Edge AI silicon and endpoint compute covers the chips and local compute boards that let cameras, vehicles, robots, industrial controllers, wearables, and defense endpoints run sensing, control, and inference near the machine instead of sending every decision to a cloud data center. The stack includes system-on-chip processors, field-programmable gate arrays, microcontrollers, automotive ADAS processors, connectivity chips, software toolchains, and production programs at device makers and auto suppliers. The node converts into company economics when design wins become shipped units, reported quarterly revenue, durable gross margin, clean inventory, and cash conversion. AMBA ranks first because its low-power vision SoCs have the cleanest direct tie to edge perception and robotics; NXPI, QCOM, and LSCC add scaled auto, industrial, on-device AI, and low-power control exposure.
What the stack is: endpoint compute is the silicon layer inside the device, vehicle, camera, robot, sensor hub, industrial controller, or embedded board that reads data from sensors and runs local control or AI inference before data moves to a network or cloud service.
What it does in the parent chain: physical AI and defense autonomy need machines to sense, decide, and act under latency, power, bandwidth, privacy, and reliability constraints; this node supplies the processors, accelerators, FPGAs, microcontrollers, and connectivity chips that make those local decisions possible.
Main pieces: SoC means system-on-chip, a processor package with compute, memory controllers, image or signal processing, security, and interfaces on one chip; FPGA means a programmable chip used for low-latency control and custom logic; MCU means microcontroller, a low-power embedded controller; ADAS means advanced driver-assistance systems inside vehicles.
Where it sits: the chips sit on endpoint boards, camera modules, vehicle electronic control units, industrial gateways, robotics controllers, edge servers, or connected-device motherboards, usually close to image sensors, radar, lidar, motors, batteries, or network interfaces.
How the theme uses it: a robot or vehicle captures sensor data, endpoint silicon processes the signal, local software decides whether to steer, alert, classify, or transmit, and the supplier gets paid only when customer programs reach production shipments and revenue recognition.
Terms used later: design win means a customer selected the chip for a product program; production ramp means the product is shipping in volume; attach means the chip or software is included with a platform; channel sell-through means distributors are selling to end customers rather than only holding inventory.
Report boundary: this node is a tactical report layer. It ranks the current endpoint-compute basket, source-backed exposure, selected-security chart route, and source trail after the parent theme expanded the ticker universe. Durable research, claim IDs, raw source registries, and security-lane maintenance stay in the linked knowledge pages. Price and volume context comes from read-only discovery daily_ohlc through 2026-06-12.
Current Setup
Endpoint conversion gateLocal AI demand matters only when design wins become shipped chip revenue.
The best evidence is reported production revenue, end-market mix, gross margin, inventory conversion, and customer breadth. AMBA has the cleanest edge-vision fit; NXPI, QCOM, and LSCC bring scale but carry broader semiconductor exposures.
Positive proofAMBA, QCOM, NXPI, and LSCC already disclose relevant revenue.
Watch Q2 guide delivery, auto and IoT mix, Compute and Communications growth, and distributor sell-through.
Conversion gateProduction ramps must beat roadmap language.
Named customer programs, channel inventory, gross margin, and cash flow decide whether design wins are investable.
Primary constraintIndirect exposure and stale setup levels limit the read.
Several names move on handsets, analog restocking, financing, or broad semiconductor beta rather than endpoint AI proof.
Demand sourceCameras, vehicles, robots, industry
Chip functionSense, infer, control, connect
Collection routeDesign win to production
Proof pointRevenue, margin, cash
Physical AI needs local compute when latency, power draw, connectivity, reliability, or privacy makes a cloud-only inference path impractical. The node ranks companies by source-backed exposure to edge AI silicon, endpoint compute, and autonomy hardware before technical timing. AMBA leads because its edge vision AI silicon is directly tied to camera, security, robotics, telematics, ADAS, and edge-infrastructure ramps. NXPI, QCOM, and LSCC follow because they have scaled production exposure in automotive, industrial, endpoint platforms, secure connectivity, and low-power control silicon with better customer and disclosure support than most small-cap routes.
