The event
LibreOffice Shipped 'No AI' as the Headline — and the Market Answered
On August 26, 2026, The Document Foundation released LibreOffice 26.8 with a release-line that read, in full: \"professional typography, deeper support for the world's writing systems, and no artificial intelligence.\" A week later the foundation turned a community comment — \"So, no AI is now a feature?\" — into a blog post of its own titled \"Yes, no AI is now a feature,\" and The Document Foundation's 26.8 hit 1,031,162 official downloads in its first week, more than 2 million within roughly two weeks. The release explicitly framed the absence as a feature, then defended it on audit grounds: \"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.\"
LibreOffice 26.8 first-week downloads
1.03M
Official count, week ending Sep 6, 2026
LibreOffice 26.8 total downloads (early Sep 2026)
2M+
Per LibreOffice Download Metrics, as reported Sep 10, 2026
Microsoft 365 Copilot paid seats
30M+
FY26 Q4 earnings release, Jul 29, 2026
Microsoft Copilot annualized revenue run rate
$37B
Up ~123% YoY, FY26 Q4 disclosure
Adobe–HUMAIN Saudi commitment
$4B+
Aug 31, 2026, covering 27M users
AI productivity tools market, 2026
$14.1B
Grand View Research, base for 14.5% CAGR to 2033
The conditions
Six Conditions Make 'No AI' the Hardest Bar in Productivity Software
The Document Foundation's \"Yes, no AI is now a feature\" post is not a vague privacy pledge; it is a six-point engineering specification that any vendor shipping default AI inside a productivity suite must satisfy. Today, per the post, \"there is no integration that meets all the requirements.\" That wording is the load-bearing piece: it lets LibreOffice ship extensions that connect to local models through Ollama or LM Studio while still owning the headline.
- User-controlled execution — inference runs locally, on user-controlled infrastructure, or on a service the user explicitly picks; no silent fallback to a single provider
- No unauthorized data leaving the computer — every transmission must result from a user action, never background behavior
- No telemetry of any kind — the existing LibreOffice \"no usage data collected\" rule extends to AI features
- No dependence on a single provider — interfaces must be implementable by multiple backends
- No compromises on ODF format — generated output must retain structure, styles, and semantics, blocking the proprietary lock-in path
- Entirely optional — AI features must be installable, removable, and absent from the UI for users who do not authorize them
The signal
The 2 Million Downloads Land on Top of a Quiet Microsoft Pullback
Read against the rest of 2026, LibreOffice's bet stops looking fringe and starts looking like a counter-cyclical indicator. On April 15, 2026, Microsoft began removing Copilot Chat from Word, Excel, PowerPoint and OneNote for commercial customers with 2,000-plus seats, restricting rather than removing it for smaller accounts, and relabeling the tiers as Copilot Chat (Basic) versus M365 Copilot (Premium). Two weeks before LibreOffice 26.8 shipped, on August 31, Adobe announced a partnership with Saudi Arabia's MCIT and HUMAIN to deliver 12 months of free Firefly Standard plus Adobe Express Premium to 27 million residents — a $4 billion-plus commitment that effectively prices Adobe's AI creative stack at zero in one market. On September 29, OpenAI launched Space, a shared workspace inside ChatGPT with agent personas called Dots, directly attacking Microsoft's workplace-software moat from inside its own largest investor's portfolio.
| Vendor | Product | AI posture | Pricing/monetization |
|---|---|---|---|
| LibreOffice (TDF) | LibreOffice 26.8 | No AI in default install; extensions allowed to local models | Free, open source; downloads = 2M+ in first weeks |
| Microsoft | Microsoft 365 Copilot | AI bundled across Word/Excel/PPT/Outlook; Copilot Chat cut from apps for >2,000-seat accounts Apr 15, 2026 | 30M+ paid seats; ~$37B Copilot ARR (FY26 Q4) |
| Adobe | Firefly + Adobe Express | AI-first creative; bundled free in Saudi Arabia via HUMAIN deal | $4B+ commitment covers 27M users for 12 months |
| Alphabet | Workspace with Gemini | Gemini in side panel across Gmail/Docs/Sheets/Slides | $7–$22 per user/month across Business tiers; Gemini Enterprise from $21/seat/month |
| OpenAI | ChatGPT Space (Sep 29, 2026) | AI-native shared workspace with agent Dots | Bundled with $20/month ChatGPT Plus; Work Cloud/Local require paid tier |
Three of the five entries above are pushing AI harder into the default surface area; one is holding it back. The fourth — OpenAI — is essentially taking Microsoft's bundle from the inside out via Dev Day. The third — LibreOffice — sells the absence. The pattern is not \"AI is failing\"; it is that the demand curve is bifurcating between AI-first bundles and AI-optional alternatives, and the incumbents have not yet decided which side to fall on.
