The latest release from the Marvell Chipematic Universe is out:

JPMorgan analyst Harlan Sur hosted a meeting with Marvell Technology President and COO Chris Koopmans and Senior VP of Investor Relations Ashish Saran on Monday amid reports that the chip company was poised to lose business from its two biggest hyperscaler custom chip clients: Amazon and Microsoft.

Benchmark downgraded the company on Monday, citing a loss of Trainium3 and 4 business, while The Information said on Friday the latter was planning on shifting its business to Broadcom. Shares tumbled 7% on Monday, erasing all of its post-earnings bounce, and are down again on Tuesday.

The message communicated to Sur from Marvell is, in short, one of Vince Vaughn’s quotable lines in “Wedding Crashers”: “Erroneous! Erroneous on both counts!”

“At our meeting yesterday, the Marvell team reiterated securing purchase orders for all of CY26 for the next-gen Trainium 3 XPU ASIC program at AWS and that the Microsoft 3 nanometer Maia AI XPU ASIC program remains on track to ramp back-half of calendar year 2026 and into calendar year 2027,” Sur wrote in a note to clients on Tuesday. “Moreover, the team reiterated that they are already working on next-gen 2 nanometer XPU programs for both customers.”

The analyst maintained a $92 price target and “overweight” rating on the shares.

Sur added that Marvell’s management “remains perplexed/frustrated at all of the ‘noise’ in the market.”

This whole thing is starting to have the feel of a three- to four-episode subplot arc from HBO’s “Billions.”

Accenture is rising after the consulting giant announced a multiyear partnership with Anthropic to become “a premier AI partner for coding with Claude Code.” This includes a joint offering for AI-enabled software development with a focus on regulated industries including finance, healthcare, life sciences, and the public sector.

It comes on the heels of Accenture’s partnership with OpenAI earlier this month to utilize ChatGPT Enterprise in its consulting work. It’s also recently invested in AI-powered customer research platform WEVO and expanded its collaboration with cloud-based data company Snowflake to better utilize data using AI tools.

As Sherwood News’ Hyunsoo Rim recently flagged, the consulting business has hit an AI-shaped wall, with employment in the industry peaking shortly after the launch of ChatGPT.

Charitably, Accenture’s management is eagerly embracing how the consulting business may be radically altered in a world where corporate AI adoption is ubiquitous, and reacting accordingly. Uncharitably, it’s the best “training your replacements” company out there.

The emphasis on regulated industries as potential customers for this partnership is noteworthy. Jordi Visser of 22V Research recently discussed at length how GenAI tools that enable “vibe coding” reduce barriers to entry for software development, prompting a need to focus on industries where quality and safety are paramount.

“Where software is mostly a polished UI on CRUD, vibe coding is existential,” he wrote. “Where software is inseparable from life, safety, or regulated liability, AI deepens the moat.”

CoreWeave’s stock and bonds are well off their highs of the year, but investors seem to have a healthy appetite for its convertible notes.

The neocloud company said Tuesday morning that its $2 billion convertible note offering announced on Monday has been upsized to $2.25 billion. The boosted deal has the potential to be upsized further by another $337.5 million.

These notes, which mature on December 1, 2031, have a coupon of 1.75% and a conversion price of about $107.80 a share.

That’s around 25% above where it ended on Monday and 41% below its peak closing price, but also a level where the stock has traded within the past month.

Shares sank as much as 9.2% at their lows yesterday, but recovered to finish down 2.4%.

The company plans to spend about 13% of the proceeds (or a little less than $300 million) to enter into capped call transactions that limit any potential dilutive impact of this offering, with the remainder to be used for general corporate purposes.

CoreWeave arguably never took advantage of its very lofty equity price from June through October to raise money via equity, though one could argue that its failed bid to acquire Core Scientific in an all-stock deal was just that. So now, it’s raising more money through debt that could turn into equity.

There’s no easy fix to Nvidia’s China problem.

The world’s most valuable publicly traded company had extended Monday’s gains during after-hours trading yesterday on the heels of US President Donald Trump’s Truth Social post indicating that the chip designer could begin to sell its H200 chips to China, with 25% of the proceeds going to the US government.

However, the company is reversing those gains this morning, with Nvidia dipping into the red relative to yesterday’s close at its lows, after the Financial Times reported that “regulators in Beijing have been discussing ways to permit limited access to the H200,” according to two people familiar with the matter.

Per the FT, buyers would likely need to go through a lengthy approvals process to get their hands on the H200s — Nvidia’s most advanced chip in its Hopper line, which has since been replaced by the Blackwell generation — and would need to provide an explanation as to why domestic Chinese chips couldn’t perform the tasks at hand.

This feels like déjà vu all over again for Nvidia.

Export restrictions put in place in mid-April during the height of US-China trade tensions prevented the chip designer from sending its H20 chip, a nerfed version of its premium Hopper offering, to China. After that export ban was lifted months later, demand from China “never materialized,” Nvidia CFO Colette Kress said following the company’s Q3 earnings report. Reports suggested that China banned its leading technology giants from purchasing these semiconductors, instead pushing them toward domestic alternatives.

The difference between the H20 and the H200 is one zero (and a lot of computing power). Zero is also the amount of interest that Chinese policymakers would prefer their leading tech companies to have in Nvidia’s chips.

China’s seemingly measured response to its renewed ability to access these chips suggests that the heady thoughts of a $10 billion to $15 billion boost to Nvidia revenues, which Bloomberg Intelligence analysts had anticipated following Monday’s announcement, may need to be tempered.

A bipartisan group of senators doesn’t want China to have access to advanced US chips. Chinese leadership seemingly doesn’t want their tech champions to rely on them. President Donald Trump and Nvidia CEO Jensen Huang, on the other hand, don’t mind if they do.