The current evidence is strongest where production revenue, guidance, or disclosed mix already connects to endpoint compute. Ambarella reported Q1 FY2027 revenue of $100.4 million and guided Q2 to $105 million to $111 million; NXP reported Q1 2026 revenue of $3.181 billion with Automotive and Industrial & IoT as major end markets; Qualcomm reported Q2 FY2026 QCT automotive revenue of $1.326 billion and IoT revenue of $1.726 billion; and Lattice reported Q1 2026 revenue of $170.9 million with 68.8% gross margin and record Compute and Communications revenue. The next useful proof is Q2 guide delivery, named production ramps, stable gross margin, clean channel sell-through, and evidence that design wins are becoming revenue rather than only roadmap language.
The weak points are concentration, indirect exposure, and valuation after large semiconductor moves. AMBA still has WT and Asia concentration, inventory and free-cash-flow pressure, and LTAs that should not be treated as backlog. AMBQ and BZAI have direct endpoint-AI narratives but short public history, limited 100-week chart coverage, customer concentration, and cash-burn or financing risk. STM, MCHP, and TXN are useful embedded-silicon comparables, but their current evidence is broad industrial, analog, MCU, data-center, or restocking recovery rather than pure edge-AI silicon conversion. Watch earnings guides, customer and distributor concentration, inventory days, AMI and Silicon Labs transaction terms, foundry and export-policy disclosures, and whether product claims become reported revenue with margin and cash conversion.
Selected-security right-rail charts call the report API with as_of=latest. Static setup thresholds from the prior 2026-06-05 page are stale because local daily_ohlc now runs through 2026-06-12; refresh setup levels before using them as current trading evidence.
Basket
This basket is hand-curated from the refreshed parent theme, the knowledge lanes, official company releases, read-only discovery market data, and the prior ranking work embedded in this node. Ranking uses source-backed node exposure and operating proof before technical timing. Core names have the clearest combination of endpoint-compute relevance, production scale, and disclosure. Option names have real node exposure but need cleaner conversion evidence. Watch and comparable names stay in the basket because the parent universe includes them, but their endpoint-AI evidence is either indirect, transition-heavy, or too speculative.
Direct low-power edge-vision SoC supplier tied to cameras, robotics, security, telematics, and vehicle perception.
Market cap$3.3B
Next earningsNot confirmed
Latest qtr revenue$100.4M
Role in stack
Ambarella sells edge-AI vision SoCs to camera, security, robotics, telematics, ADAS, and edge-infrastructure customers. The gate is whether CV and N-series programs, LTAs, and customer ramps become shipped SoC revenue with high gross margin and lower distributor concentration.
Revenue mix
One reportable segment. Management mix puts IoT at roughly three-fourths of Q1 FY2027 revenue and automotive at roughly one-fourth; the knowledge lane records edge AI as about 80% of FY2026 revenue.
Latest qtr revenue
Q1 FY2027 revenue was $100.4M for the quarter ended April 30, 2026, up 16.9% year over year; source is Ambarella's May 28, 2026 Q1 FY2027 earnings release and linked AMBA lane.
Scaled automotive and industrial embedded-control supplier with clean production revenue but broader exposure than pure edge AI.
Market cap$74.4B
Next earningsNot confirmed
Latest qtr revenue$3.181B
Role in stack
NXP sells embedded processors, secure connectivity, RF, automotive and industrial control silicon, and data-center control-plane chips to OEM, industrial, and infrastructure customers. Conversion depends on content per system, design-in durability, channel sell-through, margin, and free cash flow.
Revenue mix
Q1 2026 revenue was $1.782B Automotive, $628M Industrial & IoT, $391M Mobile, and $380M Communication Infrastructure & Other.
