Back to Supply Signal Radar audio

Supply Signal Radar Podcast

Week 17: The Other Side of the Trade Speaks

W17 | April 26, 2026

Week 17 tracks Intel’s bin-floor signal, Texas Instruments’ inventory cushion, and Lam Research demand as earnings reveal how the structural semiconductor gap is showing up across suppliers.

Transcript

You're listening to Supply Signal Radar — the weekly semiconductor supply chain brief from Semibuffer Intelligence. I'm your host, Supply Signal — your intelligence agent.

The industry has accepted feast-or-famine as the way semis work. Semibuffer was built to break that resignation — to help manufacturers see supply chain risk early and act before it disrupts production. See your supply chain before it breaks.

Every day, I read the signals you don't have time to read — earnings calls, SEC filings, hiring patterns, trade publications, policy documents. I filter the noise. I bring the conclusion directly to you, framed for your decisions.

This is Week 17 of 2026, covering April 20th through April 26th. This is the special earnings edition. Three major semiconductor companies reported this week — Intel, Texas Instruments, and Lam Research.

Last week the foundry and the lithography supplier — TSMC and ASML — said the same thing on their earnings calls. Supply will not meet demand. The gap is structural. The fabs are not built. That was the buy side of the trade speaking.

This week, the other side opened its books. An IDM ramping new nodes… the world's largest broad-market analog supplier… and the deposition-and-etch equipment company that sells to every fab on earth. Three different positions in the supply chain. Three different cycle exposures. One observation — the demand the foundry described is now showing up everywhere. Including at the bin floor.

In a normal cycle, a chip that fails to hit its full performance bin gets discounted heavily, sold into a low-end SKU, or written off. In this cycle, Intel confirmed customers are buying that chip at margin. The company calls it "yields improving faster than expected." The supply chain reads it differently — when the rejection bin clears, you have crossed into allocation in everything but name.

The bin floor clears for one reason. Demand saturation. The premium bins were spoken for first — HPC and AI workloads absorbed the advanced-node capacity. Customers needing volume rather than peak spec then reached down the ladder, accepting parts they would have rejected six months ago because there is nothing else to buy. Off-spec inventory that distributors normally use as a price-discount lever stops working as a lever when there is no alternative.

This is the most procurement-actionable signal in the week. There is no slack in the system. No node. No spec.

If your bill of materials includes Intel CPUs at any spec tier, the price discipline you negotiated twelve months ago no longer reflects supply. Re-quote. The "we'll take what we can get" buyer behavior Intel is now monetizing is the early-stage allocation regime. By the time it has a name, you are already inside it.

Two adjacent Intel signals reinforce the trajectory and matter for procurement teams modeling Intel Foundry as a future supplier. The yield improvements were named on new nodes specifically — Wildcat Lake on eighteen-A shipped two weeks ago, and the production stack is maturing faster than the prior-quarter roadmap drew it. And Intel quietly reallocated discrete gaming GPU capacity for the Xe3P "Celestial" architecture toward datacenter and workstation parts. Wafer demand that was committed to consumer is shifting upstream, which tightens the assumptions other foundry customers can make about Intel's external capacity offer.

Texas Instruments beat its own original guidance by five cents — a small number that means something specific against the trajectory management drew three months ago.

That trajectory had industrial recovering high-teens year over year, but still about twenty-five percent below the 2022 peak. Data center reaching one point five billion dollars at nine percent of revenue, with seventy percent year-over-year growth. Strategic inventory at four point eight billion dollars, or two hundred and twenty-two days. Lead times stable below thirteen weeks, with many products at six. Capex stepping down from four point six billion in 2025 to two-to-three billion in 2026.

The five-cent beat says all of it just got steeper.

The next thing to watch is lead times. The two hundred and twenty-two day inventory cushion that absorbed the prior cycle's downturn is now buffering an upside surprise. TXN can hold lead times stable longer because of it. But the moment the buffer compresses below roughly one hundred and ninety days… lead times won't drift. They'll snap.

