AI & SaaS News, Nov 3: Grok Limits, Anthropic Q3

Last checked November 3, 2026 · News roundup

Today in brief: Grok Bot subscribers say their weekly usage is running out days early, and Elon Musk says serving the tool cost too much relative to revenue. Anthropic's amended IPO filing puts third-quarter revenue near $17 billion, Goldman Sachs pulled Claude Code and GitHub's coding tools from bankers and traders, and a new benchmark found the best model solving only 12% of hard professional tasks. On the business side, Amazon is canceling some Nvidia hardware orders, SUSE is cutting about one in ten jobs, and Marvell agreed to buy Lumentum in an all-stock deal.

Part 1 · AI

Grok Bot usage limits: why weekly allowances are running out early

If your Grok Bot allowance vanished faster than usual this week, you are not alone. Subscribers say their weekly usage ran out days ahead of schedule, and one widely shared report described the meter jumping from 56% to 100% within an hour, according to Gadget Review's November 2 report syndicated on Yahoo Finance. TechSpot reported that some users felt their effective limit had been cut roughly in half. These are user reports, not measurements published by xAI.

Elon Musk did not deny the squeeze. He wrote on X that "the cost of serving Grok Bot was unsustainably high relative to revenue." GuruFocus noted that he did not confirm a specific change to the caps, and no advance notice of a change has been found.

Grok Bot is sold through X's Premium ($30 a month), Super ($100) and Ultra ($200) plans, with larger allowances on the higher tiers, per the same report. It is not clear whether all three tiers were affected. Earlier, SpaceX President Gwynne Shotwell reportedly cited about 800,000 paying weekly users in early October.

What to do: check the usage meter before starting a long job, split big tasks across the week, and hold off on any annual prepayment until xAI publishes the actual limits. For a sense of how another vendor documents its caps, see our breakdown of ChatGPT Pro's weekly limits.

Anthropic Q3 revenue hits about $17 billion in the amended S-1

Anthropic's updated IPO filing, made public on November 2, shows third-quarter revenue of nearly $17 billion. That is more than $5 billion above the second quarter and roughly 12 times the year-earlier figure, Yahoo Finance reported. CNBC put nine-month revenue at $33.24 billion.

The profit picture moved the other way. Anthropic swung to a $567 million operating loss in the quarter after $974 million of operating income in Q2. The company blamed "$5.0 billion of compute under short-term commitments," which costs more than multiyear contracts. Compute and infrastructure charges reached $9.9 billion for the quarter, and the filing also lists $2 billion of non-cash charges tied to an employee donation-matching program. Total compute and infrastructure commitments come to about $895 billion, nearly half of it expected by the end of 2030. CNBC reported a Q3 net loss of $23.29 billion.

Still missing: a price range, share count and offering size. Nothing in the filing changes Claude plans or prices for users. Our October 27 roundup covers the original filing, including the ANT ticker on Nasdaq.

Goldman Sachs limits Claude Code and GitHub Copilot to its engineers

Goldman Sachs has taken Anthropic's Claude Code and Microsoft's GitHub coding tools away from bankers and traders and kept them for its roughly 12,000 engineers, Bloomberg reported on November 2. PYMNTS, citing the same report, said internal guidance published last month sets out the restriction, and that bankers had been allowed to use the tools for months before that.

A Goldman spokesperson, Nick Carcaterra, framed it as the end of an experiment: the bank wanted to put the tools "in the hands of those who are best suited to leverage them." Some traders were reportedly frustrated, since they had used the tools to watch client trading flows and to build written and graphic materials. The bank still makes other AI tools available to staff, though the reports do not say which.

The move matters beyond Wall Street. Agentic coding tools are billed per seat or by usage, and they can touch production systems, so a seat audit is a quick way to cut both cost and risk. If you manage these tools for a team, list who opened a coding agent in the last 30 days and move occasional users to a chat assistant. Our Claude Code guides and reviews cover setup and safe use.

