NASDAQ: NVDA Pre-print synthesis 14 calls | 4 supplier checks

Channel Check Synthesis – NVDA – Blackwell/Rubin Demand Read

Synthesized from 14 expert calls + 4 supplier checks | Published May 26, 2026 (1 day pre-print) | Q1 FY27 reports May 27 AMC

We aggregated client-uploaded Tegus, AlphaSense, and Third Bridge transcripts (14 calls between Apr 2 - May 24, 2026), public supplier disclosures (TSM, ASML, MU, AVGO), and our own primary supplier checks (SK hynix HBM3e, Samsung HBM4 sample shipments, Foxconn Tier-1 assembly). Output: a single ABI-synthesized read on Blackwell-to-Rubin demand transition, GM trajectory, supply bottlenecks, and how hyperscaler capex is tracking vs implied. Designed to be the single document you read 30 minutes before the print.

Expert calls

avg 38 pages

Supplier checks

TSM, MU, AVGO, ASML

Transcript pages

534

aggregate

ABI conviction

8.6/10

vs sell-side 6.8

Demand momentum

9 / 10

accelerating

Pricing trajectory

↑ +6-9%

Rubin ASP vs Blackwell

Q1 FY27 setup | ABI vs consensus

$B revenue | waterfall

LIVE

Consensus: $43.2B | Whisper: $44.6B | ABI: $45.4B (+5% beat) | Q2 guide ABI: $48.8B (vs cons $47.4B)

ABI View Source Mosaic Consensus Matrix End-Market Pricing / GM Supply Heat-map Hyperscaler Capex Source Quotes Risk to Read Ancillaries How to Position

ABI Synthesized View | Pre-Print

Five bullets. This is the conclusion. Everything below is supporting evidence.

What 14 calls + 4 supplier checks tell us

Demand for Blackwell remains supply-constrained through Q3 FY27. 11 of 14 experts described order books as "fully booked" or "fully booked through Q4 FY27." This is consistent with our supplier read on TSM CoWoS-L capacity (52K wpm exit-2026 vs Mar consensus 45K wpm) – incremental capacity is being absorbed instantly.
Rubin transition is on track and ASP-positive. 9 of 14 experts confirmed Rubin sample units shipped to top-3 hyperscalers in April. Implied ASP per system +6-9% vs Blackwell Ultra; specifically driven by HBM4 content (8 stacks vs 6, $48K ASP delta) and CPX co-packaged optics adoption.
Hyperscaler capex is not the risk many are positioning for. Combined MSFT/META/GOOG/AMZN FY26 capex now tracking $418B vs Mar implied $402B. Within that, GPU mix is rising from 31% to 38% of total infra spend. The risk is supply, not demand.
Pricing power on Blackwell holds even with AMD MI400 ramp. 8 of 14 experts characterized AMD as taking 5-8% wallet share in Q1, but at lower ASP – net NVDA volume is undisturbed. No customers reported pulling Blackwell orders to wait for MI400.
GM trajectory: short-term mix headwind, long-term tailwind. Rubin pre-ramp expenses are weighing on Q2 FY27 GM (we model 74.2% vs guide range high 75%). But Rubin annualized GM should hit 77-78% by Q4 FY27 on richer mix + CoWoS-L learning curve. This is the single most important commentary to listen for on the call.

ABI conviction 8.6 / 10 vs sell-side 6.8 / 10 expected post-print direction ▲ +4 to +9%

Source Mosaic | 18 Inputs

Every input transparently sourced and confidence-weighted. ABI weighting reflects tenure, role relevance, and historical accuracy of the source.

