Software earnings week: Anthropic expands startup credits as LibreOffice says no to AI

Anthropic used Tuesday to push further into the pipeline business. It is expanding Claude Startups, its program for founders. Startups founded within the past five years, or funded within the past two, are eligible to apply, according to CNBC.

Members can receive up to $45,000 worth of discounts and credits through what Anthropic calls the Claude Startup Stack. That package includes a one-time $1,000 API credit and a free year of Claude Team for up to five premium seats. The program launched in May and already has thousands of companies participating, the company said.

"Startups are often the first to push Claude to its limits," Beth Robertson, head of startups at Anthropic, told CNBC in a statement. "When a new model ships, they're ready that same day to take on projects that used to seem out of reach."

The timing is not neutral. CNBC notes that the vast majority of Anthropic's revenue comes from business customers. The company has spent much of the last year fending off OpenAI and Google, which run their own startup programs, and it is widely expected to pursue a blockbuster IPO. Startup credits are cheap customer acquisition when the alternative is competing on model benchmarks alone, and the credits buy logos that benchmarks cannot.

The other side of the same bet

Not every software maker is adding AI. The Document Foundation, which makes the open source editor LibreOffice, said in a blog post this week that it "will not add" AI to its software for the foreseeable future, TechCrunch reported on Tuesday. The group had already said in late August, with the latest release, that the software "contains no generative AI features."

Its argument is auditability. "Any organisation handling confidential, legally privileged or personal data needs to know where that data goes, and the only assurance that survives an audit is that it does not leave the machine," the organization said in the post. Users who want AI can install extensions that connect to local models. The foundation said its criteria are not a "definitive rejection" but an assessment of the technology as it stands.

That position cuts against the annual earnings story that software vendors have been telling investors. AI features are the pitch for higher prices and stickier seats. LibreOffice is arguing the opposite, that not shipping them is itself a feature, and it is doing so while claiming tens of millions of users.

For a software company reporting quarterly numbers, that is the uncomfortable part. Buyers who care about data residency now have a free option, and it costs nothing to evaluate.

Security costs land on the same balance sheet

The week's largest data incident came out of Denmark. Unauthorized parties accessed the names, addresses and personal identification numbers of about 8.8 million people, living and dead, in the national population register, the country's digitalization ministry said on October 5, as reported by The Hacker News. The route was a private Danish company's lawful right to look up records in the Central Person Register.

The register holds about 11 million people, so the access covered roughly four in five of them. The access lasted about 10 days in September, according to Christina Egelund, the minister responsible for digitalization, who spoke to the news agency Ritzau. An employee noticed unusual activity on Friday, October 2. Datatilsynet, the data protection authority, said a very large number of automated lookups were made to identify valid CPR numbers, and its notice, dated October 5, describes the numbers as allegedly retrieved. The 8.8 million figure is the ministry's and is not final.

Three questions remain open in the official statements: how the attackers got into the company's systems, whether the data has been retained or used, and who they are. The ministry points users to sikkerdigital.dk and to the government cyberhotline.

This is the kind of event that shows up in enterprise software budgets rather than in a single vendor's revenue line. BleepingComputer, in a sponsored piece published Tuesday, laid out the scale of the management-plane problem: in September 2026, N-able shipped an emergency hotfix for CVE-2026-86218, a maximum-severity pre-authentication remote code execution flaw in its N-central RMM platform, its fourth hotfix in five weeks, with roughly 1,500 servers exposed online. The same article cites Microsoft SharePoint's "ToolShell" zero-days in July 2025, exploited before a patch existed, with at least 85 on-premises servers compromised.

Every one of those numbers is a future line item, either for the vendor or for the customer.

Money moving into the harder half of the stack

Away from the AI feature arms race, the hardware and climate side of the market is pulling capital into things that take longer to ship. Glimpse, which sells image-processing software for CT scanners, told TechCrunch it has more than 100 customers including Anker, Lucid and the U.S. Navy, and generates revenue in the "mid seven-figure" range, according to co-founder and CEO Eric Moch. It is launching a product called Explore for aerospace, automotive, electronics and medical device customers, and says its software lets battery makers inspect cells 10 to 30 times faster on existing scanners.

The reference point Moch gives is a recall: improperly installed parts inside LG battery cells sidelined every Chevy Bolt made between 2017 and 2022 and cost LG $1.9 billion.

MIT Technology Review published its annual list of 10 climate tech companies to watch on Tuesday, and its editors were explicit about what changed. The 2026 list is largely composed of companies based outside the US, after federal regulatory rollback and cuts to government incentives, with two Chinese firms included: WeLion, building semi-solid-state batteries for EVs, and Envision Energy, installing wind power and storage. Energy Dome, Moment Energy and Form Energy make the list on the storage side, and the editors name rising electricity demand from AI data centers as one of the biggest stories of the year.

Canada Rocket Company is making a similar bet on infrastructure with a longer horizon. The company said on October 1 that it will build what it calls Canada's first large-scale rocket engine test facility at London International Airport, a $30 million complex, leasing 50 acres for up to three test stands and a 12,000-square-foot office and shop building, according to NASASpaceFlight.com. It expects the site to be fully operational by 2028 and about 40 skilled jobs within 18 months. "Canada's path to its own access to space starts on the ground," CEO and co-founder Hugh Kolias said.

None of this will appear in a software earnings call this quarter, which is rather the point. The companies selling AI features are competing for the same startup logos with credits; the ones selling test stands, scanners and storage are competing for capex that takes years to convert. Both are being priced as growth.

One more data point from the past week sits awkwardly between the two. Magnitude, an open source inference engine for AI agents, claims in its GitHub README that open models run up to 2x faster than llama.cpp on it, with 92% faster decode on Metal and 19% on CUDA, kernels tuned on the user's own hardware, 27% less memory per agent, and an Apache 2.0 licence. It ships with a desktop app and connects to Pi, OpenCode, Hermes, Codex, Claude Code and others. The benchmark claims are the company's own and have not been independently verified, but the pitch, performance per dollar on hardware you already own, is the one that tends to survive a downturn.