Most IP vendor coverage reads like a product announcement. The more useful document is the financial filing, because it shows where the money actually comes from — and that determines how a vendor behaves when your programme hits trouble.
This piece reads VeriSilicon's reported 2026 results as engineering intelligence. All figures below are from the company's own disclosures and national trade press reporting of them; anything not directly attested is marked as needing confirmation.
The Model, Stated Plainly
VeriSilicon Microelectronics (688521.SH), founded in August 2001 and headquartered in Shanghai, runs a model it calls Silicon Platform as a Service. Two revenue streams: semiconductor IP licensing (upfront licence fees plus per-unit royalties) and chip customisation services (design plus mass production).
The company reports six categories of processor IP — graphics, neural network, video, digital signal, image signal and display processors — alongside more than 1,700 analog/mixed-signal, RF and interface IP. Reported design wins include 4nm FinFET and 22nm FD-SOI.
Why this matters to a buyer: a vendor earning from both IP and turnkey execution has an incentive to keep your programme moving through production, which is different from a pure IP licensor whose involvement effectively ends at delivery. It is also a different conflict profile, discussed below.
The 2026 Numbers
First quarter 2026: revenue RMB 836 million, up 114.47% year on year, with a net loss attributable to parent of RMB -341 million. New orders signed between 1 January and 29 April 2026 reached RMB 8.24 billion, of which roughly 91.37% was described as AI-compute related.
Breaking the Q1 revenue down by stream is where the useful signal sits: IP licence fees RMB 143 million (+52.97%), royalties RMB 32 million (+19.26%), chip design RMB 193 million (+57.60%), and mass production RMB 467 million (+219.93%). Mass production is not a side business; by Q1 it was the largest single contributor.
First half 2026: revenue RMB 1.864 billion with a net loss attributable to shareholders of RMB -612 million. IP licensing service revenue was RMB 349 million, about 18.73% of total, split into licence fees of RMB 293 million (+4.32%) and royalties of RMB 56 million (+10.85%). Backlog reached RMB 12.449 billion at end of June, up 142.52% from end of Q1, with over RMB 10 billion of that in mass-production services.
- Growth is concentrated in execution volume, not licence volume — licence fees grew single digits while mass production grew triple digits.
- Royalty growth is modest, which suggests current shipment volume behind licensed IP is still building.
- Backlog growth of this size is a capacity-commitment question as much as a demand signal.
What 77% IP Gross Margin Actually Tells You
IP licensing was reported at roughly 77% gross margin in the first half of 2026, having been cited near 90% historically. The direction of travel is worth noting even though the absolute level is high.
High licence margin is structural: development cost is front-loaded and the marginal cost of one more licence is near zero. The practical consequence for a buyer is that licence pricing has real negotiating room, particularly where several vendors offer an equivalent block. Royalty rates in the industry are commonly cited in the 1–2% range per shipped part, though the actual rate is whatever the specific contract says.
The falling blended margin reflects revenue mix rather than IP becoming less valuable — mass production services carry structurally lower margin than licensing. Reading a margin decline as a decline in IP quality would be a misread.
What This Means If You Outsource to This Model
The advantage is handoff. A single counterparty carrying IP selection, design execution and production removes the gaps where programmes usually lose weeks — the interface between whoever designed the block and whoever has to get it built.
The trade-off is optionality. Once your design is built around one vendor's IP and inside one vendor's production flow, moving is expensive. If multi-foundry resilience or second-sourcing matters to your programme, it has to be stated before the architecture is fixed, not raised later.
One further discipline point: an IP block being available is not the same as it being qualified for your process and library version. That distinction is decided per PDK, per release, and it is the buyer's verification step regardless of who supplies the block.
- State second-sourcing requirements before architecture freeze, in writing.
- Confirm which specific PDK and library release each block is characterised against.
- Ask what happens to your queue position if their backlog absorbs your slot.
- Separate licence negotiation from integration scope so neither is hidden inside the other.
Where This Analysis Stops
Everything above comes from public filings and trade press reporting of them. Two things are not established here: whether any specific IP block is available and characterised for your node and library version, and the forward projections, which are the company's own guidance rather than independent fact.