#	Source type	Role / company	Tenure	Network firm	Date	Pages	Conf weight
1	Former employee	VP Data Center \| NVDA (2021-25)	4 yrs	Tegus	May 22, 2026	42	9.4
2	Former employee	Director Compute Architecture \| NVDA (2019-24)	5 yrs	AlphaSense	May 18, 2026	38	9.2
3	Customer	CTO \| Top-5 Hyperscaler (anon)	n/a	Third Bridge	May 14, 2026	52	9.6
4	Customer	Head of AI Infra \| Tier-1 Cloud (anon)	n/a	Tegus	May 12, 2026	46	9.2
5	Customer	VP Procurement \| Top-3 AI Lab	n/a	AlphaSights	May 9, 2026	34	8.8
6	Supplier	Senior Director \| TSM CoWoS	9 yrs	Tegus	May 16, 2026	42	9.0
7	Supplier	Mgr \| SK hynix HBM Engineering	11 yrs	Third Bridge	May 8, 2026	36	8.8
8	Supplier	Sales VP \| Foxconn AI Server Group	15 yrs	AlphaSense	May 6, 2026	28	7.6
9	Partner	Sr. Engineer \| Astera Labs (PCIe retimers)	3 yrs	Tegus	May 4, 2026	22	7.4
10	Partner	Director \| Credo Tech (AECs)	4 yrs	Third Bridge	Apr 30, 2026	26	7.8
11	Former employee	Mgr Networking \| Arista Networks (2022-25)	3 yrs	AlphaSense	Apr 28, 2026	30	7.2
12	Customer	Engineering Director \| Enterprise AI (Fortune 50)	n/a	Tegus	Apr 25, 2026	38	7.0
13	Former employee	Director \| Mellanox/NVDA Networking (2018-24)	6 yrs	Third Bridge	Apr 21, 2026	42	8.6
14	Supplier	Sr. Engineer \| ASML EUV Service	8 yrs	AlphaSights	Apr 18, 2026	24	6.8
15	ABI primary	TSM Hsinchu site supplier (anon)	n/a	ABI primary	May 23, 2026	8	9.0
16	ABI primary	SK hynix Icheon HBM4 line check	n/a	ABI primary	May 20, 2026	6	9.2
17	ABI primary	Foxconn server assembly Vietnam/Mexico	n/a	ABI primary	May 15, 2026	10	8.6
18	ABI primary	Tier-2 PCB substrate volume read	n/a	ABI primary	May 11, 2026	10	7.4

Avg confidence weight: 8.2 / 10 | 9 of 18 sources at "high confidence" (≥8.5). Source overlap (same hyperscaler/supplier captured by multiple analysts) flagged in full report. Pre-print run is intentionally conservative on customer-side calls due to selection bias.

Consensus Matrix Across 14 Expert Calls

For each key question, % of expert responses falling into each bucket. Designed to surface where the source set agrees vs disagrees.

Demand trajectory | "Is Blackwell demand accelerating, stable, or softening?"

14 of 14 responded

Accelerating | 11 of 14 (79%)

Stable | 3 of 14 (21%)

Pricing | "Is ASP rising, flat, or compressing?"

12 of 14 responded

Rising | 8 of 12 (67%)

Flat | 3 of 12 (25%)

Down 1 (8%)

Supply bottleneck | "Is supply the binding constraint in 2H FY27?"

14 of 14 responded

Yes, supply is binding | 12 of 14 (86%)

No | 2 of 14 (14%)

Rubin readiness | "On schedule for 2H FY27 ramp?"

13 of 14 responded

On track | 10 of 13 (77%)

1-quarter slip | 2 (15%)

2-qtr+ slip 1 (8%)

AMD MI400 share threat | "Material to NVDA share?"

14 of 14 responded

Not material | 9 of 14 (64%)

5-8% | 4 (29%)

10%+ 1 (7%)

Hyperscaler capex direction | "FY27 vs FY26?"

12 of 14 responded

Higher | 10 of 12 (83%)

Flat | 2 (17%)

What the matrix tells us: The signal is unusually clean. Demand acceleration, pricing power, supply bottleneck, and Rubin readiness all show 64-86% agreement in the bullish direction. The single biggest disagreement is on the AMD MI400 threat (29% of experts see 5-8% wallet shift). This is the bear case to monitor on the call – listen for unprompted commentary on Hopper-to-Blackwell upgrade cycle and any softness in mid-tier customer mix.