GDPx benchmark: the best AI model finished only 12% of real-work tasks

SemiAnalysis launched GDPx on November 2. It tests agents in realistic environments across 11 domains where the authors expect AI to earn the most money: recursive self-improvement, hardware design, chip design, cybersecurity, biology, finance, legal, quant trading, materials science, alignment and software engineering. The top overall score was 12%, and the authors call it a true frontier benchmark.

Each lab wins somewhere. Opus 5.5 is the strongest all-rounder and leads quant trading, hardware design and chip design. Gemini 4 Carbon leads biology and software engineering, while GPT-6 Astra leads recursive self-improvement and materials science. Finance and legal tasks were so hard that nearly every model scored 0% under pass-or-fail grading, so the authors report partial credit there. They also note that safety classifiers pull down scores for Opus 5.5 and GPT-6 Astra in cyber, biology and alignment.

For buyers, the lesson is plain: test the model on your own domain instead of trusting one leaderboard rank.

Meituan LongCat-2.5 Preview: a 1.6-trillion-parameter model with open weights promised

Meituan, the Chinese delivery giant, released LongCat-2.5 Preview on November 2. TechNode reports 1.6 trillion total parameters with about 48 billion active per token, which is a mixture-of-experts design that keeps running costs closer to a mid-size model. Text and images share one token framework, aimed at visual coding, creative design and 3D tasks. Meituan says it trained the model on long task sequences and turned real agent failures into reproducible training tasks for large codebases. AI Weekly lists a 1-million-token context window.

It is available now through Meituan's chat service and API. GitHub and Hugging Face releases are "planned to follow," so open weights are promised but not yet out. No pricing or benchmark results were published.

US teams should treat it as a model to watch rather than to deploy. Wait for the weights and an independent evaluation, and run any API trial on non-sensitive data only, since requests go to a China-based provider.

Part 2 · SaaS & business

Amazon canceled DGX Spark and Nvidia GPU orders: what buyers are reporting

Some Amazon shoppers buying AI hardware are seeing orders canceled, and in rare cases their accounts closed, Wccftech reported on November 2. One buyer, posting as @nicksaraev on X, said four separate Nvidia DGX Spark orders placed October 31 were all canceled, followed by an email saying Amazon had completed a verification process and decided to close the account. Epic Games CEO Tim Sweeney said orders were canceled shortly before delivery after he ordered two Nvidia GPUs. His account stayed open.

Amazon has not commented. Nobody knows yet whether this is a glitch or a deliberate limit aimed at scalpers buying several units at once.

If you are buying a DGX Spark or a high-end GPU: order one unit at a time, make sure the shipping and billing names match the account, and keep the confirmation emails. If an order is canceled, check that the charge was reversed before you reorder. Buying from Nvidia directly or from another authorized retailer avoids the problem for now.

SUSE layoffs: about 10% of staff cut, with AI named as the reason

SUSE, the enterprise Linux and Rancher vendor, will cut about 10% of its roughly 1,900 employees, The Register reported on November 2 from a staff memo. CEO Dirk-Peter van Leeuwen wrote that the company had reviewed its "future strategy, required capabilities, and team structures" and that "the requirement for some existing roles has greatly diminished." The memo names "automation and AI" as the driver.

The timeline is short. Affected staff hear by November 4, employee representatives are nominated the next day, and one-to-one meetings start the following week. SUSE will share its fiscal 2027 strategy internally on November 16. The company is owned by private equity firm EQT.

Customers have not been told about product changes. The memo talks about "deprioritizing" some areas without naming them. If you run SUSE Linux Enterprise or Rancher, note your renewal date and ask your account team about the roadmap for the specific products you rely on after the November 16 strategy update.

Marvell buys Lumentum: the all-stock deal behind AI data center lasers

Marvell Technology agreed to buy Lumentum, which makes the lasers and optical parts that move data between AI servers. Under the merger filing, each Lumentum share converts into 4.5 Marvell shares. Marvell holders will own about 67% of the combined company, and the deal is expected to close in the first half of 2028. CNBC put the implied enterprise value at $115 billion, or about $1,132 per Lumentum share, a 23.9% premium to the October 28 close.

