What are the key sales KPIs for the Semiconductor Foundry industry in 2027?

The outcome you should expect

When these nine metrics are instrumented and reviewed on a fixed cadence, the foundry sales function stops behaving like an order-taking desk and starts behaving like a capacity-allocation business. That shift is the outcome, and it shows up in four concrete ways within two to three quarters.

First, allocation conversations move eighteen to twenty-four months forward. A foundry that only tracks booked wafer revenue is negotiating against a quarter that is already fixed — the fab schedule was locked when the equipment was ordered. A foundry that tracks tape-out count alongside wafer revenue is negotiating against the design starts that will become revenue in 2029. Leading-edge capacity at the top of the industry has repeatedly been described as effectively sold out a year or more ahead; if your sales team is discovering demand in the same quarter it ships, the metric set is wrong.

Second, pricing stops being blended. Blended ASP across a mixed node portfolio is one of the most misleading numbers in the industry, because a foundry running 28nm analog alongside 3nm HPC has two businesses whose per-wafer prices differ by roughly an order of magnitude. Leading-edge 300mm wafers at the most advanced nodes have been reported in the tens of thousands of dollars each, while mature 28nm-and-above wafers sit in the low single-digit thousands. Averaging them produces a number that moves whenever mix moves and tells you nothing about pricing power. Once ASP is tracked per node and per customer tier, sales can see which accounts are actually paying for the leading edge and which are riding a legacy price sheet.

Third, the depreciation math becomes visible to the people selling capacity. A leading-edge fab is a multi-billion-dollar asset depreciated over roughly five to seven years. The fixed cost per wafer is enormous and does not care whether the fab is full. Utilization is therefore not an operations metric that sales can ignore — it is the single largest determinant of whether a given wafer price is profitable. Sales teams that can see utilization by node stop discounting into a full fab and stop holding price in an empty one.

Fourth, customer concentration becomes a managed position rather than an accident. Pure-play foundries at the top of the market draw a large majority of revenue from their ten largest accounts, and the very top accounts — the large AI accelerator and smartphone SoC customers — can each represent a double-digit share on their own. That concentration is not inherently a defect; multi-year prepayments and capacity commitments from those accounts are precisely what makes a twenty-billion-dollar fab bankable. But it has to be measured, stress-tested, and re-forecast every quarter, because a single account reallocating a quarter of its volume changes the capital plan.

The realistic expectation is not that these metrics increase revenue directly. It is that they convert a manufacturing P&L into a sales-legible one, so that every allocation decision, price negotiation, and node commitment is made against the same numbers the CFO uses.

What drives that outcome

The mechanism is a chain: a design start becomes a tape-out, a tape-out becomes a node commitment, a node commitment becomes wafer starts, wafer starts become utilization, utilization determines whether the depreciation load is absorbed, and absorbed depreciation determines gross margin, which funds the next node. Every one of the nine metrics sits somewhere on that chain, which is why tracking them in isolation produces so little insight.

Tape-outs are the true leading indicator. A customer commits to a process node when it starts the design, not when it orders wafers. Design-to-volume lead times on a leading-edge node run roughly eighteen to twenty-four months. So the tape-out count on a given node this year is a fairly direct read on that node's wafer revenue two years out. A foundry watching only bookings is watching a lagging indicator of a decision made two years ago. Track tape-outs by node, by customer, and by platform — HPC, smartphone, automotive, IoT — because the platform mix determines the packaging load downstream.

Node choice sets the entire economic profile of the wafer. The same customer, on the same fab floor, generates radically different revenue depending on node. Leading-edge nodes carry a substantial per-wafer premium over the prior generation — the step from one leading-edge node to the next has historically carried a premium in the range of tens of percent, not single digits. Mature nodes carry a fraction of that price but also a fraction of the capital load and a much longer depreciation runway. Neither is inherently better; a mature-node specialty fab running RF-SOI or silicon photonics at high utilization can out-earn a leading-edge fab running at 65%.

