Telsys: Nvidia, subcontractors, and the real quality of the distribution backlog

The main article already established that Telsys's distribution rebound was one of the biggest moves in the 2025 cycle, but also that the key question was not whether orders rose, but what kind of orders sat inside that rise. This follow-up isolates only that issue: how much of the distribution backlog is actually firm, how dependent it is on Nvidia through subcontractors, and what would have to happen for 2026 to look like durable growth rather than a good year built on timing and one customer ecosystem.

Four points frame the picture:

• Backlog almost quadrupled, but the hard core got smaller. Distribution backlog rose to $51.7 million at the end of 2025 from $13.1 million at the end of 2024, yet the non-cancelable piece fell to $2.6 million from $4.7 million.
• The annual customer agreements do not really lock in volumes. The company says those agreements usually set prices and terms, but not quantities. The real quantities arrive through separate orders, and some blanket orders merely force Telsys to hold minimum inventory.
• Cancellation rights are not legal fine print. If the customer delays a pull or cancels an order, the company bears the cost without a return right to the supplier, except for limited scrap arrangements.
• Nvidia is not just another large customer. It is the demand axis of the rebound. The end customer Nvidia bought $35.982 million in 2025, about 26.4% of consolidated sales and roughly 64% of the segment's external revenue, through direct and indirect purchases via subcontractors.

The headline is backlog, but the quality deteriorated

This is the core of the story. At the end of 2025, distribution backlog looked almost four times larger, $51.7 million against $13.1 million a year earlier. But inside that jump there was a sharp downgrade in quality. At the end of 2024 about 35.9% of backlog was non-cancelable. At the end of 2025 that share had fallen to about 5.0%. Put differently, Telsys ended the year with more backlog, but with less contractual certainty.

The implication is not just statistical. The company explicitly says that annual agreements with some customers usually set prices and terms, but not quantities. The real quantities are ordered from time to time. That means the enlarged backlog cannot be read as if it were a hard purchase commitment. It is better read as a much more flexible purchasing intention that the customer can still change until 45 days before the planned delivery date.

That also explains why the end-2025 comparison against end-2024 flatters the headline so strongly. The company itself says the unusual increase in backlog came from the fact that its main end customer, Nvidia, entered most of its 2025 orders only at the beginning of 2025 rather than during 2024. In other words, part of the improvement is an order-timing effect, not necessarily a broader and more durable demand base.

The cancellation right matters because the risk still sits with Telsys

In electronics distribution, backlog quality is not defined only by whether an order can be canceled. It is defined by who carries the cost when something changes. At Telsys the answer is blunt: if the customer decides not to pull on time or to cancel, the company bears the cost without a return right to the supplier, except for limited scrap arrangements with some suppliers. At the same time Telsys acts as principal, recognizes revenue on a gross basis, carries inventory risk and manages that inventory itself.

That is why cancelable backlog is not just "lower quality" backlog. It is backlog where the option sits with the customer and the risk stays with the distributor. There are some limited return arrangements, generally around 2% to 5% of purchases made before the return date, and in the case of Lattice the allowance is 3.5% of purchases made in the prior six months. But that is nowhere near enough to turn $49.1 million of cancelable orders into a minor exposure.

The early-2026 cancellation data also needs to be read in proportion. From January 1, 2026 through February 23, 2026, cancelable orders worth about $2 million were canceled, roughly 4.1% of the cancelable backlog. That is clearly better than the 6.6% cancellation rate recorded in early 2025 against the end-2024 backlog, but it is still only a sub-two-month window. It says the first cancellation wave was not abnormal. It does not prove that the backlog as a whole is high quality.

There is another layer that matters. Some blanket orders include minimum quantities that the company is required to hold throughout the period. So even before demand becomes firm, Telsys can already be required to carry inventory against it. That means the 2026 question is not only how much of backlog stays in place, but how much inventory the company may have to carry while serving a backlog that remains mostly customer-flexible.

Nvidia is the demand center, not one layer inside a broad customer mix

Once you see 17 customers accounting for 85% of sales, it is easy to assume the mix is reasonably diversified for a distributor. That is only a partial read. The company itself identifies Nvidia as the segment's material end customer and says purchases for that customer are made both directly and indirectly through subcontractors. In 2025 that reached $35.982 million, almost triple 2024, and against the segment's external revenue it was equal to roughly 64%.

That chart matters because it shows that the rebound and the increase in concentration happened in the same move. So it is hard to argue today that Telsys returned to growth through a broader customer base. A better way to read it is that the company benefited from a very strong acceleration in one customer ecosystem, and that it still has not proven that this momentum has broadened into several independent demand engines.

This is also why the end-2025 backlog looked so impressive. The issue is not just whether Nvidia bought more. The issue is also when those orders were entered, through which buying entities they flowed, and how much of them remained open to cancellation until close to delivery. That is exactly the difference between volume and quality.

Subcontractors blur the sense of diversification

Another number that forces a reread is the customer mix of the segment. In 2025 about 67% of distribution revenue came from Contract Manufacturing companies, versus 59% in 2024 and 58% in 2023. Electronics companies themselves fell to 33% of revenue. The implication is that anyone reading the customer list or the country map has to remember that a growing share of demand is coming through the manufacturing and assembly layer, not necessarily through a broader base of end customers.

This is where geography also needs to be read carefully. In 2025 about $28.1 million of distribution sales were booked in Singapore, against only $1.9 million in 2024, while China fell to $0.5 million from $5.0 million. That is not proof of newly diversified end markets. In a setup where the company itself stresses purchases through subcontractors, the customer country often tells you more about the manufacturing and assembly point through which the order flowed than about the true breadth of the end-customer base.

That is why customer counts alone can mislead. If the same end customer feeds demand through several manufacturing entities, the legal spread of invoices will look wider than the real economic spread. That does not mean the demand is fake. It means the right test is the breadth of end programs and end customers, not just the number of invoices or ordering entities.

What would make 2026 durable

The investor presentation gives a clue to what management is trying to build next. It points to high demand in the Enterprise segment, a positive trend in industrial, a large scale of design opportunities in Enterprise and Military, and an expansion of the Foundry Service activity with GlobalFoundries. That is the right direction. But for now it is still not proof that the economic spread has already changed.

For 2026 to look more durable, three things need to happen together. First, a bigger share of backlog needs to become NCNR, or at least show a low cancellation rate over time rather than only in a short early-year window. Second, the new opportunities need to turn into orders coming from more programs and more end customers, not just into an even heavier Nvidia weight through subcontractors. Third, the volume needs to start producing better economics as well. In 2025 the distribution segment generated $4.023 million of segment profit on $56.042 million of revenue, roughly 7 cents of segment profit for every revenue dollar. The volume is working, but it is still not a strong profit engine.

That is the conclusion of this follow-up: Telsys's distribution backlog looks much stronger than the economic quality it currently represents. There is real demand recovery here, there is a strong end customer, and there are also signs that management is trying to build a broader opportunity funnel. But until the hard portion of backlog grows, until concentration widens beyond Nvidia, and until the volume flows through a mix that is less dependent on subcontract manufacturers, 2025 still looks more like a powerful opportunity built on one ecosystem than a full proof point of a stable, broad distribution engine.