In a shortage, revenue growth and market-share loss coexist — check units, not dollars

Claim

Intel's Q2 2026 was its fastest revenue growth since 2011 (+25%), with Data Center & AI +59% at a 39.3% segment operating margin. Over the same window Mercury Research has Intel's x86 share falling 72.9% → 67.4%, AMD reaching a record 32.6% overall, 33.2% of server units — and 46.2% of server revenue.

Both are true. The reconciliation is a physical CPU shortage: server CPU prices up 10–20% since March 2026 with another 8–10% expected in 2H26, against an 11–14M unit global shortfall, with Intel reportedly filling only ~40% of distributor allocations on 30+ week lead times.

The revenue line was measuring the shortage. The share line was measuring the business.

Two second-order observations worth carrying forward:

• AMD held 33.2% of server units but 46.2% of server revenue. When the revenue-share gap exceeds the unit-share gap, the loser is retaining the cheap sockets and losing the profitable ones. That is a worse position than the unit number alone implies, and it is invisible in either company's revenue growth rate.
• A vertically integrated manufacturer losing share because it cannot make enough product is indicting its own core structural advantage. "It's capacity, not product" is a weaker defence for an IDM than for a fabless firm.

Guard

When a cyclical posts unusually strong revenue growth, before crediting it to competitive recovery:

1. Find the unit-share series (Mercury, IDC, Gartner, Counterpoint — by segment) and check direction. Revenue share and unit share diverging is the signal.
2. Check whether ASPs moved industry-wide. If every player's ASPs are up, the growth is the market's, not the company's.
3. Ask what the number looks like without the shortage before annualising it into a valuation. In the INTC case the difference between annualising the record quarter and normalising it was roughly $67–83 versus $26–49 of fair value — larger than any other assumption in the model.
4. Date the normalisation. Shortages end; put the expected quarter in triggers so the recheck fires on the event rather than on an interval.

What would falsify this

Unit share stabilising or reversing while the shortage persists would show the growth had a product component after all. For INTC specifically: x86 share holding above ~67% into 2027, or DCAI sustaining >20% growth after supply normalises.

Related

[[principle-down-a-lot-is-not-cheap]] · [[principle-story-vs-revenue]] · [[pattern-ai-levered-fields-trade-above-own-band]] · [[Themes/semi-equipment-and-test]]

History

• 2026-08-04 — recorded during the INTC re-analysis, where it was the decisive tension between the Fundamentals and Moat readings and the reason the verdict came out TRIM despite two consecutive large earnings beats.

---

Amendment 2026-08-27 - the falsification condition FIRED, against the bearish read

This note set an explicit test: "Unit share stabilising or reversing while the shortage persists would show the growth had a product component after all." Mercury Research Q2 2026 delivers exactly that, and the note must be narrowed accordingly.

Segment (units)	Q1'26	Q2'26	YoY
Overall x86 (AMD)	32.6%	34.1%	+4.7pt
Client	29.6%	30.3% (first time >30%)	-
Desktop	33.1%	34.9%	+2.7pt
Mobile	-	28.9%	+8.4pt
Server (ex-IoT)	33.2%	34.5%	+7.3pt

The decisive detail: Intel added millions of units of mobile capacity in Q2 after two supply-constrained quarters, materially narrowing supply against demand - and still lost 8.4 points of mobile share year-over-year. Supply relief did not restore share. That is the falsification condition met on its own terms.

What survives, and what is narrowed

Narrowed. The note was written from Intel's side and read AMD's gains as partly shortage-assisted. On the Q2 evidence AMD's share gains are structural. The guard should not be applied symmetrically: for the loser in a shortage, revenue growth and unit share genuinely decouple; for the winner whose unit share rises in every segment while ASPs rise industry-wide, the growth has a real product component.

Survives, and still bites. "Before annualising, split the growth into share, price and mix." AMD's server revenue growth (and its guided +>70% for 2027) is still part share and part a 10-20% industry-wide server CPU price increase since March 2026. And the shortage is actively costing AMD elsewhere: Client & Gaming operating income fell 24% YoY on +6% revenue, with desktop unit shipments down >20% globally - AMD is taking a bigger slice of a shrinking pie, and Gaming is -31%. A rising unit share is not the same as a rising profit pool.

Correction to this note's own numbers - a basis conflation

This note recorded 46.2% as AMD's Q1-2026 server revenue share. The 46.4% now widely quoted for Q2-2026 is a different metric - Mercury's narrow-basis unit share counting only Xeon SP vs EPYC, which strips out Intel's low-end/networking/storage silicon. Press coverage (Tom's Hardware included) blurs the two, and no distinct Q2-2026 server revenue-share figure could be located.

Do not carry "46.2% revenue share" forward as a Q2 number. The figure to stand behind is 34.5% server units. The unit/revenue divergence conclusion - Intel retaining the cheap sockets - still holds and is one quarter deeper, but state which basis you are on every time.

The second-order finding that outlives the shortage

The x86 pool itself is shrinking as a share of datacenter compute, and not because of Intel. Mercury Q2-2026 puts Arm at 13.6% of server units and 15.3% of client, both records. Every Nvidia rack ships an Arm CPU and every AMD rack ships an EPYC - so AMD's CPU franchise is partly hostage to its own GPU share. Winning x86 share is compatible with losing datacenter CPU relevance. (Separately-circulated claims of "Arm >45% of datacenter CPU revenue" count Grace/Vera CPUs bundled into GPU racks at rack ASPs - a different basis from Mercury's 13.6%. Do not average them.)

History (continued)

• 2026-08-27 - falsification condition fired during the first-ever /analyze AMD. Claim narrowed to the losing side of a shortage; a basis conflation in this note's own evidence corrected; Arm added as the durable second-order threat. decay: cycle retained - the shortage is still on (EPYC sold out through year-end, 30+ week lead times, ~70% of 2026 DRAM allocated to AI).