The Information Machine

NVIDIA Launches Vera CPU and Vera Rubin NVL72 at COMPUTEX / GTC Taipei · history

Version 12

2026-06-03 18:39 UTC · 250 items

What

NVIDIA's Vera Rubin NVL72 has cleared two independent rack validation milestones: Dell delivered the world's first fully validated NVL72 to CoreWeave on May 31 [1][3], and Jensen Huang announced at COMPUTEX 2026 (June 1) that Microsoft completed bring-up of its first Rubin VR200 NVL72 rack via Foxconn as ODM [7]. Jensen also disclosed that wafer-level Rubin production has started but rack-level mass production has not yet begun [7]. NVIDIA's COMPUTEX announcements extended to software: official Marvell and NVIDIA press releases frame the $2B NVLink Fusion investment as enabling 'semi-custom AI infrastructure' for third-party silicon [9], and NVIDIA's NemoClaw blueprint for industrial AI agents has been adopted by Cadence, Siemens, Dassault Systèmes, and Synopsys [11]. SemiAnalysis rated the COMPUTEX keynote 'F tier,' finding no new AI datacenter products [12].

Why it matters

Two separate ODM chains completing NVL72 bring-up within 48 hours confirms the Rubin validation ecosystem is operational across multiple operators simultaneously. The gap between wafer-level and rack-level production start [7] remains the concrete signal for when large announced commitment volumes — from Microsoft, CoreWeave, and others — can actually ship. HBM4 shortage projected to persist until 2028 [16][18] and a 600kW per-rack power requirement mandating greenfield construction [19] are binding ceilings on that pace.

Open questions

  • Rack-level mass production of Rubin VR200 NVL72 had not started as of the COMPUTEX keynote [7] — when does it begin, and does the wafer-to-rack lag create a delivery gap for the large commitment volumes announced by Microsoft, CoreWeave, and others?

  • Both CoreWeave and Microsoft have achieved single-rack bring-up (L11), but full-cluster L12 validation with scale-out networking has not been demonstrated by either [3][6][7] — when does the first L12 milestone occur, and at what cluster scale?

  • NVIDIA's official NVLink Fusion press release describes it as enabling 'semi-custom AI infrastructure' [9], while Jensen's COMPUTEX framing emphasized platform openness — does NVLink Fusion represent genuine architectural openness to alternative GPU suppliers or a mechanism to retain interconnect control over competing silicon?

  • Micron's IR press release asserts high-volume HBM4 production for Vera Rubin [24] while multiple independent sources conclude NVIDIA designated only Samsung and SK Hynix as suppliers [25][26][27] — has Micron secured an actual production allocation or is the claim aspirational?

Narrative

NVIDIA's Vera Rubin NVL72 crossed its first two concrete deployment milestones in rapid succession. On May 31, Dell delivered the world's first fully validated Vera Rubin NVL72 rack to CoreWeave, with L11 diagnostics confirming the rack's internal NVLink/IMEX scale-up domain operational [1][2][3][4][5]. In NVIDIA's three-stage validation hierarchy, L10 certifies single-server firmware, L11 certifies a single rack's scale-up domain, and L12 certifies a full compute cluster with scale-out networking [6]. The following day, at COMPUTEX 2026, Jensen Huang announced that Microsoft has completed bring-up of its first Rubin VR200 NVL72 rack via Foxconn as ODM [7] — a second validation chain becoming operational within 24 hours of the first. Jensen also disclosed that wafer-level mass production of Rubin has started while rack-level mass production has not yet begun [7], a distinction that matters for assessing when the large announced deployment volumes can actually ship.

The COMPUTEX and surrounding announcements extended beyond hardware validation to ecosystem and software. Official press releases from Marvell and NVIDIA frame the $2B NVLink Fusion investment as enabling 'semi-custom AI infrastructure' — allowing hyperscalers to integrate custom silicon into NVIDIA's NVLink interconnect fabric [8][9]. Jensen publicly predicted Marvell would become the next trillion-dollar company [10]. Separately, NVIDIA's NemoClaw blueprint for industrial AI agents — providing a secure runtime, model router, and NeMo customization libraries — has been adopted by Cadence, Dassault Systèmes, Siemens, and Synopsys to compress weeks-long engineering tasks such as RTL verification and thermal simulation into hours [11]. SemiAnalysis graded the COMPUTEX keynote 'F tier,' stating Jensen announced nothing new on the AI datacenter side and that the headline Windows-on-NVIDIA-ARM announcement faces structural conditions unlike Apple's x86-to-M1 transition [12].