Demand by End-Market

Hyperscalers | Sovereign AI | Enterprise | Inference. Where the incremental dollars are coming from.

MSFT Azure

FY26E capex $112B (vs implied $108B)

+38% YoY

GPU mix rising to ~42% of FY26 capex. OpenAI partnership commitments still flowing through; 2H FY26 build-out concentrated in TX, WI.

Pricing & Gross-Margin Trajectory

ASP path from Hopper through Rubin Ultra. GM walks through the Rubin transition.

Per-system ASP | Hopper → Rubin Ultra

$K per system | ABI estimate aggregating supplier checks

Gross margin | expected path

%, GAAP. Rubin pre-ramp is the Q2 FY27 dip.

Per-system content walk | Rubin vs Blackwell Ultra supplier-derived

Component	Blackwell Ultra $K	Rubin $K	Δ	Driver
GPU compute die (TSM N3P/N2)	$108	$132	+$24	N2 node ASP step + larger die
HBM (6→8 stacks \| HBM3e→HBM4)	$84	$148	+$64	Stack count + HBM4 ASP ~$11/GB
CoWoS-L packaging	$28	$36	+$8	CoWoS-L pricing absorption
Substrate (ABF)	$12	$14	+$2	Larger size, tight supply
NVLink switch + retimers	$18	$24	+$6	NVL576 fabric scaling
Other (PCB, power, cooling, networking)	$32	$42	+$10	Cooling + 1.6T optics
Total BOM cost	$282	$396	+$114	+40%
Per-system ASP (estimate)	$680	$748	+$68 / +10%	Premium for performance/W
Gross margin per system	58.5%	47.1%	-11pp pre-yield	Yield ramp recovers ~9pp
GM at mature yield (12m)	58.5%	56.4%	-2pp	Slight mix headwind

Read: Rubin GM dilutive at launch but recovers within 4 quarters as yields normalize. The system-level GM step-down (-2pp) is more than offset by software/services attach (NVAIE, Cuda Foundry). Aggregate GM remains 74-77% range – the swing factor on the print and guide.

Supply Chain Bottleneck Heat-Map

Per-component supply tightness across 4 forecast quarters. 1 = abundant, 5 = severely constrained.

Component

Q1 FY27 (current)

Q2 FY27

Q3 FY27

Q4 FY27

HBM3e (SK hynix | Micron)

5 | severe

4 | tight

3 | improving

HBM4 (SK hynix lead, Samsung)

5 | severe

4 | tight

CoWoS-L (TSM)

5 | severe

4 | tight

3 | improving

Advanced packaging – generic CoWoS

4 | tight

3 | improving

2 | OK

ABF substrate

4 | tight

3 | improving

800G/1.6T networking switches

3 | improving

2 | OK

Liquid cooling (CDU, manifolds)

4 | tight

3 | improving

2 | OK

PCIe 6.0 retimers (ALAB | CRDO)

3 | improving

2 | OK

EUV wafer capacity (TSM N3/N2)

3 | improving

Critical path: HBM4 and CoWoS-L are the binding constraints. SK hynix HBM4 sample shipments started in late April; mass production confirmed for Q3 FY27. TSM CoWoS-L exit 2026 capacity now 52K wpm vs Mar consensus 45K. The supply story is improving but Q2 FY27 still tight.

Hyperscaler Capex Tracker | Top 8

Disclosed FY26 and implied FY27. ABI estimate combines management commentary with infra build-out evidence.

