-average returns)

Profit flows along the value chain

Industry chain research has one core proposition: where does money come from, who holds it now, and where will it flow next. In an upswing, profit usually lands first on the segment with the tightest capacity; in a downturn it stays with whoever has the strongest bargaining power (closest to demand or holding a quasi-monopoly). Researching an industry chain means tracking where profit flows.


2. Three Classic Industry Chains Dissected

The smartphone chain

SegmentRepresentative ActivitiesProfit LevelLandscape Traits
UpstreamChips (SoC/memory/displays), optical lenses, CMOS sensorsHighHighly monopolized; head players take the lion's share
MidstreamMainboards, batteries, structural parts, whole-device assemblyLow-to-midFierce competition; assembly net margins in single digits
DownstreamBrands (Apple/Huawei/Xiaomi), channels, operating systemsHighBrand concentration; top brands capture most industry profit

📖 The classic phenomenon of the Apple supply chain

In the Apple supply chain, whole-device assemblers (e.g., Foxconn) earn hard-earned money, while suppliers of core components — lenses, chips, displays — enjoy far higher margins than assembly: the added value sits not in assembly but in design and core components.

The EV chain

SegmentRepresentative Player TypesCurrent Profit Traits
UpstreamLithium mines, cobalt/nickel, cathode/anode/electrolyte/separatorViolent price cycles: windfall profits in 2021-2022, then retreat after overcapacity set in
MidstreamBattery makers, motors & controls, vehicle manufacturingBatteries highly concentrated (CATL/BYD duopoly); vehicle assembly fiercely competitive
DownstreamBrand automakers, charging networks, mobility servicesBrand divergence; intelligence, channels, and after-sales are the new profit battlegrounds

The semiconductor chain

SegmentContentBarrier Traits
UpstreamEDA software, semiconductor equipment, photoresist/wafers and other materialsHighest barriers; chokepoint segments; long qualification cycles
MidstreamWafer fabrication, packaging & testingMassive capex; advanced nodes run by a duopoly
DownstreamChip design (Fabless), end applicationsDesign firms are asset-light with high gross margins, but depend on foundries and IP licensing

💡 One rule common to all three chains

The shared pattern: the further upstream, the more monopolized; the further downstream, the more fragmented (except consumer-facing terminal brands). For any industry chain, draw this structure first, then map profit onto it — get the structure right and half the analysis is done.


3. Profit Distribution Along the Chain: The Smile Curve

What the smile curve is

Acer founder Stan Shih proposed that value-added along an industry chain traces a "smile"-shaped curve — both ends (R&D/design, branding/marketing) are high-value; the middle (manufacturing/assembly) is lowest.

Smile curve: value-added peaks at both ends, troughs at manufacturing/assembly

Why design, brand, and chips earn more

SegmentWhy It Earns MoreCase Traits
Chips / IP / designPatents form quasi-monopolies with near-zero marginal cost — selling one unit costs about the same as selling 100 millionNVIDIA GPU gross margins consistently above 60%
BrandsA brand premium is the trust cost consumers willingly pay extra — and it's nearly impossible to copyMoutai's gross margin hovers around 90%
Channels / retailThey own the consumer entry point and extract payment terms and rebates from upstreamTop retailers' bargaining power over suppliers

Why assembly earns less

  • Low technical barrier: differences in assembly craftsmanship are hard to sustain; substitutability is high.
  • Fully contested: anyone can do it, so price wars inevitably compress margins.
  • Squeezed from both ends: upstream core components raise prices while downstream brands push them down; assembly passively absorbs both.
  • Real-world picture: top contract manufacturers' net margins sit around 3%-6%, while brand and core-component players often earn several times that.

Using the smile curve for investment judgment: at comparable quality, companies closer to the "corners" of the curve (design/brand ends) carry economic moat premiums; those closer to the "chin" (manufacturing/assembly) depend on scale and efficiency. Exceptions exist — if the manufacturing side builds unique process barriers (precision manufacturing, proprietary materials), it can climb off the bottom of the curve.

💀 Iron rule: contract manufacturing earns hard-won money, not an economic moat

At comparable quality, companies nearer the "corners" (design/brand) carry moat premiums; those near the "chin" (assembly) rely on scale and efficiency. Top contract manufacturers net 3%-6% while brands and component makers often earn multiples of that — so "big scale" ≠ "big profits." In any value chain, first ask which segment captures the profit.