The performance picture for Vera Rubin has two distinct layers. Phoronix benchmark results show a 1.5x overall advantage over 128-core x86 and a 1.6x geometric mean improvement over NVIDIA's Grace CPU [13][14]. SemiAnalysis adds the critical nuance: Rubin GPU FP4/FP8 FLOPs scale approximately 3.5x over GB200 while FP16 gains are only ~1.6x and HBM capacity remains flat [15]. The efficiency claim is workload-specific, concentrated in low-precision inference rather than distributed across training or high-precision workloads.

Two hardware constraints establish binding ceilings on deployment pace. HBM4 supply is dominated by SK Hynix (~70% of NVIDIA's orders) and Samsung (2026 allocation sold out), with shortage projected until 2028 and rack prices at $8.8M [16][17][18]. The NVL72's 600kW per-rack power requirement is incompatible with most existing data center infrastructure, requiring greenfield construction [19][20]. NVIDIA holds approximately $674M in equity in Nscale [21] — making the large deployment announcements co-publicized by NVIDIA and Nscale non-arm's-length transactions [22]. Taiwan's manufacturing ecosystem provides the assembly backbone, with more than 500 NVIDIA partners assembling over one million MGX rack components across 25 factory sites [23].

Timeline

  • 2026-01-05: NVIDIA debuts Rubin chip at CES: 336 billion transistors, 50 petaflops AI performance. [45]
  • 2026-01: Jensen Huang announces at CES 2026 that Vera Rubin NVL72 is in full production. [46][47]
  • 2026-02: SK Hynix begins HBM4 mass production shipments to NVIDIA, holding approximately 70% of NVIDIA's HBM4 orders. [48][16]
  • 2026-03-17: Nscale acquires 8GW Monarch Compute Campus in West Virginia; Microsoft signs 1.35GW LOI co-announced with NVIDIA and Caterpillar. [49][50][22][51]
  • 2026-05: Samsung sells out entire 2026 HBM4 supply; rack prices reach $8.8M; HBM4 shortage projected to persist until 2028. [40][18][41][17]
  • 2026-05: Multiple analyses document Vera Rubin NVL72's 600kW per-rack power requirement as incompatible with existing data centers, requiring greenfield construction. [19][20][42]
  • 2026-05: NVIDIA equity stake in Nscale confirmed at approximately $674M; Microsoft's Rubin GPU deployment via Nscale revised upward to 130,000 units. [44][21][32]
  • 2026-05-18: First Vera CPUs hand-delivered to OpenAI, Anthropic, and other leading AI labs. [52][53][54]
  • 2026-05-18: Jensen Huang keynotes Dell Technologies World: projects $3–4T AI infrastructure buildout by 2030 and flags memory supply as primary bottleneck. [28][55][56]
  • 2026-05-21: NVIDIA reports Q1 2026 earnings: $81.6B revenue, up 85% year-over-year. [29][57]
  • 2026-05-21: NVIDIA GTC Taipei: Vera Rubin NVL72 wins Computex Best Choice Golden Award; Meta, Google Cloud, and Microsoft formalize partnerships; NVIDIA announces $2B NVLink Fusion investment in Marvell. [58][59][60][61][31][10]
  • 2026-05-26: Phoronix benchmarks of Vera CPU published: 1.5x overall x86 advantage, 1.6x geometric mean over Grace CPU, 90% peak bandwidth utilization. [13][14]
  • 2026-05-31: Dell delivers world's first fully validated Vera Rubin NVL72 rack to CoreWeave; L11 diagnostics confirm scale-up domain operational. [1][2][3][6][4][5][33]
  • 2026-06-01: Jensen Huang at COMPUTEX announces Microsoft has completed bring-up of its first Rubin VR200 NVL72 rack via Foxconn as ODM; wafer-level production started, rack-level mass production not yet begun. [7]
  • 2026-06-01: SemiAnalysis rates Jensen's COMPUTEX keynote 'F tier': no new AI datacenter products announced; Windows-on-NVIDIA-ARM transition unlikely to succeed. [12]
  • 2026-06-01: Official NVIDIA and Marvell press releases frame NVLink Fusion as enabling 'semi-custom AI infrastructure' for third-party silicon integration. [8][9]
  • 2026-06-02: NVIDIA NemoClaw industrial AI agent blueprint announced with adoptions from Cadence, Siemens, Dassault Systèmes, and Synopsys, compressing RTL verification and simulation from weeks to hours. [11]

