A100 Compute Box 4ch PCIe
The A100 Compute Box 4ch PCIe is a four-camera PCIe card carrying DeepGrid's SoC2, sold to OEMs and integrators who need full surround perception inside their own host architecture. At ₹3,00,000 it is the highest-priced compute SKU in the portfolio, with an 89% gross margin and management-projected FY2032 revenue of ₹36 Cr at 1,200 units.
- Listed price
- ₹3,00,000
- FY2032 revenue
- ₹36 Cr
- Share of plan
- 3.2%
- Gross margin
- 89%
- First revenue
- FY2028
- Volume plan
- 1,200 @ FY32
From input to outcome
Proposed product architecture. The scene illustrates its application, not a shipped kit.
Four cameras give full surround coverage
The card runs perception across all four streams
Results go to the host over PCIe
The host system builds its own product on top
Drag to rotate. Conceptual layout, not a mask.
On the die
The case in brief
- At ₹3,00,000 per unit and 89% gross margin, the A100 4ch PCIe is the highest-priced compute SKU in the portfolio, making the FY2032 revenue case dependent on margin rather than volume.
- Management projects ₹36 Cr at 1,200 units by FY2032, a consistent annual doubling from FY2028, against a $4.8B addressable market requiring only 0.20% capture, per management assumptions.
- The card's dependency on host thermal and power budget narrows the addressable pool to integrators with sufficient platform headroom, concentrating early revenue in a small number of direct design-in accounts.
What it is used for
Multi-camera automotive edge-AI
- Buyer
- OEMs and integrators running complete multi-camera stacks on their own platform
- The job
- These buyers need full surround perception across four camera feeds but want to integrate the compute into their own host architecture rather than adopt a turnkey system.
- What DeepGrid supplies
- A PCIe card carrying the SoC2 with four camera inputs, an SDK and model toolchain, and structured object lists output to the host over PCIe, so the integrator builds its own product on top without designing its own board.
Industrial inspection and infrastructure monitoring
- Buyer
- OEMs and integrators
- The job
- Industrial inspection and infrastructure monitoring applications require simultaneous multi-camera perception at high utilisation without the cost or complexity of a custom board design.
- What DeepGrid supplies
- The A100 4ch card runs parallel perception across all four streams on the SoC2, with each compute cube maintaining its own local DMA against an eight-bank L2 to keep utilisation high, then passes results to the host system.
Security and surveillance compute
- Buyer
- OEMs and integrators
- The job
- Security and infrastructure monitoring deployments demand multi-camera coverage with on-card compute so that the host system receives processed alerts rather than raw video streams.
- What DeepGrid supplies
- Up to four camera feeds enter via the PCIe interface; the SoC2 runs parallel perception on all four channels and outputs object lists and alerts to the host, with the customer's software consuming structured perception data directly.
The story
The A100 4ch PCIe is a four-camera SoC2 card that drops full surround perception into a customer's own host architecture.
The A100 Compute Box 4ch PCIe is a PCIe card carrying DeepGrid's SoC2 with four camera inputs, an SDK, and a model toolchain. The host system provides the enclosure and power; the card provides the perception compute. At ₹3.00 L it is the highest-priced compute SKU in the portfolio and the closest the module line gets to being a complete system.
- ₹3.00 Lunit price
- Systemssegment
- Computeoperating domain
- FY2028first revenue
- Four-channel SoC2 PCIe cardcore component
- Camera input harnessingcomponent
Four parallel camera streams run perception on the SoC2 and deliver structured results to the host over PCIe.
Four cameras provide full surround coverage; the SoC2 runs perception across all four streams simultaneously, with each compute cube maintaining its own local DMA against an eight-bank L2 cache, the architecture that keeps utilisation high rather than starving on a single shared memory path. Structured results pass to the host over PCIe, and the integrator builds its own product on top. The binding constraint is the host system's thermal and power budget, which determines which customers can adopt the card.
- Four camerassurround coverage
- Per-cube local DMAmemory access
- Eight-bank L2cache architecture
- Results over PCIehost interface
- Host thermal & powerkey dependency
OEMs and integrators building multi-camera perception into their own platforms buy this card to avoid custom silicon work.
The buyer is an OEM or integrator already running a host machine who needs to add full surround perception without designing a custom board, robotics platforms, industrial vehicles, and infrastructure monitoring deployments where the host receives processed object lists rather than raw video. The channel is direct design-in with larger integrators. The FY2032 volume of 1,200 units represents a 0.20% capture of the automotive edge-AI accelerator modules market, sized at $4.8B in 2025; capture rates are management assumptions and pool sizes are third-party published estimates (Dataintelo: $4.8B 2025 to $22.6B 2034 at 21% CAGR).
- OEMs & integratorsbuyer
- Direct design-inchannel
- $4.8B (2025)addressable pool
- 21% CAGR to 2034pool growth
- 0.20% capture FY2032management assumption
Management projects ₹36 Cr at 89% gross margin by FY2032, making this the largest revenue contributor in the compute-box family.
The A100 4ch PCIe is management-projected to reach ₹36 Cr at 1,200 units in FY2032, representing 3.2% of group revenue at an 89% gross margin, the largest of the three compute-box SKUs. Revenue doubles each year from FY2028, reflecting design-in cycles rather than a consumer volume ramp. The card requires no vehicle integration, no installer, and no homologation.
- 1,200 unitsFY2032 units (mgmt proj.)
- ₹36.0 CrFY2032 revenue (mgmt proj.)
- 3.2% of grouprevenue share
- 89%gross margin
- Host thermal & poweradoption constraint
- ₹0 CrFY270 units
- ₹2.25 CrFY2875 units
- ₹4.5 CrFY29150 units
- ₹9 CrFY30300 units
- ₹18 CrFY31600 units
- ₹36 CrFY321,200 units
What has to go right
The foundational dependency is ASIC tapeout and PCIe interface validation; without a functioning SoC2 die, no compute-box SKU ships, and the A100 4ch PCIe is among the nine products gated by that milestone. Host thermal and power budget is the second constraint: customers whose enclosures cannot accommodate the card's requirements are excluded from the addressable pool, limiting the market to integrators with sufficient platform headroom. The direct design-in channel means revenue is concentrated in a small number of larger integrators, so a delay or cancellation in any one design-in has an outsized effect on the FY2028–FY2030 ramp. Volume at 1,200 units by FY2032 remains modest in absolute terms, so the business case depends on margin rather than scale, and any erosion of the ₹3,00,000 price point would materially affect the revenue line.
Key dependency ASIC tapeout and PCIe interface validation.
Further reading and viewing
Related products
Prices, volumes, revenues and margins are management projections. The text is written from the linked documents, and every figure in it appears in the source it came from.