4. Investment Logic for Each Segment

Upstream: watch price and supply

Watch PointContent
Core variableProduct prices (spot/futures/contract prices)
Supply sideTiming of new capacity, mine/line build-out cycles, inventories
Demand sideDownstream operating rates, demand growth
Typical logicSupply contraction (shutdowns, output curbs, mine accidents) + demand recovery = upward price elasticity
RiskPrice cuts both ways: windfall profits in an upcycle, but once supply is released, price and profit collapse together

The essence of upstream investing is betting on the price cycle (see Article 04): buy when losses force capacity out, sell when windfall profits trigger expansion. For upstream companies, PE is a trap — product prices and spreads are the anchor.

Midstream: watch capacity and cost

Watch PointContent
Core variableCapacity utilization, unit costs, expansion plans
Key questionIs industry capacity excessive? Are many new entrants coming?
Competitive strategyVertical integration to cut costs, economies of scale, process leadership
Typical logicHigh utilization + stable landscape = volume and price rise together; overcapacity + price war = margins shaved
RiskThe midstream most easily shows "revenue growth without profit growth" — rising revenue cannot mask falling gross margins

Downstream: watch demand and brand

Watch PointContent
Core variableEnd sales volumes, penetration rate, brand share, channel inventory
Key questionIs demand a real breakout or short-term stimulus from subsidies/discounting?
Competitive strategyBrand premium, channel density, repurchase rates and user stickiness
Typical logicRising demand + brand concentration = volume-price double gain, share and profit rising together
RiskSales data distorts easily under promotions and channel stuffing; "sell-through" is truer than "shipments"

Quick-reference table across the three segments:

SegmentLogic in One LineKey DataKey Risk
UpstreamPrice and supplyPrice, inventory, outputPrice reversal
MidstreamCapacity and costUtilization, spreads, expansionsOvercapacity
DownstreamDemand and brandSales, penetration, shareDemand falsified

5. The "Sell Shovels" Logic

What selling shovels means

In a gold rush, the steadiest money is not made by prospectors but by those selling shovels, water, and jeans — no matter who strikes gold, the toolmaker gets paid first.

MappingGold RushModern Industries
ProspectorsMinersApplication/device makers (AI apps, carmakers, game studios)
Shovel sellersToolmakersCompute equipment, semiconductor equipment, battery equipment, test instruments, materials suppliers
TraitWinner takes allWinners still take all, but toolmakers don't bet on any single player

Three essentials of the shovel-seller logic:

  1. Don't bet on winners: you don't need to pick which AI application wins — as long as "everyone needs compute," compute-equipment and materials sellers benefit.
  2. Prosperity transmits early: when an industry takes off, capex hits equipment and materials first, so shovel sellers book orders earliest.
  3. But shovels also become oversupplied: after every capex frenzy, equipment and materials face overcapacity too — the shovel seller merely defers risk rather than eliminating it.

✅ Takeaway: the shovel seller defers risk rather than eliminating it

Shovel sellers only postpone risk; they don't remove it. After every capital-spending frenzy, equipment and materials face the same overcapacity — so "sell shovels" is not a sure-win grail but deferred gains ("earn first, give back later"). When prosperity ebbs, toolmakers get crushed by overcapacity just like everyone else.

Typical "shovel" segments

IndustryShovels
SemiconductorsLithography/etching equipment, photoresist/wafer materials
New energyBattery-manufacturing equipment, solar PV equipment (expansion phase), inverters
AI computeGPU/AI chips, HBM memory, optical modules, liquid cooling, servers, data-center power and infrastructure
Innovative drugsCXO (R&D outsourcing), lab instruments, consumables

6. Industry Chain Research Methods

Method 1: Find the bottleneck

Every industry chain has a segment that acts as the "bottleneck" — everyone else waits for its capacity while it sets prices. Bottleneck = highest-margin segment = strongest bargaining power.

Criteria for identifying a bottleneck:

  • A persistent supply-demand gap (utilization stays elevated);
  • High technology/certification barriers (qualification cycles start at 2-3 years);
  • Long expansion cycles (building plus ramp-up takes 2+ years);
  • Customers cannot route around it (no substitute exists).

✅ Takeaway: bottleneck = highest margin = strongest bargaining power

The bottleneck is the highest-margin, most powerful segment. Every chain has a link others wait on — find the bottleneck first, then see which companies occupy it. That is step one, more important than looking at market caps.