Perspectives

NVIDIA / Jensen Huang

Maximally bullish: Q1 2026 earnings ($81.6B, +85% YoY) validate parabolic AI demand; COMPUTEX framing positions this as 'the greatest era in history to build software'; $2B Marvell NVLink Fusion investment and NemoClaw industrial agent blueprint extend the platform to third-party silicon and enterprise software.

Evolution: Expanded: official NVLink Fusion press release framing as 'semi-custom AI infrastructure' [9] and NemoClaw software announcement [11] broaden NVIDIA's COMPUTEX narrative to ecosystem and software beyond hardware alone.

[28][29][30][7][10][9][11]

SemiAnalysis (technical analysis and keynote critique)

Rubin FP4/FP8 FLOPs scale ~3.5x over GB200 while FP16 gains are only ~1.6x and HBM capacity is flat — headline efficiency is workload-specific. COMPUTEX keynote rated F tier for delivering no new AI datacenter products and an ARM transition framing unlikely to replicate Apple's success.

Evolution: Consistent; remains the primary independent critical voice against NVIDIA's deployment momentum narrative.

[15][12]

Microsoft / Foxconn / Nscale

Anchor customer for Vera Rubin NVL72 via Nscale (130,000 GPUs at Start Campus Portugal, 1.35GW LOI for West Virginia) and first hyperscaler to complete Rubin VR200 NVL72 bring-up, with Foxconn as ODM partner.

Evolution: Consistent; Microsoft and Foxconn are confirmed as the second actor pair to complete NVL72 bring-up, within 24 hours of Dell and CoreWeave.

[31][22][32][7]

Dell / CoreWeave

First-mover operators: Dell delivered the world's first fully validated Vera Rubin NVL72 rack to CoreWeave, clearing L11 diagnostics; CoreWeave is confirmed among first cloud providers adopting Vera Rubin.

Evolution: Consistent; CoreWeave's milestone is now confirmed by multiple independent outlets [4][5][33] without new technical detail.

[1][2][3][34][4][5][33]

Marvell / NVLink Fusion partners

Official Marvell press release frames the NVLink Fusion partnership as enabling customers to build 'semi-custom AI infrastructure' by integrating Marvell custom silicon with NVIDIA's NVLink interconnect fabric.

Evolution: New official voice: prior synthesis relied on Jensen's COMPUTEX statements; Marvell's own press release [8][9] now provides direct sourcing for the partnership framing.

[8][35][36][9][37][38]

Micron

Officially asserts high-volume HBM4 production specifically designed for NVIDIA Vera Rubin, directly contradicting industry reports concluding NVIDIA designated only Samsung and SK Hynix as HBM4 suppliers.

Evolution: Consistent; the contradiction with multiple independent sources remains unresolved.

[24][39][25][26][27]

Memory and supply chain analysts

HBM4 shortage is the binding structural constraint: SK Hynix holds ~70% of NVIDIA's orders, Samsung has sold out its 2026 supply, rack prices are at $8.8M, and shortage is projected to persist until 2028.

Evolution: Consistent; rack-level mass production not yet started adds a further constraint at the moment large-scale commitments are outstanding.

[16][17][40][18][41][7]

Data center infrastructure analysts

Vera Rubin NVL72's 600kW per-rack power requirement is a fundamental incompatibility with existing data center infrastructure — not a retrofit problem but a greenfield requirement, establishing a second binding structural bottleneck alongside HBM4 supply.