Value-chain read
What the Trust Ceiling Means for Microsoft, Adobe, and Google
For Microsoft, the read is the most uncomfortable. The company reported $90.0 billion in FY26 Q4 revenue (filed July 29, 2026) and disclosed Microsoft 365 Copilot had crossed 30 million paid seats with an annualized run rate of roughly $37 billion. That is the monetization engine the AI capex is being justified against. LibreOffice's signal implies that a non-trivial slice of those seats are paying for AI users do not want turned on by default — which is precisely why Microsoft started removing free Copilot Chat from enterprise Word/Excel/PPT and pushing customers to the $30/seat Premium tier. The risk is not that Copilot usage collapses; it is that the average realized price per seat compresses faster than seat growth as procurement teams audit what is actually being used.
For Adobe, the read cuts the opposite way. The HUMAIN deal — Firefly Standard and Adobe Express Premium free to 27 million Saudi residents for 12 months, with a custom image model tuned to Arabic prompts — is a defensive distribution play disguised as a free giveaway. Adobe generated $6.76 billion of revenue in Q3 FY2026 (quarter ended August 28, 2026, filed September 22), and Creative Cloud ARR carries the company; giving Firefly away in one market is the price Adobe is paying to keep the AI-optional cohort from concluding that the entire creative stack is replaceable by a local Stable Diffusion or Flux install.
For Alphabet, the read is the cleanest positive. Workspace with Gemini is priced from $7 per user per month in Business tiers up to $21 per seat per month for Gemini Enterprise. If the AI-optional segment is real and growing, Google has a structural advantage: it owns both the AI model and the productivity surface, so a customer who wants AI but insists on data-residency controls can stay inside Google's stack with a tighter contract rather than defecting to LibreOffice. The risk is opposite to Microsoft's: if AI productivity becomes a commodity, Gemini's moat collapses into Workspace's distribution moat, which Alphabet already has. Watch Gemini attach rates inside Business-tier renewals, not headline Workspace seat growth.
AI productivity tools market — projected trajectory through 2033
Market sizing by Grand View Research, 2026–2033 forecast
Unit: USD billion
2025 (base)
11.1
2026
14.1
2028 (est.)
18.5
2030 (est.)
24.2
2033 (est.)
36.4
Upstream & downstream
The Suppliers and Customers That Get Pulled In
Upstream, the read points to NVIDIA and to Meta Platforms. If enterprise AI workloads begin to bifurcate — a chunk running in vendor clouds, a chunk running locally on Llama-class models through Ollama or LM Studio — the addressable GPU demand for inference softens at the margin even as training demand stays tight. LibreOffice users, per The Document Foundation's own post, are explicitly being steered toward local OpenAI-compatible endpoints, and the dominant open-weights family in that ecosystem is Meta's Llama line. Downstream, the affected parties are enterprise procurement teams writing AI-acceptable-use clauses, regulators auditing where training data leaves the device, and the developer community that is now building \"AI-optional by default\" extensions for an open-source productivity suite with a 2-million-download floor.
TDF AI policy — six non-negotiables
Execution
User-controlled
Local, user infra, or explicitly chosen service
Data egress
Requires user action
No background transmission
Telemetry
None
LibreOffice-wide rule extends to AI
Provider lock-in
Multi-backend
Open interfaces for any vendor
Format
ODF only
Preserve structure, styles, semantics
Opt-in
Required
Installable, removable, absent from UI otherwise
Frontier-model context
Why a 'Trust Ceiling' Beats a 'Compute Ceiling' Right Now
The trust framing lands because the underlying economics have already shifted. Frontier inference pricing collapsed through 2024–2026 — token costs that were roughly $20 per million in 2022 trade near $0.05–$0.14 per million for budget-tier models today, with the deepest discounts from Chinese labs cutting API prices six times in the first half of 2026 alone. Once inference is cheap enough to run locally on a workstation-class GPU, \"the AI sent your document somewhere\" stops being a theoretical risk and becomes the deciding procurement criterion. That is the mechanism behind the LibreOffice signal: the cost of running AI privately has fallen below the cost of explaining why you did not, and the productivity-suite vendors that bundle cloud inference by default are now paying an audit tax their customers did not budget for.
