Vecstrata · Backtest · June 2026
A backtest: scored history, not a live call. We froze the vantage at 2013-05-16, the day Brian Krzanich became Intel's CEO, derived four predictions from only what was knowable that day, and scored them against what actually happened through 2020. A backtest has no edge, because hindsight is undefeatable, so this is never marketed. Its only product is what running the machinery blind on a resolved case teaches. We score backtests on a separate ledger from the live record, and the Brier below is the honest, un-cherry-picked number.
The Call (as of 2013-05-16)
Intel had just handed the CEO seat to Brian Krzanich with an explicit mandate to fix mobile. The company had missed the smartphone entirely, with no design win in any flagship phone, yet it still led the industry in process technology and printed roughly 62% gross margins. The easy read was that a dominant, cash-rich leader with a new CEO would eventually buy or build its way into mobile. The structural toolkit reads it the other way, and more precisely: it predicts not just that Intel struggles, but exactly where.
The binding question is not transistor performance, which Intel led. It is the volume that amortizes a leading-edge fab, and all the growth in silicon volume was in mobile, which Intel did not serve, while its own volume base in PCs was shrinking. Underneath that sits Christensen's resource-allocation process: a company organized to feed a 62%-margin franchise structurally starves a low-margin mobile business, whatever the CEO mandates (the machinery is spelled out below). The same fabs that are a moat in PCs are dead weight in mobile, where ARM's nickel royalty and TSMC's foundry-amortized capacity are the cost structure that fits. The scope is global: the smartphone, foundry, and datacenter silicon markets are worldwide.
So the lens makes a sharper claim than "Intel declines": disruptable where the model is misaligned, durable where it fits. The fourth call is the control built to test it. We score four calls at confidences set blind from May-2013 uncertainty, all resolved through 2020. The stack lands at Brier 0.150, with its costliest line the boldest and correct one.
The Situation
Vantage facts only: dated, sourced, knowable on or before 2013-05-16. The analysis comes after.
- 2011-05-04
Intel announces the first 3-D Tri-Gate (FinFET) transistors in high-volume manufacturing at 22nm: the process lead established, the asset everyone assumed was unassailable (Intel). - 2012-01-26
TSMC's Europe president claims ARM on TSMC 28nm beats Atom on Intel 22nm FinFET on performance per watt: the node lead does not translate to a mobile system win (Electronics Weekly). - 2012-09-18
Intel's debut phone, the Motorola RAZR i on a single-core Atom, leads on browser benchmarks but trails the dual-core Snapdragon on most others: a late, sub-flagship entry competitive only on a narrow axis (Engadget). - 2012-11-09
Qualcomm's market cap (~$105B) surpasses Intel's (~$104B): a fabless mobile-SoC designer overtakes the integrated manufacturer in value. The value is already migrating to the modular layer at the vantage (EE Times). - 2012-12-10
Apple is reported to be shifting A-series production to TSMC: leading-edge mobile volume aggregating onto the pure-play foundry (MacRumors). - 2013-02-19
Intel's FY2012 10-K: revenue $53.3B at 62.1% gross margin, the high-fixed-cost model a mobile SoC business would have to clear and cannot. The same filing shows Data Center Group at $10.7B revenue and roughly 47% operating margin, growing on cloud, the segment where Intel's model is aligned (SEC 10-K). - 2013-02-25
Intel opens its leading-edge fabs to an outside logic customer, Altera on Intel 14nm: a tacit admission the integrated manufacturer needs external volume to amortize the fab, the constraint surfacing in Intel's own actions (Intel). - 2013-04-10
Worldwide PC shipments fall 13.9% year over year, the steepest decline on record: the x86 volume base that amortizes the fab is eroding from under Intel (Business Wire / IDC).
The Mechanics
The resource-allocation process starves the mobile business
A CEO mandate is the loud signal, and it is the wrong place to look. Underneath sits Christensen's resource-allocation process, the day-to-day machinery that decides where capital and the best engineers actually go. At Intel that machinery is tuned to a 62%-margin x86 franchise. A mobile SoC sells at a fraction of that margin, so every internal hurdle rate, every capacity-allocation decision, and every promotion incentive points away from it. The business is starved from inside, structurally, no matter how much the CEO wants it to win or how much cash stands behind it.