Latest qtr revenue
Q1 2026 revenue was $3.181B, up 12.2% year over year; source is NXP's April 28, 2026 Q1 results release and linked NXPI lane.
Scaled on-device AI, automotive, IoT, connectivity, and custom-silicon route funded by the handset and licensing base.
Market cap$230.9B
Next earningsNot confirmed
Latest qtr revenue$10.599B
Role in stack
Qualcomm sells Snapdragon, automotive, IoT, connectivity, on-device AI, and emerging custom-silicon platforms into handsets, vehicles, PCs, industrial edge devices, and infrastructure customers. Conversion depends on QCT automotive and IoT growth, on-device AI attach, named OEM launches, and data-center disclosure.
Revenue mix
Q2 FY2026 mix was QCT handsets $6.024B, automotive $1.326B, IoT $1.726B, and QTL $1.382B. Handset and QTL cash flows fund the broader endpoint platform.
Latest qtr revenue
Q2 FY2026 revenue was $10.599B; automotive grew 38% year over year and IoT grew 9%. Source is Qualcomm's April 29, 2026 Q2 FY2026 results release and linked QCOM lane.
Low-power FPGA and control-silicon supplier with strong Compute and Communications growth and distributor concentration risk.
Market cap$17.4B
Next earningsNot confirmed
Latest qtr revenue$170.9M
Role in stack
Lattice sells low-power FPGAs and control silicon into AI servers, embedded control, industrial, robotics, machine vision, and edge-AI attach points. The gate is Compute and Communications growth, high gross margin, distributor sell-through, and AMI transaction execution.
Revenue mix
Q1 2026 Compute and Communications revenue was $106.6M, or 62.4% of revenue; Industrial and Embedded was $64.3M. Distributor revenue was 94%, so sell-through evidence matters.
Latest qtr revenue
Q1 2026 revenue was $170.9M, up 42% year over year, with 68.8% gross margin; source is Lattice's May 4, 2026 Q1 results release and linked LSCC lane.
Vehicle endpoint-compute supplier where EyeQ normalization and advanced ADAS wins decide conversion quality.
Market cap$7.7B
Next earningsNot confirmed
Latest qtr revenue$558.0M
Role in stack
Mobileye supplies EyeQ ADAS and higher-content autonomy SoCs into OEM vehicle programs. Conversion depends on unit normalization, Surround ADAS and SuperVision awards, Chauffeur and Drive milestones, and adjusted operating income from vehicle endpoint compute.
Revenue mix
EyeQ was about 92% of Q1 2026 revenue, and the Mobileye segment represented nearly all revenue. Advanced autonomy product economics are not separately disclosed.
Latest qtr revenue
Q1 2026 revenue was $558.0M, up 27% year over year; source is Mobileye's April 23, 2026 Q1 results release and linked MBLY lane.
Ultra-low-power endpoint-AI SoC route with high purity and short public-company proof history.
Market cap$943.9M
Next earningsNot confirmed
Latest qtr revenue$25.1M
Role in stack
Ambiq sells ultra-low-power Apollo and Atomiq SoCs and endpoint-AI software into battery-constrained device makers. The gate is production designs, customer diversification, TSMC supply, mid-40s non-GAAP gross margin, and lower cash burn as revenue scales.
Revenue mix
The knowledge lane does not support a precise end-market split. Product evidence is Apollo and Atomiq, and the top three customers were about 71% of Q1 2026 sales.
Latest qtr revenue
Q1 2026 sales were $25.1M, up 59.3% year over year; source is Ambiq's May 12, 2026 Q1 results release and linked AMBQ lane.
Broad MCU, sensor, embedded, power, analog, and custom-processing comparable with indirect endpoint-AI proof.
Market cap$52.6B
Next earningsNot confirmed
Latest qtr revenue$3.095B
Role in stack
STMicro sells MCUs, custom processing, MEMS and sensors, RF/optical, power, analog, and embedded chips into endpoint, automotive, industrial, and infrastructure systems. Conversion depends on utilization, segment margin repair, and data-center or LEO program revenue rather than pure endpoint-AI proof.