Horizons
What Moves First — and What This Becomes
Over the next two quarters, watch three things. First, Microsoft FY27 Q1 (reported late October 2026) and Q2 disclosures — does Copilot attach inside the Productivity & Business Processes segment shift from seat count toward usage minutes, and does Microsoft start disclosing \"Copilot-opted-out\" seats inside large enterprises? Second, the next Adobe earnings call — does management frame the HUMAIN deal as a one-off or extend the free-bundle template to other regions, and what does that imply for Creative Cloud pricing power? Third, any movement from OpenAI on whether Space is gated to $20/month Plus only or pushed down to ChatGPT Go ($8/month) — a price-cut would be the cleanest signal that the workplace-AI bundling model is under pressure.
Over a 1–3 year horizon, the structural read is that AI privacy becomes a brand asset rather than a niche. The Document Foundation has set the bar; Vivaldi (which shipped \"No AI Slop\" with version 7.8 on January 29, 2026 and an explicit anti-AI positioning in 8.0 on May 21) has done the same on the browser side. The open question is whether any large incumbent will copy the framing — and the betting line is that Microsoft and Alphabet will offer \"AI off\" toggles long before they offer \"no AI\" as a marketing headline, because the first is a feature checkbox and the second is a brand admission. The 2 million downloads are the early evidence that the second framing is what some buyers want to hear.
Investable names tied to the trust-ceiling signal
- Copilot's 30M paid seats and ~$37B ARR are the engine justifying FY26 Q4 capex of $30B+; if enterprise audits force AI-opt-out defaults, realized price per seat compresses even as seat count grows
- Removing Copilot Chat from Word/Excel/PPT in April 2026 for >2,000-seat customers confirms management is already repricing AI from \"free with the suite\" toward Premium-tier monetization
- Over 1–3 years, if \"no AI\" becomes a procurement requirement in EU/regulated buyers, Microsoft must ship an explicit AI-off SKU — a brand concession the company has historically resisted
- Workspace with Gemini at $7–$22/user/month lets Alphabet capture the AI-wanting-but-private-data cohort inside its own stack rather than losing it to LibreOffice
- Gemini Enterprise at $21/seat/month is the cleanest test of whether AI productivity can carry pricing power above the commodity floor set by sub-$0.10/M-token inference
- Over 1–3 years, Alphabet's combination of proprietary model and proprietary productivity surface absorbs the AI-optional segment better than vendors who have to bundle a third-party model
- The $4B+ HUMAIN deal giving Firefly free to 27M Saudi users trades short-term revenue for long-term distribution against local Stable Diffusion/Flux installs
- Q3 FY2026 revenue of $6.76B and Creative Cloud ARR depend on whether AI bundles stay premium-priced in Western markets — the Saudi template tests whether AI-led creative can hold pricing when copycats are free
- Over 1–3 years, Adobe must decide whether to keep Firefly walled (premium) or open it (volume) — the Saudi deal is the first move down the open path
- LibreOffice's local-AI extension ecosystem runs overwhelmingly on open-weight Llama models — every TDF-approved local install is a vote for Meta's distribution model over closed frontier APIs
- If enterprise AI bifurcates between cloud-vendor-AI and local-Llama-AI, Meta's open-weights strategy monetizes through engagement and ads rather than inference revenue — a different but defensible angle
- Over 1–3 years, the open-weights stack that LibreOffice users adopt becomes the de facto privacy-tier AI standard, giving Meta influence over the productivity stack without selling a productivity suite
- The 26.8 \"no AI in default\" stance, if copied by other suites, redirects a slice of inference workload from hyperscaler GPUs to local RTX-class cards — neutral on training, ambiguous on inference TAM
- Watch NVIDIA's next earnings disclosure of inference-versus-training revenue mix for any softening at the low end that signals enterprise AI workload bifurcation
- Over 1–3 years, if local AI becomes a procurement requirement in regulated industries, NVIDIA's workstation-GPU segment becomes the under-appreciated growth lever alongside datacenter