Method 2: Track price transmission

  • PPI and the cost-transmission chain: upstream input prices rise → midstream costs climb → midstream raises its prices → downstream end prices rise. Tracking the "spread" (product price − input cost) beats tracking single prices.
  • Watch how smoothly transmission flows: stable midstream gross margins = smooth pass-through; compressed margins = blocked transmission, hurting midstream profits.
  • Transmission lags: upstream hikes typically reach downstream pricing 1-3 quarters later — the "profit vacuum" during transmission is precisely the forecasting opportunity in industry chain research.

Method 3: Reverse-engineer the landscape from related-party transactions

  • Check leaders' purchasing/sales counterparties: concentration of top customers and suppliers reveals concentration and bargaining relations up and down the chain.
  • Check related-party transactions and receivables: segments with ballooning receivables are usually the ones whose customers occupy their funds (weak bargaining power).
  • Watch cross-shareholdings and strategic alliances among heads: chain alliances hint at technology routes and lock-in structures.

Method 4: Draw the industry chain map (template)

text
Upstream materials/components ──▶ Midstream manufacturing/integration ──▶ Downstream brands/end products ──▶ End demand
   ▲                    ▲                    ▲
Supply concentration   Capacity utilization   Channel inventory
Price & inventory      Spread & gross margin   Sales & penetration

Annotate each segment: representative companies, concentration, current prosperity, profit trend. One map plus four rows of notes is an updatable working draft of the chain.


7. Hands-on Exercise: The AI Compute Industry Chain

Walk through the full method from Sections 1-6 (illustrative/fictional data):

Step 1: Draw the map

LayerSegmentRepresentative ParticipantsConcentrationBargaining Power
UpstreamAI chips (GPU/ASIC), HBM memory, advanced-node foundry servicesLeading chip designers, memory makers, wafer fabsVery high (oligopoly)Very strong
UpstreamOptical modules, servers, liquid cooling, power equipmentLeading telecom/server vendorsMediumMedium
MidstreamIDC/AI data center construction & operationsTelecom operators, third-party IDCs, cloud providersFragmentedWeak-to-mid
DownstreamLLM training, inference applications, agentsCloud providers, AI application companiesFragmentedWeak (unsettled)

Step 2: Find the bottleneck

  • Bottleneck 1: advanced AI chips and HBM memory — supply far below demand, long expansion cycles, strongest bargaining power.
  • Bottleneck 2: advanced-node capacity — only a handful of foundries worldwide can produce it; a physical bottleneck.
  • Conclusion: the fattest, most certain profits sit in chips/memory/advanced nodes, not in the application layer.

Step 3: Judge transmission and prosperity

  • Cloud providers' capex is the key leading indicator: raised capex guidance → optical module/server orders → data center construction → power and cooling infrastructure, transmitting down the chain stage by stage.
  • Assumed current position: compute demand is still exploding, but watch for capex peaking signals — once big-tech capex guidance turns, shovel sellers' order growth slows first.

Step 4: Reach an actionable conclusion (example)

  • Main thesis ranking: bottleneck segments (chips/memory) > elastic segments (optical modules/liquid cooling) > lagging segments (power infrastructure).
  • Risk list: technology-route switches (in-house ASIC displacing GPU), concentrated capacity releases, capex cycle peaking.
  • Update cadence: track big-tech capex guidance, GPU delivery lead times, and memory spot prices monthly.

💡 Exercise requirement: draw a complete industry chain yourself

Requirement: apply the same four-step method to another chain of your choice (humanoid robots, innovative drugs, low-altitude economy). If you can't draw the map, you haven't gathered enough material; if you can draw it but can't explain where the money flows, your analysis isn't there yet.


⚠️ Risk Warning

⚠️ Risk Warning

Industry chain analysis delivers a "structural verdict," but structure gets broken dynamically — technology-route switches (e.g., in-house chips replacing purchased ones), geopolitics and export controls, and capacity-release timing can instantly redistribute profit. The fattest bottleneck segments also tend to carry the richest valuations: picking the right segment but buying at the top still loses money. And "selling shovels" is not "guaranteed profit" — shovels suffer overcapacity too. This is educational methodology content, not investment advice; every chain conclusion must be dynamically verified against high-frequency data such as prices and orders.

📝 行业研究篇 · 随堂测

3 concept questions · instant grading

📖 Done reading? See the real market

Find the concepts from this lesson on the live chart — understand before you continue.

Open live chart →
🤖Ask AI: 02 · Value Chain Analysis→

Related lessons

  • →01 · Industry Research Methodology
  • →03 · Competitive Landscape & Economic Moats
  • →04 · Sector Prosperity & Cycles
  • →05 · New Tracks & Theme Investing

Next

03 · Competitive Landscape & Economic Moats

→