SemiconductorsBRIEF

SpaceX Deepens NVIDIA Partnership as GPU Allocation Fears Squeeze Emerging Cloud Providers

N
NathanTechnology Editor · Technical Lead
Published · Updated
According to TechNews, SpaceX plans to build its AI data centers entirely on NVIDIA (輝達) chips and targets AI compute capacity above 2GW by end-2026, rising to near 10GW by end-2027. Bernstein analyst Madison Rezaei said the market fears NVIDIA may prioritize GPU allocation to SpaceX ahead of the cloud providers it backs — a concern surfacing just as CoreWeave and Nebius prepare to report earnings on August 11 and 12.

SpaceX's Expanded NVIDIA Partnership and Its Compute Targets

Speaking on an earnings call on August 4, Elon Musk said SpaceX's planned AI data centers will run entirely on NVIDIA (輝達) chips, according to TechNews. TechNews further reported that SpaceX expects its AI compute capacity to exceed 2GW by the end of 2026, expanding to near 10GW by the end of 2027.

Beyond terrestrial data centers, SpaceX also said it is working with NVIDIA to design the compute payload for the Starmind AI1 satellite, aiming to deliver datacenter-class computing power in space, per TechNews.

Analysts Warn of GPU Allocation Risk as CoreWeave and Nebius Head Into Earnings

Bernstein analyst Madison Rezaei said the market is concerned NVIDIA could prioritize GPU allocation to SpaceX — potentially even ahead of the cloud providers NVIDIA itself has backed, TechNews reported.

That concern lands just as two of those NVIDIA-backed cloud providers face investor scrutiny. CoreWeave is scheduled to report earnings on August 11, and Nebius on August 12, per TechNews. Citing FactSet, TechNews reported that Wall Street expects CoreWeave's per-share loss to widen substantially from a year earlier, even as revenue is projected to more than double. Nebius's per-share loss is also expected to widen from a year earlier, with revenue forecast to surge 442% year-over-year to $570 million, TechNews reported.

Nebius has additionally become a short-selling target for well-known investor Michael Burry, and its data center expansion plans face opposition from local U.S. communities, according to TechNews.

NVIDIA's Separate Mega-Deal With Meta: Scale, Concentration, and the Grace CPU Pitch

In a development reported separately by Cnyes, NVIDIA (輝達) and Meta announced on February 17 an expanded, multi-year, cross-generation partnership under which NVIDIA will supply Meta with millions of Blackwell and Rubin GPUs, along with CPUs and networking products. NVIDIA did not disclose the deal's value, but Cnyes reported that analysts estimate it could be worth as much as $50 billion.

Cnyes noted that Meta is widely regarded as one of NVIDIA's top four customers, and that in its most recent quarter, NVIDIA's top four customers together accounted for 61% of revenue — underscoring how concentrated NVIDIA's largest deals already are.

On the technology side, Ian Buck, NVIDIA's general manager of hyperscale and high-performance computing, told Cnyes that the Grace CPU has already shown energy efficiency gains in common workloads — for example, cutting power consumption by half when running databases — with the next-generation Vera chip expected to deliver further improvement.

Stock Reactions and the Search for Alternatives

According to Cnyes, Meta's stock has fallen about 3.3% year-to-date, while Microsoft (微軟)'s has dropped more than 17% over the same period.

Cnyes also reported that alongside NVIDIA's announcement of the expanded Meta partnership, reports emerged that Meta is separately pursuing its own AI chip program and discussing with Google the possible use of Google's Tensor Processing Units (TPU) for AI computation.

Key Figures at a Glance

MetricValueSource
SpaceX AI compute capacity (end of 2026)>2GWTechNews
SpaceX AI compute capacity (end of 2027)~10GWTechNews
CoreWeave earnings dateAug. 11TechNews
Nebius earnings dateAug. 12TechNews
Nebius revenue growth forecast (YoY)+442% to $570MTechNews
NVIDIA–Meta deal estimated value$50 billionCnyes
NVIDIA top-4 customer revenue concentration61%Cnyes
Grace CPU database workload power savings50%Cnyes
Meta stock, year-to-date-3.3%Cnyes
Microsoft stock, year-to-date-17%Cnyes

What This Means

Taken together, the two reports describe NVIDIA's supply chain pulling in two directions at once. On one side, SpaceX's own compute ramp — from over 2GW by end-2026 to near 10GW by end-2027 — comes with Musk's pledge to run entirely on NVIDIA chips, which Bernstein's Rezaei says is fueling market worry that NVIDIA-backed cloud providers could be deprioritized. That worry precedes earnings from CoreWeave and Nebius, both of which are already showing widening per-share losses even as revenue grows sharply (Nebius +442%), and Nebius separately faces a short-seller and local community pushback. On the other side, NVIDIA's top four customers — including Meta, per Cnyes — already account for 61% of its revenue, and the up-to-$50-billion Meta deal only deepens that concentration. Yet Meta's own stock has underperformed (-3.3% YTD) and it is reportedly exploring Google's TPUs as an alternative — suggesting that even NVIDIA's largest, most concentrated customer relationships are not insulating either side from supply and diversification pressure.

📊 Evidence

FAQ

When are CoreWeave and Nebius scheduled to report earnings?

According to TechNews, CoreWeave is scheduled to report on August 11 and Nebius on August 12.

How large is NVIDIA's expanded partnership with Meta estimated to be?

NVIDIA has not disclosed the deal's value, but Cnyes reported that analysts estimate it could be worth as much as $50 billion.

📎 Sources

  1. technews.tw
  2. news.cnyes.com
N
NathanTechnology Editor · Technical Lead

Related

BRIEF

Anthropic Makes Claude Code's Auto Mode the Default Starting August 14

According to TechCrunch and ITHome, Anthropic will set auto mode as the default permission setting for Claude Code Pro, Max, and Team accounts from August 14, 2026, while Enterprise, API, and cloud-platform users must opt in manually. Anthropic's internal study of 1,053 paid testers found auto mode blocked 89% of harmful actions versus 13.6% for manual human review, and separately reported that users had been approving 97% of permission prompts before this change.

Nathan ·
BRIEF

Meta Returns to Open Source With Muse Glimmer, a 30B Apache 2.0 Agent Model

According to VentureBeat, Meta released Muse Glimmer on August 10, 2026 — a 30-billion-parameter, Apache 2.0-licensed open-weight model built to run AI agents on consumer hardware. TechCrunch reports it is an open version of Meta's closed Muse Spark model, framed by CEO Mark Zuckerberg as evidence of his "personal superintelligence" vision.

Nathan ·
BRIEF

NVIDIA Partners With Six Wall Street Firms on $500 Billion AI Financing Platform

NVIDIA (輝達) has signed memorandums of understanding with six Wall Street firms — Apollo, BlackRock (貝萊德), Blackstone, Brookfield, Goldman Sachs (高盛), and KKR — to launch a computing financing platform targeting over $500 billion in third-party funding for AI infrastructure, according to CNA (中央社) and Liberty Times (自由時報) reports. The move follows Moody's warning that unprecedented capital spending is squeezing tech firms' free cash flow, per TechNews (科技新報). NVIDIA has not disclosed financing terms, individual commitment amounts, or a deployment timeline for the $500 billion target.

林紀旭 James Lin ·