Revenue mix
Broad IDM mix across Analog, MEMS and Sensors; Power and Discrete; Embedded Processing; and RF/Optical Communications, with end markets including auto, industrial, personal electronics, communications, computers, and peripherals.
Latest qtr revenue
Q1 2026 revenue was $3.095B, up 23% year over year; source is STMicroelectronics' April 23, 2026 Q1 release and linked STM lane.
Broad embedded-control and connectivity comparable that needs restocking to become durable sell-through.
Market cap$53.6B
Next earningsNot confirmed
Latest qtr revenue$1.311B
Role in stack
Microchip sells MCUs, analog, FPGA, timing, Ethernet, PCIe, security, memory, and embedded-control chips into industrial, automotive, data-center, defense, consumer, and communications endpoints. Conversion depends on broad embedded restocking and quantified AI/data-center connectivity revenue.
Revenue mix
FY2026 sales were 50.0% mixed-signal microcontrollers, 28.2% analog, and 21.8% other; distributors were 47% of sales.
Latest qtr revenue
Q4 FY2026 net sales were $1.311B for the quarter ended March 31, 2026; source is Microchip's May 7, 2026 Q4 and FY2026 results release and linked MCHP lane.
Analog and embedded-processing context comp where the direct endpoint-AI link is weaker than broad industrial and auto recovery.
Market cap$261.9B
Next earningsNot confirmed
Latest qtr revenue$4.825B
Role in stack
Texas Instruments sells analog power, signal-chain, and embedded processing that supports sensors, power, control, and endpoint electronics. Conversion depends on broad industrial and auto recovery, factory loading, and free cash flow per share.
Revenue mix
Q1 2026 revenue was $3.924B Analog, $723M Embedded Processing, and $178M Other; FY2025 industrial and automotive were each about 33% of revenue.
Latest qtr revenue
Q1 2026 revenue was $4.825B, up 19% year over year; source is TI's April 22, 2026 Q1 results release and linked TXN lane.
Direct programmable edge-AI inference option where revenue conversion, customer concentration, liquidity, and dilution dominate.
Market cap$266.0M
Next earningsNot confirmed
Latest qtr revenue$2.7M
Role in stack
Blaize sells programmable edge-AI inference hardware, software, and AI Services into rugged edge, industrial, sovereign AI, and hybrid compute customers. Conversion requires NeoTensr, Winmate, Nokia, Datacomm, and service opportunities to ship, be accepted, be collected, and diversify revenue.
Revenue mix
One operating segment. Q1 2026 revenue was 94.9% from one related-party customer; future mix between proprietary hardware, third-party systems, software, and AI Services is not mature.
Latest qtr revenue
Q1 2026 revenue was $2.738M; source is Blaize's May 14, 2026 Q1 results release and linked BZAI lane.
Market caps use read-only Discovery instruments.market_cap values queried on 2026-06-13. Nasdaq earnings pages for the basket were checked on 2026-06-13 and did not provide confirmed next earnings dates, so every card labels the date field Not confirmed. Latest-quarter revenue metrics use the linked security lanes and official company releases.
What Confirms Or Weakens
AreaWhat confirmsWhat weakens or invalidatesWatch next
01Node thesis
Design wins become production
What confirms
New design wins, production ramps, named OEM, Tier 1, robotics, security-camera, defense, or industrial customer evidence show that endpoint silicon demand is becoming reported revenue.
What weakens or invalidates
Reported growth comes mainly from unrelated handsets, licensing, analog restocking, data-center infrastructure, or legacy products rather than local AI and endpoint control.
Watch next
AMBA CV and N-series ramps
QCOM auto, IoT, and custom silicon
NXPI and LSCC production mix
02Economics
Revenue, margin, cash
What confirms
Revenue lands within or above guidance while gross margin, inventory, channel sell-through, and free cash flow improve enough to show production economics.