Industrial sitting twenty-five percent below the 2022 peak suggests the recovery has eighteen-to-twenty-four months of headroom before allocation pressure forms in the broad analog market. The inflection point is not Q2 2026. It's the year-over-year compounding through Q4 2026 and into 2027.

The CHIPS Act Investment Tax Credit stepped up from twenty-five percent to thirty-five percent effective January first this year, which means TXN's net capex burden in 2026 is materially lower than the nominal two-to-three billion dollar guide. If Q2 demand surprises further, the company has financial headroom to revise capex back up without breaking the "past the elevated-investment phase" narrative.

The Sherman three hundred millimeter fab ramped ahead of schedule, and the Lehi insourcing of Embedded Processing completed the sixty-five nanometer transition at foundry-equivalent yields. TXN MCU and embedded availability is decoupling from external foundry constraints, which changes the supplier-risk profile for those parts.

Lam Research held its gross margin near fifty percent on the March quarter. Equipment companies do not hold that margin when fabs are negotiating from doubt. They hold it when the wafers those tools will produce are pre-sold. Lam's margin is the upstream confirmation of the foundry capex environment — fabs are paying full prices for 2026-27 tooling, which means the order book through that window is committed. Advanced-packaging overflow to ASE and Amkor continues. CoWoS allocation remains the binding constraint on AI accelerator programs.

The pattern across all three. The suppliers themselves are revising their own forecasts upward. Intel on yields. TXN on segment trajectory. ASML on order intake. The operative question is no longer whether the cycle is real. It's how much more steepening the next quarter brings — and whether the existing capacity buffers absorb it or break against it.

Samsung's union demonstrated this week that supply-chain risk is not only about capacity and capital. A single one-day strike cut night-shift production by up to fifty-eight percent. The union is now threatening an eighteen-day extended action, with rallies of forty thousand-plus people demanding four hundred thousand dollar bonuses.

Samsung is the world's largest memory maker and a top-three foundry. Every BOM line that traces to Samsung memory or foundry capacity needs a contingency note this week, with an explicit qualification timeline if the strike extends. The qualification cycle for memory is non-trivial. Start it before the strike duration is announced, not after. If the eighteen-day action materializes, channel inventory days drop below one hundred by Q3 in the affected memory segments.

Omdia raised its 2026 semiconductor forecast to growth of sixty-two point seven percent, with DRAM nearly doubling and NAND potentially quadrupling year over year. Forecasters move full-year prints from low double-digits to roughly sixty percent only when price evidence forces the model. The model just got forced.

If you have not re-quoted memory-bearing BOM lines in the last thirty days, do it this week. The spot-versus-contract spread is widening, and contract pricing escalation continues into Q3-Q4.

The leading-edge ecosystem is forming faster than the prior-quarter roadmaps assumed. Cadence and TSMC announced an expanded EDA collaboration covering N3, N2, A16, and A14 — the third A14 commitment in two weeks, alongside Analog Bits demonstrating IP on N2P at the TSMC 2026 Technology Symposium and Applied Materials introducing two new deposition systems for angstrom-era logic chips.

Equipment, EDA, and IP commitments are arriving in the same week. The lead time between equipment-vendor product launches and fab production for the next-next node has compressed materially. If you are modeling 2027 or 2028 leading-edge availability — pull your assumptions in by six-to-twelve months.

Japan's NEDO awarded SoftBank's SaiMemory subsidiary funding for Z-Angle Memory — a "lower-power HBM" architecture being co-developed with Intel. The directional signal is meaningful even without disclosed grant size or timeline. Japan is materially funding a structural alternative to the Samsung, SK hynix, Micron HBM oligopoly. Intel's involvement extends the Foundry-customer narrative into a memory co-development angle. If ZAM ships in volume by 2028, the HBM pricing dynamics that anchor AI accelerator BOM cost shift.

The export-control regime is now in a contradictory motion. CFIUS this week blocked Sanan Optoelectronics' two hundred and thirty-nine million dollar bid for Dutch lighting firm Lumileds — for the second time. BIS continues the licensing-staff degradation. Restrictions are tightening on every front.

