Variscite: memory shortage, inventory, and capacity discipline after 2025

The main article already established that Variscite is the real value engine inside Telsys. This follow-up isolates only the bottleneck that opened after growth returned: not end demand for modules, but the ability to secure memory on time, fund the inventory layer that now sits behind that effort, and decide when capacity is still an advantage and when it starts turning into a capital commitment.

That matters now for a simple reason. In 2025 Variscite returned to growth, its 2026 backlog jumped, and management said in the presentation that it met delivery dates for practically all Q4 2025 orders. But at the same time a global shortage in DRAM and flash pushed component costs higher, led to earlier procurement, and started moving cash from the consolidated headline into inventory, payables, and bank debt. So 2026 looks less like a demand year and more like a proof year for supply discipline, pricing, and capital allocation.

The four non-obvious points are these:

• The shortage already hit margin, but it has not yet broken delivery. Variscite's revenue rose to $80.280 million in 2025 and segment profit rose to $39.280 million, but gross margin slipped to 53.1% from 55.0%. At the same time the presentation says quarterly component mark-ups can reach 40% and more in some cases, yet the company still met delivery dates for practically all Q4 orders.
• The inventory build is mainly a supply buffer of components, not finished goods waiting for buyers. Variscite does not manufacture without a prior customer order, it holds raw components for production, and it assembles the module only after receiving the order. In 2025 it stopped destocking and increased memory-component inventory among other items.
• Early procurement was not funded only from cash on hand. Customer advances at Variscite rose to $10.703 million from $5.814 million, receivables at Variscite actually fell by about $0.79 million because more shipments moved toward upfront-payment customers, supplier balances at Variscite rose by $10 million, and Variscite's total bank debt rose to $19.702 million from $6.855 million.
• Capacity has not yet hit a physical wall, but it is not yet an automatic expansion case either. Variscite is running at 60% to 70% utilization, the current site can support 3 additional production lines and more than 50% extra capacity, but the large Kiryat Gat plant still remains at the land-pricing and objection stage.

The value engine sits here, which is why memory matters more than the headline

This is the core of the issue. In 2025 Variscite generated about 90.7% of the combined segment profit of Telsys's two operating engines, $39.280 million against $4.023 million in distribution. So the memory shortage is not a side discussion about one component line in the bill of materials. It is a question about whether the engine that creates most of Telsys's value can keep converting demand into profit, or whether it starts requiring more and more capital just to preserve the same delivery performance.

That chart shows why a surface-level read of 2025 can mislead. Revenue and segment profit did rise, but gross margin fell by 190 basis points in the same year that the company described a global memory shortage and the presentation already pointed to quarterly mark-ups of 40% and more in some cases. The most reasonable interpretation is that the pressure did not disappear. It was temporarily contained through price increases, early procurement, and delivery discipline.

That is also why the presentation matters. If Variscite was still able to meet delivery dates for practically all Q4 2025 orders despite the allocation crisis, then the 2025 problem was not demand collapse or shipment failure. The problem was that supply became more expensive, more dependent on memory availability, and more capital-intensive.

There is another layer that makes flexibility less open-ended than it first appears. On one hand, the presentation says Variscite is expanding both vendor and sales channels to meet component demand. On the other hand, the annual report makes clear that processor dependence remains high: around 70% of Variscite products are based on NXP processors, and NXP represented about 38% of supplier purchases in 2025. Once a module's development is complete, the processor cannot simply be changed. So procurement flexibility is real, but partial. The company can broaden sourcing around memory, but it cannot keep redesigning the product architecture every time the component market tightens.

The inventory build is a supply layer, not a demand bet

One of the easiest mistakes in reading 2025 is to look at the inventory increase and treat it as a return to excess-risk inventory. That is incomplete. Variscite explicitly says it does not manufacture without a prior order, that it holds component inventory only for production, and that actual assembly takes place only after an order is received. In other words, what was built here is mostly a protective layer of components, not finished boxes sitting around waiting for customers to commit.

The 2025 story is also very specific. After inventory had already returned to normal levels at the end of 2024, the company stopped the reduction process, adjusted procurement to match recent activity, and increased memory-component inventory among other items. This is a supply response to a shortage, not just a generic inventory expansion.

The chart highlights an important paradox. The company did stop destocking, yet inventory days actually improved, from 283 to 252 on an annual average basis and from 259 to 240 in the fourth quarter. At the same time backlog for the next delivery year jumped to $46.8 million from $29 million. That means inventory rose because Variscite brought procurement forward into stronger demand, not because product got stuck.

That does not eliminate the risk. It defines it correctly. Anyone reading the numbers as if the company returned to dead inventory is missing what is really happening: Variscite is buying itself supply continuity inside a tight component market. If memory prices keep rising, and if allocation remains difficult, the question will not be whether the company has inventory. The question will be whether it bought early enough and at a cost that still allows margin protection.

Who is funding this supply buffer

The most interesting point in 2025 is that early procurement did not rely only on internal cash. It relied on a combination of customers, suppliers, and banks, which means that even if the operating response to the shortage was correct, it is already sitting on a capital structure that needs discipline.

The first layer is customers. Variscite's usual customer terms are typically about 50% advance payment at order and 50% before shipment, and only distributors receive 30+ to 60 day credit. Consistent with that, customer advances, all within the SOM segment, rose to $10.703 million from $5.814 million. At the same time the directors' report says Variscite's receivables actually fell by about $0.79 million because the shipment mix moved toward customers paying advances rather than normal credit terms. That is crucial. Part of the early inventory layer was pre-funded by customers, not only by external finance.