What weakens or invalidates
Guidance is cut, gross margin compresses, inventory or distributor receivables rise faster than revenue, or design-win language stays unquantified.
Watch next
AMBA Q2 FY2027 guide
LSCC distributor sell-through
AMBQ gross margin and cash burn
03Customer
Program breadth
What confirms
Customer concentration falls, distributor sell-through stays clean, and named customer programs move beyond demos, reference platforms, or potential-value LTAs.
What weakens or invalidates
WT, top-customer, related-party, OEM, Tier 1, distributor, TSMC, STMicro, China, tariff, or export-policy exposure becomes an order, margin, or working-capital problem.
Watch next
AMBA WT and Asia mix
AMBQ top-three customer share
BZAI related-party concentration
04Funding
Balance sheet and dilution
What confirms
Cash generation funds growth, transactions close with acceptable leverage or dilution, and smaller names show revenue growth without escalating share issuance.
What weakens or invalidates
AMI or Silicon Labs financing absorbs too much flexibility, small-cap burn persists, going-concern language remains, or dilution grows faster than production evidence.
Watch next
LSCC AMI financing
TXN Silicon Labs terms
INDI, AMBQ, and BZAI cash use
05Policy and supply
Foundry and export gates
What confirms
Foundry, OSAT, memory, and component availability support shipments, while tariff and export-control disclosures do not disrupt customer ramps or product margins.
What weakens or invalidates
Advanced-node cost, TSMC reliance, China exposure, export restrictions, tariff costs, or memory shortages reduce deliveries, margin, or customer demand.
Watch next
QCOM handset memory commentary
AMBA advanced-node and Asia disclosures
AMBQ TSMC sole-wafer reliance
06Stale condition
Refresh trigger
What confirms
Discovery daily_ohlc remains current, no new company release changes the basket read, and source lanes still match the latest public filings and releases.
What weakens or invalidates
A new trading session, earnings release, investor day, transaction update, customer disclosure, export-policy action, or source-lane refresh arrives before this page is refreshed.
Current public-company results, latest quarterly revenue, revenue mix, guidance, margin, and product evidence used to refresh stale or thin local claims without replacing durable knowledge lanes.
Nasdaq earnings pages and company IR/event routes were checked on 2026-06-13 for the 12-name basket. No source provided a company-confirmed next earnings date during this worker pass, so Basket cards use Not confirmed instead of estimates.
Read-only local status, lineage spot checks, market-cap query, and latest-table rows for all 12 tickers. DuckDB coverage showed all 12 tickers current through 2026-06-12; AMBQ has 220 daily rows and BZAI has 355 daily rows, so both still have shorter public chart history than mature semiconductor peers.
node metadata; local discovery daily_ohlc; python -m discovery.cli status AMBA --json; python -m discovery.cli lineage-status --ticker AMBA --dataset daily_ohlc --limit 3 --json; python -m discovery.cli lineage-status --ticker AMBQ --dataset daily_ohlc --limit 3 --json; read-only DuckDB queries against daily_ohlc and instruments.market_cap.
Chart route
Selected-security charts render in the right rail from the report API, with weekly bars derived from daily_ohlc; weekly aggregation uses first open, weekly high, weekly low, last close, and summed volume, with 20-week and 100-week EMAs when history is long enough. The visible horizon is three years, and right-rail chart order follows Basket rank.
Static chart setup thresholds from the prior page used 2026-06-05 data and are stale after the 2026-06-12 Discovery refresh. Several broad semiconductor names can move on handset, auto, industrial, data-center, rate, or inventory factors that are not endpoint-compute evidence. Product-line margins are missing for many edge-AI claims. Earnings dates are not confirmed. AMBQ and BZAI have limited trading history. BZAI remains watch-only until shipped, accepted, collected, and diversified revenue is visible.