Evolution: Consistent.

[19][20][42]

Tensions

  • Micron's official IR press release states high-volume HBM4 production for NVIDIA Vera Rubin [24], while TechPowerUp and multiple independent analyses conclude NVIDIA designated only Samsung and SK Hynix as HBM4 suppliers [25][26][27] — two claims that are mutually incompatible unless they refer to different allocation tiers. [24][25][26][27]
  • NVIDIA markets Vera Rubin on a 10x cost-per-token reduction, but SemiAnalysis documents FP4/FP8 gains of ~3.5x over GB200 while FP16 gains are only ~1.6x and HBM capacity is flat [15] — the efficiency claim is workload-specific, not uniform across training or high-precision inference. [15][18][19]
  • NVIDIA holds approximately $674M in Nscale equity [21] while publicly describing Nscale only as a commercial partner — making large deployment announcements co-publicized by both companies non-arm's-length commercial transactions [22][43]. [43][44][21][22]
  • Official NVLink Fusion press materials frame the partnership as enabling 'semi-custom AI infrastructure' for third-party silicon [9], while the structural terms — third-party chips must adopt NVIDIA's NVLink interconnect — leave unresolved whether this is genuine platform openness or a mechanism for NVIDIA to retain ecosystem control. [8][9][10]
  • Wafer-level mass production of Rubin has started but rack-level mass production has not yet begun [7] — creating a gap between NVIDIA's production narrative and actual rack availability at a moment when large-scale deployment commitments from Microsoft and others are outstanding. [7]
  • SemiAnalysis rated Jensen's COMPUTEX keynote 'F tier' for announcing no new AI datacenter products [12], while NVIDIA's promotional coverage treats the same event as a major milestone featuring Microsoft's NVL72 bring-up, NVLink Fusion official launch, and NemoClaw ecosystem expansion [34][7][9][11]. [12][34][7][9][11]