This is the call the easy read cannot make. A surface analysis sees cash, a process lead, and executive will, and concludes Intel will eventually force its way in. The resource-allocation lens says the very thing that makes Intel strong, a high-margin franchise the whole company is organized to feed, is the thing that makes the mobile effort impossible to fund to victory.
Mechanism: a high-margin incumbent's resource-allocation process cannot clear the internal hurdle for a low-margin adjacent business, so it starves the effort regardless of mandate or cash.
The moat rests on the volume base, not the lead
Intel's process leadership looked unassailable, and the toolkit predicts it falls, which was the minority view in 2013. The reason is a theory-of-constraints reading of a thing usually told as an execution story. A leading-edge fab is a fixed-cost asset, and the binding constraint is the volume that amortizes it. As node costs explode, the foundry that pools the entire fabless ecosystem's volume can out-invest a captive manufacturer running only its own shrinking x86 base. The lead is downstream of the volume, and the volume was leaving: mobile growing on TSMC, PCs declining, Intel opening its own fabs to Altera to find more.
So the right thing to watch was never the current node lead. It was the volume base beneath it, which the vantage signals already showed eroding. The lead was a lagging indicator of a constraint that had already turned.
Mechanism: when a moat rests on a fixed-cost asset, the binding constraint is the volume that amortizes it, and pooled ecosystem volume out-amortizes a captive base as costs rise.
The lens discriminates: resilience where the model is aligned
The toolkit would be worthless if it just said "Intel declines." Its value is that it predicts where. In datacenter, Intel's cost structure and value network fit: high margin, performance-led, x86-locked, riding the cloud build-out. The same lenses that call disruption in mobile call resilience here, because none of the misalignment that dooms the mobile effort is present. Server is where a 62%-margin model belongs.
That is why the fourth call is a control. If datacenter had collapsed on the same timescale as mobile, the lens would not be discriminating, it would just be bearish on Intel. Its holding is what shows the reading is structural: disruptable where the model is misaligned, durable where it fits.
Mechanism: the same cost-structure and value-network lenses predict resilience where an incumbent's model fits, so a correct call splits the firm by alignment rather than condemning it whole.
The Predictions
Each card was derived blind from the vantage facts, scored at its May-2013 confidence, and resolved against the record. The fourth is a control that tests whether the lens discriminates.
1 · Intel fails to build a profitable mobile-SoC business ·✓ HIT 2019-07-25
Despite a new CEO and multi-billion-dollar investment, Intel fails to win meaningful, profitable smartphone application-processor share by the horizon. Its resource-allocation process and 62%-margin cost structure structurally block a low-margin mobile business, so the effort is abandoned, not won.
Confidence 68% · Horizon 2020-12-31
Wrong if: Intel captures ≥10% of smartphone AP unit share, or runs its phone-SoC business at sustained operating profit.
2 · Intel loses process leadership to TSMC (the contrarian leg) ·✓ HIT 2020-07-23
Intel loses the process-technology lead to TSMC by the horizon: the pure-play foundry reaches volume on a leading node ahead of Intel, and Intel publicly concedes it is behind. At the vantage Intel held a clear multi-year lead, which is what makes this the minority call.
Confidence 52% · Horizon 2020-12-31
Wrong if: Intel ships its leading-edge node in volume at or ahead of TSMC throughout the horizon, keeping undisputed leadership.
3 · Value migrates to the modular layer ·✓ HIT 2020-11-10
The value displaced from the integrated manufacturer migrates to the modular layer, foundry (TSMC), architecture (ARM), and fabless designers (Nvidia, Qualcomm, Apple silicon). By the horizon at least one of TSMC or Nvidia overtakes Intel by market cap, and a marquee customer drops Intel for in-house ARM silicon.
Confidence 58% · Horizon 2020-12-31
Wrong if: Intel remains the most valuable US chipmaker and keeps its marquee CPU sockets, with no major customer defecting to in-house ARM silicon.
4 · The datacenter franchise holds (the control, where the model is aligned) ·✓ HIT 2018-12-31
Where Intel's cost structure and value network are aligned, datacenter and server, the same toolkit predicts resilience, not disruption. The server franchise holds up materially better than mobile, dominant well into the horizon before any erosion begins. If this control also collapsed early, the lens would just be saying "Intel is bad at everything."