The companies expect more than $1 billion in yearly cost savings once the deal is done, mostly in cost of goods sold. The deal includes break fees: Marvell pays $5.8 billion in some cases and $3.2 billion if regulators block it, while Lumentum owes $2.9 billion in matching cases. Marvell CEO Matt Murphy stays on as chairman and CEO, and three Lumentum directors will join the board.

Software buyers see no direct change, but optical links are one of the bottlenecks in AI cloud capacity, and a long regulatory review is likely given the size.

Oracle, Alphabet and Meta bond sales: the AI debt wave in one table

Three of the biggest AI spenders went to the debt and equity markets within days of each other. Oracle began selling $10 billion of stock and $10 billion of senior notes, and CNBC reported its shares fell more than 6%. Alphabet sold €6 billion of euro bonds. Meta is sounding out investors for a bond of at least $20 billion, according to Bloomberg.

Why it matters to SaaS customers: this money pays for the data centers behind cloud AI services. None of the three has announced customer price changes, but borrowing at this scale shows the companies expect AI demand to keep growing. If you are negotiating a multiyear cloud or AI contract, ask for price caps on renewal.

Saazora take: two AI billing changes to check before December

Two older announcements now have dates close enough to act on. Google will start moving Gemini Gems to the new skills feature on November 17, according to TheStreet, and it is not confirmed that Gems with attached files, default tools or share links will carry over intact. Atlassian starts billing extra Rovo AI usage on December 3: Jira and Confluence include 25, 70 or 150 credits per user each month on Standard, Premium and Enterprise, and each credit beyond that costs $0.01. Extra usage is on by default, and admins can set a spending limit or switch it off in Atlassian Administration.

Our SaaS price change tracker follows changes like these across tools. Before December 3, an Atlassian admin should check last month's Rovo credit use and set a cap. Before November 17, copy the instructions of any Gem you depend on into a document so you can rebuild it if the migration drops something.

Four questions about today's AI and SaaS news

Why is my Grok Bot usage running out so fast?

Users report weekly allowances emptying days early, and some believe the effective limit was roughly halved. xAI has not published the old or new caps. Elon Musk said serving Grok Bot cost too much relative to revenue, which suggests the tighter limits are deliberate rather than a temporary bug.

When is the Anthropic IPO?

No date is set. The amended S-1 filed publicly on November 2 adds third-quarter results but still leaves out a price range, share count and offering size. Those normally appear in a later amendment shortly before the roadshow, so the listing date should become clearer once that filing arrives.

Which AI model scored highest on GDPx?

Claude Opus 5.5 had the best overall score at about 12%, ahead of Gemini 4 Carbon and GPT-6 Astra, based on the launch summary. Rankings change by domain: Gemini 4 Carbon leads software engineering and biology, while GPT-6 Astra leads materials science and recursive self-improvement.

Will Atlassian charge for Rovo AI credits?

Yes, from December 3, 2026, but only beyond the included allowance. Jira and Confluence plans include 25, 70 or 150 credits per user per month depending on tier, pooled across the organization. Extra credits cost $0.01 each, and admins can cap or disable extra usage.

Sources: Yahoo Finance / Gadget Review, Nov 2, 2026 · GuruFocus, Nov 2, 2026 · Yahoo Finance, Anthropic S-1, Nov 2, 2026 · CNBC, Nov 2, 2026 · PYMNTS, Nov 2, 2026 · SemiAnalysis, Nov 2, 2026 · AI Weekly, Nov 2, 2026 · TechNode, Nov 2, 2026 · Wccftech, Nov 2, 2026 · The Register, Nov 2, 2026 · Lumentum Form 425 (SEC), Nov 2, 2026 · CNBC, Marvell–Lumentum, Nov 2, 2026 · The Next Web, Nov 2, 2026 · CNBC, Oracle, Nov 2, 2026 · TheStreet, Sep 30, 2026 · Atlassian Support, checked Nov 3, 2026