Sources

  1. [1] BREAKING NEWS: COREWEAVE & DELL IS THE FIRST CLOUD TO ANNOUNCE THAT THEY HAVE RUBIN VR200 NVL72 WITH FULLY PASSING L… — SemiAnalysis Twitter (2026-05-31)
  2. [2] Dell just made history this weekend and it is the culmination of an execution streak that no other company in enterprise… — Milk Road AI Twitter (2026-05-31)
  3. [3] Notably, passing L11 diags means that this rack is up and running, including the IMEX channels on the NVL72 scale-up dom… — SemiAnalysis Twitter (2026-05-31)
  4. [4] CoreWeave Completes Industry-First Bring-Up And Validation Of NVIDIA Vera Rubin NVL72 — reactive:nvidia-vera-computex-launch
  5. [5] HPCwire - Since 1987 – Covering the Fastest Computers in the World and the People Who Run Them — reactive:nvidia-vera-computex-launch
  6. [6] At L10 your Firmware/BIOS and OS works on a single server, at L11 a single rack or scale-up domain works, and then at L1… — SemiAnalysis Twitter (2026-05-31)
  7. [7] BREAKING NEWS: JENSEN JUST ANNOUNCED MICROSOFT HAS FINISHED BRING UP ON THEIR FIRST RUBIN VR200 NVL72 RACK with their OD… — SemiAnalysis Twitter (2026-06-01)
  8. [8] Marvell and NVIDIA to Provide Custom Solutions for Advanced AI Infrastructure — reactive:nvidia-vera-computex-launch
  9. [9] NVIDIA Unveils NVLink Fusion for Industry to Build Semi-Custom AI Infrastructure With NVIDIA Partner Ecosystem | NVIDIA Newsroom — reactive:nvidia-vera-computex-launch
  10. [10] The CEO of NVIDIA, looked at Matt Murphy and said "The next trillion dollar company, ladies and gentlemen." (Save this). — Milk Road AI Twitter (2026-06-02)
  11. [11] Industrial Software Leaders Build Secure, Autonomous AI Engineers With NVIDIA NemoClaw — NVIDIA Blog (2026-06-02)
  12. [12] F TIER KEYNOTEMAX: Jensen ComputeX presentation was one of the worst keynotes he has done. He announced nothing new on t… — SemiAnalysis Twitter (2026-06-01)
  13. [13] NVIDIA Vera CPU Is ‘Packing a Heavy-Hitting Punch’ Against Competition — NVIDIA Blog (2026-05-26)
  14. [14] NVIDIA published a report on Vera CPU benchmarks, done by Phoronix. — Rohan Paul Twitter (2026-05-28)
  15. [15] for more details on Nvidia's VR NVL72 Oberon and future roadmap, check out our article from February: — SemiAnalysis Twitter (2026-05-31)
  16. [16] SK Hynix Secures 70% of Nvidia's HBM4 Orders - Semicon — reactive:nvidia-vera-computex-launch
  17. [17] SK Hynix Surges 15% to New High: HBM Shortage Until 2028, How Much Longer Can AI Memory King Rise? — reactive:nvidia-vera-computex-launch
  18. [18] Nvidia's memory costs soar 485%, latest AI systems now cost $7.8 ... — reactive:nvidia-vera-computex-launch
  19. [19] The Data Center Isn't Ready. NVIDIA's Vera Rubin platform ships in… — reactive:nvidia-vera-computex-launch
  20. [20] NVIDIA Vera Rubin: 600kW Racks by 2027 | Introl Blog — reactive:nvidia-vera-computex-launch
  21. [21] UK AI Infrastructure Startup Nscale Receives $674 Million (£500 ... — reactive:nvidia-vera-computex-launch
  22. [22] Nscale acquires 8GW Monarch Compute Campus, Microsoft signs on for 1.35GW of compute - DCD — reactive:nvidia-vera-computex-launch
  23. [23] Taiwan’s Industry Titans Turbocharge World’s AI Infrastructure Buildout With NVIDIA — NVIDIA Blog (2026-06-01)
  24. [24] Micron in High-Volume Production of HBM4 Designed for NVIDIA ... — reactive:nvidia-vera-computex-launch
  25. [25] Micron Is Locked Out of HBM4 in NVIDIA's Vera Rubin Systems — reactive:nvidia-vera-computex-launch
  26. [26] NVIDIA to Use SK hynix and Samsung HBM4 for "Vera Rubin" Without Micron | TechPowerUp — reactive:nvidia-vera-computex-launch
  27. [27] Why Nvidia Snubbed Micron For Samsung, SK Hynix - Dailymotion — reactive:hbm-memory-supply-squeeze
  28. [28] NVIDIA CEO Jensen Huang at Dell Technologies World: ‘Demand Is Going Parabolic, Utterly Parabolic’ — NVIDIA Blog (2026-05-18)
  29. [29] NVIDIA just dropped $81.6B in Q1 revenue up 85% YoY 🤯 — reactive:nvidia-vera-computex-launch (2026-05-21)
  30. [30] Jensen Huang just said this is the greatest era in history to build software. AI agents will not kill software. They wil… — Rohan Paul Twitter (2026-06-01)
