If you are sourcing PCBs for AI or data-center server projects in the Middle East, here is the direct answer: the surface finish you specify on the drawing is the single most important line item that determines both your supplier’s quotation and your board’s long-term reliability in high-temperature, high-current environments. We are Informic, and we coordinate turnkey PCBA for teams like yours. We write this from our daily work, not from a brochure. When we receive an RFQ with a vague finish callout, we stop and ask questions. You should too.
Most procurement buyers we speak with are not surprised by the cost difference between finishes. They are surprised by how much the finish choice affects lead time, solder joint integrity, and the number of boards that fail during power-on testing. The finish is not a cosmetic detail. It is the metallurgical interface between your expensive processor and the copper traces that carry data and power.
This article is a practical guide to specifying surface finishes for server-class boards. We will separate what we know from what we recommend, and we will give you specific questions to put in your next RFQ.
Why surface finish matters more for AI server boards
AI accelerator boards and data-center server motherboards share two traits that make finish selection critical: high layer counts and dense BGA packages. A typical board we coordinate has 16 to 24 layers, with multiple 0.8 mm pitch BGAs. The finish must provide a flat surface for solder paste printing, maintain coplanarity during reflow, and resist oxidation over a long shelf life. If the finish is uneven, you get solder voids. If the finish is too thick, you get solder balls. If the finish is wrong for the assembly process, you get intermittent failures that are hell to debug in the field.
We have seen procurement teams in Dubai and Riyadh receive excellent quotations from Asian manufacturers, only to discover that the finish specified was not compatible with the halogen-free solder mask or the specific reflow profile used by the assembler. That is a coordination failure, not a component failure. Our job is to prevent that.
The four finishes we actually see in server board RFQs
There are many finishes on the market, but for AI and data-center servers, we see four main options in real quotations. We will describe each briefly, then give you a comparison table. We are not going to cite industry test data because we have not verified it for this article. Instead, we will tell you what we observe in our coordination work and what questions to ask your supplier.
1. HASL (Hot Air Solder Leveling)
This is the old workhorse. Lead-free HASL is still used for low-cost boards, but we rarely recommend it for server-class BGAs. The finish is uneven, and the surface can have peaks and valleys that are too large for fine-pitch components. If you are building a simple power supply board, fine. For an AI accelerator, no.
2. ENIG (Electroless Nickel Immersion Gold)
This is the most common finish we see for high-reliability server boards. The nickel layer provides a barrier against copper diffusion, and the gold layer protects the nickel during storage. ENIG is flat, solderable, and works well with fine-pitch BGAs. The main risk is "black pad" – a brittle intermetallic that can form if the immersion gold process is poorly controlled. That is a process issue, not a material issue. Ask your supplier how they monitor the nickel bath. A good supplier will have a clear answer.
3. ENEPIG (Electroless Nickel Electroless Palladium Immersion Gold)
This is ENIG with an extra palladium layer. We see ENEPIG specified more often for boards that need multiple reflow cycles or wire bonding. For AI servers, it is a premium option. It costs more, but it eliminates the black pad risk and provides a better surface for repeated thermal cycling. If your board has a mix of BGA and through-hole components that require two or three passes through reflow, ENEPIG is worth the premium.
4. OSP (Organic Solderability Preservative)
OSP is the cheapest finish, and it is perfectly fine for many boards. But it has a short shelf life (we recommend using it within six months of production), and it is not suitable for boards that will be handled repeatedly or stored in humid environments. In the Middle East, humidity varies greatly between coastal and inland locations. If your boards are going to sit in a warehouse in Jebel Ali for eight months before assembly, OSP is a bad choice. If you are assembling within weeks, OSP can save you money.
| Finish | Typical cost per sq ft (relative) | Flatness for fine-pitch BGA | Shelf life | Our typical recommendation for AI servers |
|---|---|---|---|---|
| Lead-free HASL | $ (lowest) | Poor | 12+ months | Not recommended |
| OSP | $ | Good | 3-6 months | Only for quick-turn assembly |
| ENIG | $$ | Excellent | 12+ months | First choice for most server boards |
| ENEPIG | $$$ | Excellent | 12+ months | For multi-reflow or high-reliability |
This table is a starting point, not a specification. The actual cost difference between ENIG and ENEPIG varies by supplier and board size. We have seen quotes where the difference was only 3% of the total board cost, and others where it was 12%. That is why we always ask for a line-item breakdown in the quotation.
How we handle surface finish in an RFQ
When we receive an RFQ for a server board, we do not only pass the Gerber files to a manufacturer. We review the finish callout against the BOM, the component package types, and the intended assembly process. Here is the specific sequence we use. You can use it too.
Step 1: Check the drawing for a finish callout. If it says "ENIG" or "HASL" without a thickness spec, we ask for the IPC class (Class 2 or Class 3) and the specific finish thickness. For ENIG, we ask for nickel thickness (typically 3-6 microns) and gold thickness (typically 0.05-0.1 microns). A supplier that cannot give you these numbers is not controlling their process.