Buyer	FY25 capex ($B)	FY26E ($B)	YoY	FY27E ABI ($B)	YoY	Implied NVDA $ (ABI)
Microsoft (Azure)	81.2	112.0	+38%	136.0	+21%	$28-34B FY27E
Meta	88.4	118.0	+33%	142.0	+20%	$32-38B FY27E
Amazon (AWS)	76.4	93.0	+22%	114.0	+23%	$22-26B FY27E
Google (GCP)	76.6	95.0	+24%	112.0	+18%	$14-18B FY27E
Oracle (OCI)	21.4	28.0	+31%	36.0	+29%	$8-12B FY27E
CoreWeave / private clouds	11.4	22.0	+93%	34.0	+55%	$14-18B FY27E
xAI	8.6	18.0	+109%	28.0	+56%	$10-14B FY27E
Sovereign + others (Humain \| G42 \| Tata \| Mistral)	4.0	15.0	+275%	26.0	+73%	$14-18B FY27E
Top 8 total	$368B	$501B	+36%	$628B	+25%	$142-178B

Calibration: Implied NVDA FY27 revenue (data center) from top-8 buyers $142-178B; current sell-side consensus FY27 NVDA total revenue $192B. Inference: consensus has limited room for an MI400/TPU/custom-silicon share loss before models break down. Our base case = NVDA FY27 $214B on supply expansion and sovereign upside.

Source Quotes | Top 10 Most-Cited

Anonymous expert quote snippets. Each is from a confidence-weighted ≥7.0 source. Quotes selected for signal, not narrative fit.

"There is no scenario in any planning document I've seen where Blackwell unit demand softens in the next four quarters. Allocation discussions with hyperscalers are about which configuration they get, not whether they get product."

– Former VP Data Center | NVDA (2021-25) | Tegus | May 22, 2026 | weight 9.4

"We've internally moved our 2027 GPU mix from 30% to 41% of capex. The thing nobody is modeling is the inference acceleration – token volume across our top 4 customer use-cases is up 6x year-over-year."

– CTO | Top-5 Hyperscaler | Third Bridge | May 14, 2026 | weight 9.6

"AMD MI400 is a real product. It's not a fake competitor like MI300 was. But the moat is software – to get same time-to-deployment as NVIDIA, you need a 6-month head start, and we don't have that head start."

– Head of AI Infra | Tier-1 Cloud | Tegus | May 12, 2026 | weight 9.2

"HBM4 yields at SK hynix are tracking 64% in May, target was 60. Samsung is closer to 52%. The constraint will be Samsung. SK hynix will outship plan."

– Mgr | SK hynix HBM Engineering | Third Bridge | May 8, 2026 | weight 8.8

"Rubin samples to top-3 hyperscalers happened first week of April. Validation is going faster than Blackwell did – we used the same test harness."

– Former Director Compute Arch | NVDA | AlphaSense | May 18, 2026 | weight 9.2

"Sovereign deals are 2-3x more profitable than hyperscaler at the gross-margin line. Saudi-Humain alone is a $4-6 billion ARR opportunity over 3 years. Nobody on the sell-side is modeling these correctly."

– Former VP Data Center | NVDA | Tegus | May 22, 2026 | weight 9.4

"CoWoS-L exit 2026 capacity is 52K wafers per month, not 45K. The 45K number is from the January roadmap. There was a January-March expansion announcement that the Street missed."

– Senior Director | TSM CoWoS | Tegus | May 16, 2026 | weight 9.0

"What you should listen for on the call is anything they say about Q2 GM. If they guide 74% or below, the market panics. If they say "above 75% on mix", the stock is up 8%. The thing they'll actually say is probably 74.5-75."

– Tier-1 sell-side analyst (debrief, not direct quote) | ABI synthesis | May 24, 2026

"Foxconn AI server volume in May was up 38% sequentially. Three new lines coming online in Mexico for Q3 ramp. The ground truth is brutal – they are building everything they can build."

– Sales VP | Foxconn AI Server Group | AlphaSense | May 6, 2026 | weight 7.6

"PCIe 6.0 retimer attach on Blackwell B200 is 12 per system. On Rubin it goes to 16. We are sold out through Q1 FY28 already. The market is not pricing this enough into ALAB or CRDO."

– Sr. Engineer | Astera Labs | Tegus | May 4, 2026 | weight 7.4

Risk to the Read | What Could Be Wrong

Our internal "bear case checklist" – four ways our pre-print synthesis could be wrong.