Advanced packaging is now a second, coupled business. For AI accelerator customers, the wafer is only part of the deliverable. 2.5D and 3D packaging — chip-on-wafer-on-substrate and its successors — has been the binding constraint on AI compute supply, with lead times reported well beyond a year at points in the cycle. That means a sales organization can be sitting on available leading-edge wafer capacity and still be unable to serve a customer, because the packaging slot is gone. Advanced-packaging revenue therefore belongs on its own P&L line with its own capacity metric, not buried inside wafer revenue.

Utilization is where the chain either closes or breaks. Because so much of a leading-edge fab's cost is fixed depreciation, there is a utilization threshold below which the node loses money regardless of price. Above it, incremental wafers drop through to margin at a very high rate. This is why the industry's gross-margin spread is so wide: the leader running near-full leading-edge utilization can post margins in the high fifties, while a second-source foundry under-filling comparable capacity posts margins a wide double-digit gap below that, and a start-up foundry business with negligible external revenue runs deeply negative.

The diagram makes the loop explicit, and the loop is the point. Sales does not sit at the end of this chain receiving a price sheet — it sits at the top of it, because the tape-out commitments the sales team wins this year are the input to next year's capital plan.

Benchmarks and realistic ranges

Absolute numbers in this industry move fast, so the durable value is in the ranges and thresholds rather than any single quarter's figure. These are the bands a foundry sales organization should be measuring against in 2027.

Wafer revenue by node. Report it split, never blended. The industry has a very steep revenue distribution: the leading pure-play foundry operates at a revenue scale roughly an order of magnitude above the next-largest pure-play competitors, and the field below that clusters in the mid-single-digit to high-single-digit billions annually. A useful internal benchmark is the share of revenue growth attributable to node mix versus volume — if revenue grows but mix-driven growth is near zero, you are selling more wafers at flat prices, which is the profile of a mature-node business regardless of what the roadmap says.

Advanced-node mix. Share of wafer revenue from 7nm-and-below, 3nm-and-below, and 2nm-class. The leading foundry draws roughly three-quarters of wafer revenue from 7nm and below. A credible second-source leading-edge foundry sits materially lower, often in the neighborhood of a third to two-fifths. A deliberate specialty foundry may be at zero by design and be perfectly healthy — GlobalFoundries exited the leading-edge race in 2018 and built a differentiated FDX, RF-SOI, and silicon-photonics business instead. The diagnostic: if you claim to be a leading-edge foundry but sub-7nm mix is under 50%, you are a mature-node fab with a leading-edge press release.

Capacity utilization. Roughly 80% is the practical depreciation breakeven for a leading-edge fab; below about 70% the node reliably loses money. Healthy operating range is 85–95%. Above 95% sustained means you are turning away demand and should be pricing higher or building. Mature-node fabs tolerate lower floors because the equipment is largely depreciated. Track it by node and by fab, never company-wide — a blended 85% can hide a 98% leading-edge node subsidizing a 60% legacy line.

ASP per 300mm wafer equivalent. Normalize everything to a twelve-inch equivalent so 200mm capacity is comparable. Leading-edge wafers price in the tens of thousands of dollars; mature 28nm-and-above wafers price in the low thousands. The generation-over-generation premium at the leading edge has been substantial — reported node-to-node step-ups in the range of roughly 30% to 50%. The metric to watch is not the level but the trend per customer tier: if your largest account's ASP is flat while node mix richens, you are absorbing a mix shift without capturing its price.

Gross margin. The industry leader has been operating in the mid-to-high fifties. A second-source leading-edge foundry under-filling capacity can be in the single digits to low teens. A specialty mature-node player typically sits in the high twenties to low thirties. A foundry business still ramping external customers runs negative. The gap between the leader and the field is roughly thirty percentage points, and essentially all of it is explained by utilization times node mix.