Step 2: Cross-reference the finish with the BGA pitch. If the board has BGAs with 0.5 mm pitch or smaller, we immediately flag HASL as a risk. We also check for mixed-technology boards. If you have a through-hole connector that requires wave soldering, the finish on the through-hole pads matters just as much as the finish on the BGA pads.
Step 3: Ask about the assembly location. This is a Middle East-specific concern. We coordinate boards that are manufactured in Asia and assembled in the UAE, or vice versa. The transit time and storage conditions between board fabrication and assembly can be weeks. If the finish is OSP, we tell the client to plan for a shorter shelf life. If the finish is ENIG, we are more relaxed.
Step 4: Get a DFM (Design for Manufacturing) review in writing. We do not trust a verbal "okay" from a manufacturer. We ask for a written DFM report that specifically addresses the finish. The report should confirm that the finish is compatible with the solder mask, the surface mount components, and the reflow profile. If the manufacturer cannot provide this, we treat it as a red flag.
Here is a concrete example of a question to put in your next RFQ, word for word: "Please confirm the surface finish is ENIG with nickel thickness of 4-6 microns and gold thickness of 0.05-0.1 microns, and confirm that this finish is compatible with a 0.8 mm pitch BGA and a lead-free reflow profile with a peak temperature of 245°C. Provide a written DFM note confirming this."
That single question will tell you more about the supplier than any marketing material. We ask it every time.
What we recommend for AI and data-center server teams
Based on our coordination work, here is our practical recommendation. For most AI server boards, we recommend ENIG as the default. It gives you the flatness, shelf life, and solderability you need without the premium cost of ENEPIG. If your board has more than two reflow cycles, or if you are using a very large BGA (over 40 mm), we recommend ENEPIG. If you are building a prototype or a low-volume run and you know the boards will be assembled within two months, OSP is a legitimate cost-saving option.
We do not recommend HASL for any board with a BGA pitch below 0.8 mm. That is a hard rule for us. We would rather lose the order than ship a board that fails in the field.
One more thing we do in our daily work: we ask the PCB manufacturer for a cross-section photo of the finish on the first article. This is a simple, low-cost verification that shows the nickel and gold layers are at the right thickness and are uniform. You should ask for the same. It is not a test report; it is a process check. A reliable manufacturer will send it without complaint.
FAQ: Surface finish for AI and data-center servers
1. How do I specify the surface finish correctly in my RFQ to avoid quotation delays?
Put the finish callout in the drawing and in the RFQ text. Do not rely on the drawing alone. Write the full finish name (e.g., ENIG), the thickness range, and the IPC class. Then add the compatibility question we provided above. This forces the supplier to confirm they can meet your requirement before they quote. If they ask for clarification, that is a good sign. If they quote without asking, be cautious.
2. What is the practical difference between ENIG and ENEPIG for a server board, and is the extra cost worth it?
The practical difference is the palladium layer. ENEPIG adds a layer between the nickel and gold, which prevents the brittle intermetallic formation that causes black pad. For most server boards, ENIG is sufficient. For boards that go through multiple reflow cycles (e.g., a board with BGAs on both sides) or for very large BGAs, ENEPIG reduces the risk of field failures. We recommend ENEPIG when the board is expensive to debug or replace. The cost difference is often small relative to the cost of a failed board in a data center.
3. Can I save money by using OSP finish for a prototype AI server board?
Yes, if you control the timeline. OSP is significantly cheaper than ENIG. But you must assemble the boards within a few months of fabrication. We also recommend that you do not use OSP if the boards will be shipped across the sea in a humid container or stored in a non-climate-controlled warehouse. In the Middle East, summer heat and humidity can degrade OSP quickly. If you are doing a quick-turn prototype and the boards stay in your facility, OSP is a fine choice. If there is any uncertainty, pay for ENIG.
We are Informic, and we coordinate turnkey pcbManufacturing and pcba for teams across the region. When you send us a components list and a set of Gerbers, we do not only quote. We review the finish, the BOM, and the assembly process together. If you want to discuss a specific board, send us a quote request. We will give you a direct answer.
Surface finish is a small line item on a large BOM, but it is the line item that determines whether your board works on the first power-up. Ask the right questions, get the DFM in writing, and do not let a supplier save you a dollar on finish that costs you a server rack later.
FAQ
What do we review first for pcb surface finishes?
We begin with the functional requirement, the current revision-controlled data package, critical components, expected volume, quality requirements, and delivery deadline.
How do we make purchasing risk easier to compare?
We compare date code, traceability, lead time, MOQ, substitute status, quality checks, and the supplier assumptions behind each quotation instead of comparing only a unit price.
What do we need for an accurate quotation?
We need the correct document revision and, where relevant, the BOM, Gerber or ODB++ files, centroid data, target quantity, application, test expectation, quality requirement, and requested delivery date.
Technical reference: For related engineering context, see IPC standards and industry resources.