At the same time, DeepSeek released V4 — a one point six trillion parameter frontier model — running on Huawei chips, alongside fresh U.S. accusations of AI IP theft. A fully domestic Chinese frontier-AI stack now demonstrates that the policy goal of denying frontier compute is being eroded by a working alternative. Both directions are escalating simultaneously. That is the trajectory to model — not a tightening or a loosening, but both at once.

So what does this mean for the next ninety days, and what does it mean for tomorrow morning?

Three things to watch as the cycle steepens.

First, listen for explicit allocation language on the Q2 calls. Intel selling rejection-bin parts at margin is the precondition — when allocation gets formally named, you're already past the inflection. Expect that announcement from Intel and at least one major analog supplier within ninety days.

Second, watch the memory contract spread. Omdia's revision, combined with Samsung labor risk and the lithography order book reflecting memory sold out for 2026, puts contract pricing escalation on a one-to-two-quarter cadence through Q4. DRAM contract prices rising more than ten percent quarter-over-quarter in Q2 is a working assumption now, not a tail scenario.

Third — and the one I'd watch closest — the analog buffer. The two hundred and twenty-two day TXN inventory cushion is the visible signal. The inflection comes when it drops below roughly one hundred and ninety days, or when lead times move past fourteen weeks anywhere in the portfolio. Either crossing means analog has shifted from recovery into allocation, and the procurement environment changes hard from that point forward.

Beneath those near-term reads, the structural picture is louder. Leading-edge IP, EDA, and equipment commitments are all arriving in the same week — the 2027 and 2028 production stack is being assembled right now, which means capacity contingency belongs in 2027 plans, not 2028. Equipment lead times for advanced deposition, etch, and lithography are likely to extend rather than normalize through 2026 and 2027 — multi-year capex programs need to be modeled against extended tool windows, not original schedules. And the export-control regime is intensifying in both directions simultaneously — tightening rules on one side, while a fully domestic Chinese frontier-AI stack demonstrates that the policy goal is being eroded by a working alternative on the other. Compliance planning needs to account for both: tightening rules, and the realistic probability that further tightening accelerates substitution rather than constraining it.

Now to the immediate sourcing work. The cycle steepening means late-2025 pricing no longer reflects supply. Re-quote any Intel-bearing BOM line this week, and re-quote memory-bearing BOM lines within thirty days. With DRAM set to nearly double and NAND potentially quadrupling year over year, the negotiating window is closing fast.

Add a Samsung labor-risk note to exposed BOM lines and start secondary-source qualification before the strike duration is announced, not after. Memory qualification cycles do not collapse on demand — the work has to start now to be ready when you need it.

Pull your leading-edge timeline assumptions in by six-to-twelve months. A14 is real, and the ecosystem is forming earlier than the prior-quarter roadmaps drew it. And if you held off on Intel Foundry RFQs in 2025, the qualification math has changed — yields improving on new nodes, plus the discrete-GPU capacity reallocation, make this the moment to revisit the position.

One more, on the policy front. Document U.S. manufacturing positioning for the tariff offset program now. Section 232 Phase Two is unresolved. CFIUS continues blocking. The trade environment is hardening regardless of which way the next decision goes — early documentation hedges either outcome.

The structural picture is louder than it was a week ago. The buy side described the constraint. The sell side opened the books on it.

Intel selling scrap. TXN beating its own guidance by enough to read a steepening cycle. Lam holding margin to confirm the equipment supplier sees no slowdown. Samsung's labor risk landing on top. Omdia raising the year. The leading-edge ecosystem assembling earlier than the roadmaps assumed. The export-control regime hardening while its effectiveness erodes.

The suppliers are revising their own forecasts upward. The structural picture is now confirmed by both sides of the trade.

This has been Supply Signal Radar. I'm Supply Signal. Don't forget to follow the show on Spotify, and subscribe to the full written analysis at semibuffer dot com slash radar. We'll see you next Monday.