CapEx intensity. Annual capital expenditure divided by annual revenue. The healthy band for a leading-edge foundry is 30–50%; the industry leader has guided to roughly 38–40%. Below 25% at a company claiming leading-edge ambitions means under-investment and a lost node two years out. Above 50% sustained without pre-sold capacity is the classic build-ahead-of-demand failure. For a mature-node specialty foundry, 15–25% is normal and healthy.

Advanced-packaging revenue. For foundries with a packaging business, this can contribute a mid-single-digit to mid-teens percentage of total revenue and has been growing faster than wafer revenue. Track capacity in wafers per month alongside revenue, plus committed allocation by customer, because the capacity number is what your sales team is actually selling.

Top-10 customer concentration. Pure-play leading-edge foundries commonly run 70–75% of revenue through the top ten accounts, with the single largest account in the high teens to low twenties. Specialty foundries often run top-three concentration above 40%. There is no "correct" number — the correct posture is knowing it, having a prepayment or capacity-commitment structure behind each large account, and modeling the P&L if any one of the top three reallocates 25% of volume.

Tape-out count and yield-ramp time. Count new tape-outs per node per quarter against the same node's cumulative target — leading nodes accumulate hundreds of tape-outs over their life, and the first-year count is the best available forecast of peak-year revenue. Yield-ramp time, measured from first silicon to a defect-density-corrected yield the customer will accept for volume, has run roughly eighteen to twenty-four months at 3nm-class nodes and is expected to be longer at 2nm-class gate-all-around nodes. A ramp that slips past thirty months is the single most reliable predictor of a thin external customer pipeline two years later.

Risks, edge cases, and failure modes

Four failure modes account for nearly every foundry that has lost a generation, and each one is visible in the metric set before it is visible in the P&L.

Node-yield slip. A leading-edge node that misses its yield curve by six months does not just lose six months of revenue — it loses the next design cycle, because customers make their next tape-out decision on observed ramp performance, not on roadmap promises. The industry has a clear example: a gate-all-around 3nm ramp that ran past thirty months coincided with a visibly thin external customer pipeline for that foundry. The early-warning metric is yield-ramp time against the published curve, tracked weekly with a stoplight on the three critical-path items — equipment install, process qualification, and customer PDK release. Sales needs this weekly, not quarterly, because the moment the ramp slips is the moment to renegotiate commitments rather than the moment the customer discovers it independently.

CapEx ahead of demand. Committing twenty-plus billion dollars to a fab twenty-four months before customers commit volume creates a depreciation cliff that lands whether or not the customers arrive. This is the structural risk facing any foundry building leading-edge capacity without a pre-sold book, and it is exactly why concentration and prepayment structures exist at the leader. The early-warning metric is the ratio of committed customer wafer demand to planned capacity for each build, tracked by node, by quarter, out twenty-four months. If that coverage ratio for a build two years out is under roughly half, the build should be phased rather than committed in full.

Customer-concentration unwind. Concentration funds CapEx right up until it doesn't. When the largest account changes — and it does change; the top account at the leading foundry shifted from the smartphone SoC customer to the AI accelerator customer as AI demand scaled — the entire allocation plan re-sorts. The failure mode is not losing an account outright; it is a top-three account moving 20–30% of volume to a second source to de-risk its own supply chain, which is a rational thing for them to do and a painful thing for your utilization. Stress-test this every quarter as a standing exercise, not as a crisis response.

Geopolitical capacity stranding. Export controls cap what some foundries can build regardless of demand, and offshore fabs built for supply-chain resilience carry a structural cost premium versus the home fab. A wafer produced in a higher-cost geography at the same price is a lower-margin wafer, and if customers will not pay the differential, that capacity is stranded at the margin level even at full utilization. The metric here is gross margin by fab, not just by node — if two fabs run the same node and one is ten points lower, that gap is either a cost-structure problem or a pricing problem, and sales owns the pricing half.

