Treat supply as CRM data
In a market where fabrication, packaging and lead times can make or break a quarter, supply assumptions need to live in the opportunity record, not in post‑sales spreadsheets. Practical fields include expected silicon windows, packaging dependencies, lead‑time risk, and customer timing flexibility so that a late‑stage deal without validated delivery assumptions is never mistaken for a true commit. Making these operational fields mandatory helps separate commercial conviction (customer intent) from fulfilment confidence (can we actually deliver on time). (ico-optics.org) (igorslab.de)
A chip sale is not really a sale when the package does not exist yet. That sounds obvious. It is not how many sales systems work. In a lot of semiconductor companies, the CRM still treats demand as the main fact and supply as a later check. A rep marks an opportunity as likely. Finance rolls that into a forecast. Operations keeps the messy part in separate files and side conversations. In ordinary markets, that can be clumsy and survivable. In the AI chip market, it can wreck a quarter. The reason is simple. The bottleneck is no longer just the wafer fab. It is the whole chain after the design win. Advanced-node capacity is tight. High-bandwidth memory is tight. Advanced packaging is tight. SEMI said on April 1 that 300mm fab equipment spending is still climbing hard in 2026 and 2027 because AI demand is forcing another round of capacity buildout across logic and memory. That is not what an industry looks like when supply is easy. (semi.org) Packaging is the part many commercial systems still fail to model. TSMC’s 2024 annual report makes clear how central advanced packaging has become to modern AI hardware. CoWoS, InFO, SoIC, and related technologies are not side services. They are part of the product. The report also describes TSMC’s role as a “trusted technology and capacity provider,” which is a useful phrase here because capacity is not an operational afterthought. It is part of the promise being sold. (investor.tsmc.com) That is why the card’s prescription is more than process hygiene. If a sales team is booking “commit” deals without fields for expected silicon window, packaging dependency, lead-time risk, and customer timing flexibility, then the company is mixing up two different things. One is commercial conviction: the customer wants the part. The other is fulfillment confidence: the company can actually ship on the date that matters. Those are not the same number. They should not live in the same checkbox. The industry itself keeps saying this in corporate language. Intel’s foundry messaging over the past year has leaned on the same idea from the other direction: customers do not just buy process technology, they buy manufacturing and packaging capacity, plus a supply chain they believe will hold. At Intel Foundry Direct Connect in April 2025, the company said its job was to “earn their trust” and framed advanced packaging and manufacturing as part of that trust. Later in 2025, Intel’s packaging group wrote that chiplet-era products require not only packaging technology but also “the capacity and secure supply chain to meet this demand.” Strip away the branding and the point is blunt. Supply credibility is part of the deal before the deal closes. (intc.com) Research on semiconductor lead times points the same way. Order lead time in this industry depends on a mix of order, planning, delivery, customer, and product variables, not a single promised date typed in at the end. If those variables sit outside the opportunity record, the forecast is blind at the exact moment executives most want certainty. A late-stage deal can look healthy because the customer is real, even while the package substrate, test slot, or HBM allocation is still speculative. (link.springer.com) The practical fix is boring, which is why it matters. Make supply fields mandatory before a deal can move into commit. Force someone to name the fab window. Force someone to state whether the package path is standard, constrained, or not yet reserved. Force someone to record whether the customer can absorb a slip of two weeks, one month, or not at all. Once that data lives in CRM, the forecast stops pretending that demand alone ships product. That turns one vague pipeline number into two hard ones. The first says the customer intends to buy. The second says the company can deliver on time. In a market where CoWoS lines have been expanded at speed, where TSMC has had to keep adding packaging capacity, and where even new facilities still take time to come online, that split is not bureaucracy. It is the difference between a real quarter and a spreadsheet quarter. (trendforce.com)