  31. [31] Microsoft's strategic AI datacenter planning enables seamless, large ... — reactive:nvidia-vera-computex-launch
  32. [32] 130,000 Rubin GPUs Are Being Deployed at Nscale For Microsoft, Further Showing Massive Interest In NVIDIA's Next-Gen AI Chips — reactive:nvidia-vera-computex-launch
  33. [33] CoreWeave Completes Industry-First Bring-Up and Validation of NVIDIA Vera Rubin NVL72 - Las Vegas Sun News — reactive:nvidia-vera-computex-launch
  34. [34] NVIDIA AI Cloud Ecosystem Expands Worldwide to Meet Global AI Compute Demand — NVIDIA Blog (2026-06-01)
  35. [35] Marvell and NVIDIA partner on NVLink Fusion | Marvell Technology posted on the topic | LinkedIn — reactive:nvidia-vera-computex-launch
  36. [36] Marvell today announced it is teaming with NVIDIA to offer NVLink ... — reactive:nvidia-vera-computex-launch
  37. [37] NVIDIA Unveils NVLink Fusion for Industry to Build Semi-Custom AI ... — reactive:nvidia-vera-computex-launch
  38. [38] NVIDIA invests $ 2 billion in Marvell Technology in silicon photonics ... — reactive:nvidia-vera-computex-launch
  39. [39] Micron Singapore - Facebook — reactive:nvidia-vera-computex-launch
  40. [40] Samsung sells out of 2026 HBM4 supply as memory resurgence ... — reactive:aws-garman-a100-demand
  41. [41] Price of Nvidia's Vera Rubin NVL72 racks skyrockets to as much as $8.8 million apiece, but server makers' margins will be tight — Nvidia is moving closer to shipping entire full-scale systems — reactive:nvidia-vera-computex-launch
  42. [42] Nvidia's Vera Rubin GPU: Redesigning Data Centres for 600kW Racks — reactive:nvidia-vera-computex-launch
  43. [43] Nvidia-Backed Nscale Plans Huge Data Center Cluster in West ... — reactive:nvidia-vera-computex-launch
  44. [44] Nvidia-backed UK AI firm Nscale raises $1.1 billion funding round — reactive:nvidia-vera-computex-launch
  45. [45] Nvidia debuts Rubin chip with 336B transistors and 50 petaflops of AI performance - SiliconANGLE — reactive:nvidia-vera-computex-launch
  46. [46] Nvidia CEO confirms Vera Rubin NVL72 is now in production — reactive:nvidia-vera-computex-launch
  47. [47] NVIDIA Vera Rubin AI Platform Hits Full Production CES 2026 ... — reactive:nvidia-vera-computex-launch
  48. [48] SK Hynix set to ship HBM4 for Nvidia's Vera Rubin this month — reactive:nvidia-vera-computex-launch
  49. [49] Nscale and Microsoft Announce Collaboration with NVIDIA and Caterpillar to Deliver 1.35GW of NVIDIA Vera Rubin NVL72 GPUs at Flagship AI Factory Campus in West Virginia — reactive:nvidia-vera-computex-launch
  50. [50] Nscale and Microsoft Announce Collaboration with NVIDIA and Caterpillar to Deliver 1.35GW of NVIDIA Vera Rubin NVL72 GPUs at Flagship AI Factory Campus in West Virginia — reactive:nvidia-vera-computex-launch
  51. [51] Nscale acquisition includes plan to build AI facility in Mason County — reactive:nvidia-vera-computex-launch
  52. [52] Vera Arrives: NVIDIA’s First CPU Built for Agents Lands at Top AI Labs — NVIDIA Blog (2026-05-18)
  53. [53] NVIDIA hand-delivers first 1.2 TB/s Vera CPUs to OpenAI, Anthropic ... — reactive:nvidia-vera-computex-launch
  54. [54] Nvidia unveils details of new 88-core Vera CPUs positioned to compete with AMD and Intel – new Vera CPU rack features 256 liquid-cooled chips that deliver up to a 6X gain in CPU throughput | Tom's Hardware — reactive:nvidia-vera-computex-launch
  55. [55] NVIDIA AI - Jensen Huang Says “Buy Dell” - LinkedIn — reactive:nvidia-vera-computex-launch
  56. [56] Jensen Huang today: Memory demand >> supply chain capacity. “Supply chain needs to be ready.” AI memory supercycle... — reactive:nvidia-vera-computex-launch (2026-05-18)
  57. [57] "Demand has gone parabolic. The reason is simple: Agentic AI has arrived." — reactive:nvidia-vera-computex-launch (2026-05-21)
  58. [58] NVIDIA GTC Taipei at COMPUTEX: Live Updates on What’s Next in AI — NVIDIA Blog (2026-05-21)
  59. [59] NVIDIA Vera Rubin NVL72 wins Computex 2026 awards for AI ... — reactive:nvidia-vera-computex-launch
  60. [60] Meta Builds AI Infrastructure With NVIDIA — reactive:nvidia-vera-computex-launch
  61. [61] NVIDIA GTC 2026: Google Cloud Deepens Partnership for AI ... — reactive:nvidia-vera-computex-launch