1. Sovereign AI orders not yet booked as revenue

Several experts cited Humain, G42, and Mistral as upside, but the contracting cadence is lumpy. Q1 FY27 print may not include the timing benefit. Risk: a "rev recognition timing" question on the call could cap upside.

2. Rubin GM transition more punitive than modeled

Yield ramp on the larger Rubin die may be slower than the Blackwell ramp. We modeled a 9pp recovery in 4 quarters; could be 6pp in 4 quarters. Risk: management guides Q2 FY27 GM to 73% (vs ABI 74.2%). Stock down 4-6%.

3. China sales restrictions extension

H20 (and successor) China access remains discretionary. A reduction or further restriction could materially impact $5-7B of FY27 revenue. None of our experts have direct policy visibility.

4. AMD MI400 share narrative

Even if our "wallet share" reads are right, the sentiment trade after AMD's June 5 product launch could spook NVDA. We expect noise but not fundamentals. Risk = sentiment-driven 5-8% pullback irrespective of print.

Implications for Ancillaries | 10-Name Read

Quick read on companies whose Q-prints will be re-rated by what NVDA says May 27. Direction-of-travel only.

NASDAQ: AVGOLong

Broadcom

Custom silicon (MTIA, TPU) + networking. Hyperscaler capex acceleration is unambiguously good. Q2 print Jun 5 – set up well. Expect +3-5% post NVDA.

NASDAQ: AMDNeutral

Advanced Micro Devices

MI400 launch Jun 5. If NVDA is bullish, AMD trades up on read-across; if NVDA mentions MI400 share, AMD trades up regardless. Asymmetric long into AMD print Jul 30.

TPE: TSMLong

Taiwan Semiconductor

CoWoS-L expansion + N2 ramp + HPC mix. The cleanest indirect NVDA play. Q2 print Jul 17. Long position pre-NVDA print justified.

AMS: ASMLLong

ASML Holding

EUV order book intact; N2 builds in 2H 2026. Less leveraged short-term but the multi-year story is intact. Hold long.

NYSE: ANETLong

Arista Networks

800G adoption accelerating. NVDA print should reinforce the front-end + back-end networking spend trajectory. +2-4% post NVDA expected.

NASDAQ: ALABLong

Astera Labs

PCIe 6.0 retimer attach rate going up with Rubin. Sole expert (Source #9) was explicit. Highest beta to NVDA print in coverage.

NASDAQ: CRDOLong

Credo Technology

AEC (Active Electrical Cables) for back-end fabric. NVDA capex growth = direct beneficiary. +4-7% post NVDA expected.

NASDAQ: MULong

Micron Technology

HBM3e tight; HBM4 sample shipments started May. HBM share could surprise. Less leveraged than SK hynix but still long.

NYSE: CLSLong

Celestica

Hyperscaler server assembly + networking infra. Foxconn read confirms volume. Hold long.

NASDAQ: SMCINeutral

Super Micro Computer

Volume tailwind exists but company-specific overhangs (compliance, GM compression) remain. Don't conflate with the read. Buy on weakness only.

How to Position | Pre-Print Risk/Reward Grid

Three scenarios. Probability-weighted return = +5.4%. Asymmetry is +9% / -6%.

55%

Bull case

+9 to +13% | Beat by 5%+ on revenue | raise Q2 guide above $48B | GM stable 74.5-75%. Sovereign AI explicitly called out.

28%

Base case

+2 to +5% | Beat in-line with whisper | guide $47-48B | GM 74-74.5%. Normal NVDA print reaction.

17%

Bear case

-4 to -8% | Light beat | soft guide | GM dip to 73% | China commentary cautious. Profit-take into print.

ABI recommended position: Hold core NVDA long. Add modestly into the print (~5% of position) to capture upside scenario. Tactical hedge via NVDA Jun-13 strangle if existing exposure is concentrated. Ancillary book: add to ALAB and CRDO into the print; trim SMCI; long-hold AVGO, TSM, MU. For AMD: long the post-NVDA gap if NVDA beats (negative-beta misread opportunity).