Nvidia said to issue memo to analysts on $4.5t valuation

Nvidia has issued a memo to stock analysts addressing recent criticism of its US$4.5 trillion valuation, which has dropped from a peak of US$5 trillion.

The company responded to claims from investor Michael Burry, and also addressed a separate Substack essay that used AI analysis of Nvidia’s public financial disclosures to allege inventories were building up and some customers could not pay.

Nvidia, which did not comment to Reuters, said in the memo that its public disclosures do not support comparisons to accounting scandals like WorldCom, Lucent, or Enron, but acknowledged its latest Blackwell chips have lower gross margins and higher warranty costs.

The memo was released by Bernstein a day after Nvidia shares fell following a report that Meta was in talks with Google to use Google’s AI chips, which compete with Nvidia’s.

Nvidia also addressed the report on X, congratulating Google’s success and saying its chips remain a generation ahead, which led some users to question why the company was defending itself on social media.

🔗 Source: Reuters

🧠 Food for thought

Implications, context, and why it matters.

Nvidia says inventory and payment concerns lack public-data support

Nvidia posted a record Q3 fiscal 2026 revenue of $57.0 billion, up 62% year over year 1, while data center revenue hit $26.3 billion with 154% growth in Q2 fiscal 2025 2.
Operating cash flow reached $14.5 billion in Q2 fiscal 2025 2, with $37.0 billion returned to shareholders via buybacks and dividends in the first nine months of fiscal 2026 1.
In a memo, Nvidia said Blackwell chips have lower gross margins and higher warranty costs, while Q3 fiscal 2026 non-GAAP (non–Generally Accepted Accounting Principles) gross margin landed at 73.6% within a mid‑70s range 1.
ceo Jensen Huang said Blackwell sales are off the charts, with cloud GPUs sold out 1.

Cross-hardware AI infrastructure creates integration opportunities for infrastructure vendors and cloud integrators

Meta’s talks with Google for AI chips suggest big firms will run heterogeneous accelerators, which will need abstraction tools to manage mixed hardware.
OpenXLA is an open-source machine learning compiler 3. Backers include Alibaba/AWS/AMD/Google/Intel/Meta/Nvidia 3. It lets teams optimize models across varied hardware, with Alibaba tests finding 72–88% speed gains on Nvidia GPUs 3.
Open Neural Network Exchange (ONNX) Runtime delivers production AI inference across platforms 4. It powers Microsoft products including Windows and Office 4. It also supports Azure Cognitive Services and Bing 4.
Infrastructure vendors and cloud integrators that build cross-hardware compilation skills can secure enterprise deals from teams adopting multi-vendor AI, while lowering lock-in risk.