Confidence 70% · Horizon 2020-12-31
Misses if: datacenter loses dominant server share (below ~75%) before ~2019. Hits if: it stays dominant and growing, eroding only late and slowly.
How they resolved. All four hit. The mobile block held exactly as the resource-allocation lens predicted: Intel never cleared 1% of smartphone share, "contra revenue" confirmed the margin trap as Intel paid OEMs to offset its cost gap, and the business was cancelled in 2016 with the modem remnant sold to Apple in 2019, abandoned not won (P1). The process lead fell when Intel admitted 7nm was a year behind while Apple's chips already ran on TSMC 5nm (P2). Value migrated on schedule: within five months in 2020 both TSMC and Nvidia passed Intel by market cap, and Apple announced and shipped its own ARM silicon (P3). And datacenter held, revenue more than doubling while share stayed near 99% through 2018, before AMD and custom silicon began a late, slow erosion (P4). The costliest line was P2 at Brier 0.230, the deliberately humble leg: predicting the fall of a multi-decade process moat was near a coin flip from the vantage, and the crossover took seven years. That the boldest, correct call also carries the highest cost is the calibration working.
Counter-signals, measured. Three independent agents re-derived the four probabilities from the vantage file alone, no outcomes and no author numbers. Their blind consensus Brier (0.212) came out worse than the author's (0.150), though Intel's decline is canonical tech history, so the premium is a weak lower bound. On the three disruption legs the skeptic, pricing Intel's cash and mandate and process lead as a real chance of success, pulled the panel down, while inside-view reasoning from cost-structure logic reached the calls confidently and beat the author. That is direct evidence the calls are structural, not narrative: mechanistic reasoning reproduces them, and doubt about Intel's resources is what loses. On the control the panel was more confident than the author (0.83 vs 0.70), reproducing the aligned-franchise resilience call independently and harder.
Backtests are scored on their own ledger, separate from the live record, and labeled as backtests: full Scorecard →.
The chain
Every framework here traces to a named source in the strategy literature. The load-bearing ones, each tagged for whether we took it from the source or inferred the application:
- Resource-allocation process. Clayton Christensen, The Innovator's Dilemma; drawn from source. The reading that a 62%-margin process starves a low-margin mobile business is our application.
- Cost structure and the value network. Clayton Christensen, The Innovator's Dilemma; drawn from source. The reading that x86 lock-in is worth nothing in the mobile value network is ours.
- The binding constraint. Eliyahu Goldratt, theory of constraints; drawn from source. Reading fab-amortizing volume, not the node lead, as the constraint is our application.
- Law of Conservation of Attractive Profits. Christensen and Raynor, The Innovator's Solution; drawn from source. The margin migrating to the modular foundry-and-fabless layer is ours.
The full chain, every lens and graph node tagged drawn-from-source or inferred, lives in the prediction files, down to the signal.
What the case taught
The case bore out the discriminating claim, disruptable where the model is misaligned and durable where it fits, and left three refinements for the live taxonomy:
- Name the role, distribute the firm, and it worked here. P3 named the roles, foundry and architecture and fabless, with multiple candidate claimants rather than betting one firm, and resolved cleanly across TSMC, Nvidia, and Apple. The named-entity humility lesson from earlier backtests paid off prospectively, so we carry it as standing practice, not just a post-mortem note.
- Watch the volume that amortizes the asset, not the asset's current lead. When a moat rests on a fixed-cost asset, the binding constraint is the volume base beneath it. This reads directly onto the live
power-binding-constraintandgraphitetopics, both fixed-asset-amortization stories. - Timing is the irreducible risk on structural calls. Every leg was right on direction; the only real uncertainty was whether it lands inside the horizon, and P2 took seven years. Price the horizon-fit risk explicitly and set the horizon generously, a discipline the live 2027 cohort already reflects.
Related analyses
- No Moat, And Neither Does the Template: the open-weight backtest carries the same name-the-role-distribute-the-firm discipline, and the same finding that timing, not direction, is the hard part of a structural call.
- The Constraint Is Not a Baton: the live power issue runs the volume-amortization lesson forward, watching the fixed-asset volume base that sets where the scarcity rent pins.