The edge cases that break the standard metric set. A specialty mature-node foundry should not be graded on advanced-node mix at all — grading it there produces a false negative on a healthy business. A foundry with a captive internal customer (an IDM running a foundry segment) has revenue and margin numbers that are not comparable to a pure-play, because internal transfer pricing distorts both ASP and gross margin; for those, external-customer revenue is the only honest metric. And a foundry ramping a genuinely new node has a legitimate period of negative incremental margin that is investment, not failure — the distinguishing test is whether tape-out count on that node is rising. Rising tape-outs plus negative margin is a ramp. Flat tape-outs plus negative margin is a stall.

A practical rollout plan

Instrumenting these metrics is a ninety-day project with a clear sequence, and the sequence matters — building dashboards before reconciling the underlying counts produces confident wrong numbers.

Days 1–30: reconcile the counts. Instrument all nine metrics end-to-end against the manufacturing execution system, ERP, and the customer-allocation system. The first real finding will be that wafer-out counts do not agree across manufacturing, finance, and customer-shipped reporting — they never do on the first pass, because each system has different cut-off rules, scrap treatment, and revenue-recognition timing. Quantify that gap and document its cause before building anything on top of it. In the same window, establish baseline gross margin per node and ASP per customer tier, and pull the last eight quarters of advanced-packaging allocation against customer demand so the size of the unserved gap is a number rather than an anecdote.

Days 31–60: build the two dashboards that matter. Ship the advanced-node mix dashboard and the CapEx-intensity dashboard. Wire CapEx on one side to the multi-year fab build schedule and on the other side to confirmed customer tape-out commitments, so the coverage ratio described above is a live number rather than a quarterly slide. Then identify the next-node yield-ramp critical path — equipment install dates, process qualification milestones, customer PDK release — and put a weekly stoplight on it against the published ramp curve. This is the artifact that lets sales renegotiate early instead of apologizing late.

Days 61–90: run the concentration review. Hold the first quarterly top-ten customer review as a formal exercise. Model each top-ten account's twenty-four-month wafer demand by node, then stress-test the plan against any one of the top three reallocating a quarter of its volume. Re-base the following year's CapEx against that demand model and take a revised operating plan to the CFO with explicit monthly checkpoints on utilization, gross margin, and advanced-packaging revenue.

The standing cadence after day 90. Daily: wafer starts by node, equipment availability, and yield-loss flags from in-line metrology. Weekly: the tape-out funnel, customer allocation requests, and advanced-packaging slot booking status. Monthly: wafer revenue by node and platform, capacity utilization by fab, ASP trends by customer, and a gross-margin walk that attributes change to price, mix, volume, and cost. Quarterly: the full P&L, CapEx pacing versus plan, advanced-node mix, top-ten concentration, packaging revenue split, and next-node yield-ramp milestones for the board. The daily and weekly layers exist so the monthly number is never a surprise; the quarterly layer exists so the capital plan is re-based against reality four times a year rather than once.

Related questions

How often should a foundry re-baseline its CapEx plan?

Quarterly, against confirmed customer tape-out commitments rather than internal demand forecasts. Fab builds have twenty-four-month lead times, so a plan re-based only annually is committing capital against demand assumptions that are already a year stale.

Should advanced packaging be a separate P&L?

Yes. It has its own capacity constraint, its own lead-time profile, and margin characteristics distinct from wafer fabrication. Burying it inside wafer revenue hides both the growth rate and the fact that packaging slots, not wafers, are frequently the binding constraint on AI accelerator supply.

Is high customer concentration a red flag for a foundry?

Not inherently. Multi-year commitments and prepayments from a small number of very large accounts are what make multi-billion-dollar fab investments financeable. The red flag is unmeasured concentration — no stress test, no prepayment structure, and no modeled scenario for a top-three account reallocating volume.

What utilization rate should a leading-edge fab target?

85–95%. Below roughly 80% the fixed depreciation load starts compressing margin materially, and below 70% the node loses money outright. Sustained operation above 95% means demand is being turned away and either price should rise or capacity should expand.