The second layer is suppliers. The directors' report says explicitly that supplier balances at Variscite increased by $10 million because the company decided to end the inventory-reduction process this year. In other words, suppliers are also absorbing part of the funding burden of the move back toward procurement.

The third layer is banks. In June 2024 Variscite took a NIS 30 million loan, and in December 2025 it took another $15 million loan. On the consolidated balance sheet, current and non-current bank debt together rose to $19.702 million from $6.855 million.

The chart does not tell the whole story because it does not include the supplier move, but it does show the direction: Variscite did not answer the memory shortage only through the cash line. It leaned on three funding channels at once, more customer advances, more supplier credit, and more bank debt.

That is exactly where easy conclusions become dangerous. The fact that the move was funded does not mean it is free. As a condition for the June 2024 loan, Variscite undertook that its surplus would not fall below its financial liabilities. After the reporting date, in March 2026, it already needed bank consent to distribute up to $10 million and had to commit to keeping surplus of at least $18 million after that distribution. This article is not a dividend deep dive, but the number still matters here because it shows that the bank is already sitting inside the picture. Once early procurement, higher inventory, and capacity decisions lean on debt as well, freedom of action stops being open-ended.

The backlog is strong, but it still needs disciplined reading

Variscite enters 2026 with a backlog that looks materially stronger than the one it had a year earlier. For 2026 it showed $46.8 million of backlog against $29 million for 2025 a year earlier, plus another $4.5 million scheduled for 2027 to 2028 against $3.1 million in the tail a year before. The presentation repeated the point directly, a 61% increase in the backlog for 2026.

But the company itself adds the most important caveat in the entire discussion: total backlog in the SOM segment does not represent the expected revenue for the coming year, because customers usually order mainly for the coming quarter and actual quarterly revenue also includes orders that arrive during the quarter itself. So the backlog is evidence of stronger demand and better visibility, but it is not a disguised annual revenue guide.

It is still higher-quality backlog than what sits in distribution, for several reasons. Variscite customers do not have cancellation or postponement rights, the company does not produce without prior orders, and the advance-payment structure is heavy. So the demand risk here does not look like a wide but customer-flexible distribution backlog. The risk is elsewhere: component allocations, memory pricing, and the company's ability to pass the higher cost through without damaging order flow.

Variscite's commercial setup adds another layer of resilience, but it does not cancel dependence. The Arrow group accounts for more than 10% of Variscite revenue, $27.327 million in 2025, and the company says it is not dependent on that distributor because it can sell directly to customers that currently buy through it if needed. That is a useful sign that the commercial chain is not the main bottleneck today. The main bottleneck remains the component layer.

Capacity is a real option, but not yet a mandatory expansion order

This may be the biggest gap between headline and economics. It is easy to take the memory shortage and the higher backlog and jump straight to the conclusion that the next constraint is plant capacity. For now, that is too early.

Variscite is currently operating at 60% to 70% of maximum production capacity. After adding another production line in the first quarter of 2023, it now has six production lines and supporting equipment with total historical cost of $6.5 million. According to the presentation, the current site can still host 3 more production lines and add more than 50% to current capacity. That means there is still room inside the existing site before the company reaches a true operating end-point.

The more important layer actually sits inside the long-term project. In July 2024 Variscite received a recommendation for roughly 21 thousand square meters of land in Kiryat Gat for a plant with at least 13.5 thousand square meters of built area. By the report date it had already received final land pricing, and after the balance-sheet date it submitted an objection to that pricing. The presentation reinforces the same point and says the company is awaiting feedback from the Israel Land Authority before deciding how to proceed.

So even the larger capacity layer has not yet become a committed capital deployment. That matters because the current memory shortage is actually pushing Variscite toward restraint. If the existing site can still expand, and the real debate is about component pricing, inventory, and debt, then the right move is not necessarily to rush straight into a new plant. The right move is to first test whether demand stays strong enough, whether pricing can hold, and whether the need for heavier inventory is structural rather than cyclical.

Conclusion

The main article showed that Variscite is the engine that explains Telsys. This follow-up sharpens the point that the 2026 test will not come first from demand, but from the quality of conversion from demand into margin and balance sheet. By the end of 2025 Variscite had already shown three very positive things: growth returned, backlog expanded, and delivery dates were largely met despite a severe memory shortage.

But what is working now comes with a clear price tag. Gross margin already slipped, the memory inventory layer was rebuilt, and the response was funded through a mix of customer advances, supplier credit, and bank debt. This is not the story of a company that has run out of capacity. It is the story of a company that still has capacity, but now has to decide how much to pay today in order to protect supply tomorrow.

Current thesis: Variscite can remain the value engine of Telsys after 2025, but only if it keeps all three lines working together, on-time supply, pass-through of memory inflation, and capital discipline that does not turn the inventory buffer into a permanent balance-sheet drag.

The strongest counter-thesis is that the market is overstating the shortage, that early procurement and broader sourcing have already solved most of the problem, and that once pricing calms down Variscite will still have a high backlog and spare capacity that can drive another leg of growth. That is possible. But as of the end of 2025, the local evidence supports the more cautious read: the bottleneck has not moved into the plant walls yet, but it is already sitting squarely inside components, inventory, and the balance sheet.