Why is blended ASP a misleading metric?

Because a foundry running mature and leading-edge nodes together has per-wafer prices differing by roughly an order of magnitude. Blended ASP moves whenever mix moves, so it reports mix shifts as if they were pricing changes. Track ASP per node and per customer tier instead.

FAQ

Which single metric matters most for a semiconductor foundry sales team in 2027?

Advanced-node mix — the share of wafer revenue from 7nm-and-below, 3nm-and-below, and 2nm-class processes. It is the metric most directly tied to per-wafer pricing power, and it determines whether leading-edge capacity utilization stays high enough to absorb the depreciation load. Its one caveat: it is the wrong grading metric for a deliberate specialty mature-node foundry, where high utilization and differentiated process offerings matter more than node leadership.

How does capacity utilization affect foundry profitability?

Because most of a leading-edge fab's cost is fixed depreciation on a multi-billion-dollar asset over roughly five to seven years, utilization is the dominant profitability lever. Roughly 80% is the practical breakeven; below about 70% the node loses money regardless of price. The healthy operating band is 85–95%. Always measure it per node and per fab, because a healthy blended figure can conceal a full leading-edge line subsidizing an empty legacy one.

Why does CapEx intensity belong in a sales metric set?

CapEx intensity — annual capital expenditure as a percentage of revenue — is what turns today's customer commitments into tomorrow's sellable capacity. The healthy band at the leading edge is 30–50%, with the industry leader guiding to roughly 38–40%. Below 25% signals under-investment that costs you a node two years out. Sales owns half of the equation because the tape-out commitments won this year are the demand side of next year's capital case.

What does ASP per 300mm wafer equivalent actually tell you?

It normalizes pricing across 200mm and 300mm capacity so mix shifts and price changes can be separated. Leading-edge wafers price in the tens of thousands of dollars each; mature 28nm-and-above wafers price in the low thousands, with node-to-node premiums at the leading edge reported in the 30–50% range. The signal to watch is the trend per customer tier — flat ASP against a richening node mix means you absorbed a mix shift without capturing its value.

How should advanced-packaging revenue be tracked?

As its own P&L line, with three attached figures: revenue, capacity in wafers per month, and committed allocation by customer. 2.5D and 3D packaging has repeatedly been the binding constraint on AI compute supply, with lead times stretching past a year at points in the cycle. A sales team can be sitting on free leading-edge wafer capacity and still be unable to serve a customer because the packaging slot is gone, which is why the two capacity pools must be reported separately.

What do tape-out count and node-yield ramp time predict?

Tape-outs are the leading indicator — design-to-volume lead times run roughly eighteen to twenty-four months, so this year's tape-out count on a node is the best available forecast of that node's revenue two years out. Yield-ramp time, from first silicon to a customer-acceptable defect-density-corrected yield, has run eighteen to twenty-four months at 3nm-class nodes and is expected longer at 2nm-class gate-all-around. A ramp slipping past thirty months reliably precedes a thin external customer pipeline.

Sources

• https://investor.tsmc.com/english/quarterly-results
• https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001709048&type=10-K
• https://www.semi.org/en/products-services/market-data/world-fab-forecast
• https://www.trendforce.com/presscenter/
• https://www.techinsights.com/
• https://www.anandtech.com/
• https://spectrum.ieee.org/topic/semiconductors/
• https://www.eetimes.com/
• https://www.semiconductors.org/data/
• https://www.bis.doc.gov/index.php/policy-guidance/advanced-computing-and-semiconductor-manufacturing-items

Related on PULSE

• [What are the key sales KPIs for the Fine-Tuning Platform industry in 2027?](/knowledge/ik0382)
• [What are the key sales KPIs for the Embeddings API industry in 2027?](/knowledge/ik0383)
• [What are the key sales KPIs for the equine breeding industry in 2027?](/knowledge/ik0451)