11 Votes, 200 Billion Gigabytes: France's Per-Byte Internet Tax and What It Costs Cloud, Video and AI

A procedural vote in Paris on 30 October 2026 has produced a tax design with cross-border consequences: companies that send data to users in France would pay EUR 0.055 for every 1GB they deliver — 5.5 euro cents per gigabyte. Dubbed the "digital data producer tax", the measure cleared France's lower house, the National Assembly, by 11 votes in favour against 10 votes against, with 14 members abstaining. Nothing is in force yet: the amendment must still pass the Senate, survive the rest of the 2027 budget procedure, and possibly be rewritten by the government itself. But for anyone who ships video, games, software updates or AI model weights, the detail that matters is not the rate — it is the unit. France is proposing to tax bytes rather than revenue.

The vote, the mandate and the numbers at a glance

• Tax unit: EUR 0.055 per 1GB of data delivered to French users, i.e. 5.5 euro cents per gigabyte.
• Vote: on 30 October 2026, France's National Assembly (lower house) adopted the amendment with 11 votes in favour, 10 votes against and 14 abstentions.
• Turnout: the National Assembly has 577 members; only 35 took part in this ballot, and just 21 chose between support and opposition — the proposal was carried by 11 ayes.
• Sponsor and vehicle: tabled by Laurent Lhardit of France's Socialist Party as an amendment to the 2027 finance bill.
• Pre-existing opposition: both the French government and the parliament's budget rapporteur had publicly rejected the idea before the vote.
• Claimed yield: about EUR 10 billion a year for the French treasury.
• Safe harbours: a company's bill may not exceed 15% of its revenue in France or 2% of its worldwide revenue.
• Tax base, as cited: using figures from French telecoms regulator Arcep, the amendment states that internet users in France receive about 200 billion GB of data per year, and that Google, Meta, Amazon, Netflix and Akamai — five companies — account for about half of it.

How the levy is designed to work

The architecture is unusual in one specific respect: the state does not measure the traffic itself. Under the proposal, internet service providers would become the metering agents, recording what each content company pushes toward French customers and filing those volumes with the French tax authorities. Content providers would then settle at EUR 0.055 per GB, with a different calculation for companies that distribute through content delivery networks (CDNs).

Design element	What the amendment says	Practical effect
Taxpayer	The content provider sending data to users in France	Not end users, and not the operators who carry the traffic
Tax unit	Each 1GB transferred	Revenue-blind: byte-heavy services are hit regardless of price
Metering	Orange, SFR, Bouygues Telecom and Free log per-provider volumes and report to French tax authorities	Competitors become the measurement infrastructure
CDN cases	Assessed on total traffic flowing out of the CDN network	Network architecture, not content, sets the bill
Caps	15% of French turnover, or 2% of global turnover	Bounds exposure for the largest platforms
Status	Lower-house amendment to the 2027 budget	Senate review and Article 49.3 still ahead

The yield arithmetic: EUR 10 billion implies far more than five companies

Set the sponsor's revenue projection against his own traffic data and a gap appears. About 200 billion GB multiplied by EUR 0.055 comes to roughly EUR 11 billion — almost the entire projected haul. Yet the five companies that the proposal is pointed at account for only about half of inbound French traffic, which would generate something nearer EUR 5.5 billion. The missing half is the rest of the internet: cloud and hosting services, streaming beyond the large platforms, software and game updates, public-sector sites and smaller French publishers.

In other words, either the levy in practice bills a far broader set of companies than the "data producers" it is branded against, or the EUR 10 billion estimate is optimistic. This is the same tension that France's minister for public action and finance, David Amiel, expressed in political terms: on his reading of the traffic data, the burden would not land mainly on US technology groups, and about half of it could fall on comparatively small companies, many of them based in France.

Metering is the weak point — legally as well as technically

Amiel's first objection was operational: attributing traffic by the source of content is not something today's networks do natively. Accurate counting, he argues, may require dedicated flow sensors inside the network, or even measurement performed through users' devices.

There is a second, harder obstacle that the amendment text does not address: EU open-internet law. Article 3(3) of Regulation (EU) 2015/2120 requires providers of internet access services to treat all traffic equally, without discrimination, restriction or interference, irrespective of the sender, receiver, content or application involved, and permits traffic management only where it is transparent, non-discriminatory, proportionate and independent of commercial considerations. Privacy and net-neutrality bodies that reviewed BEREC's implementing guidance have gone further, concluding that deep packet inspection is very unlikely to be deployable lawfully when used for commercial differentiation — a position the European Data Protection Board has also signalled. That leaves France with two imperfect routes: identify content inside the network, which collides with the open-internet and GDPR frameworks, or rely on content companies' self-declared volumes, which converts the scheme into a filing-and-audit obligation that the smaller taxpayers are least equipped to meet (see eur-lex.europa.eu, Regulation (EU) 2015/2120; epicenter.works on the BEREC guidelines).

The measurement problem cannot be separated from the commercial one. National regulators already police whether operators may handle traffic differently for business reasons: Germany's Bundesnetzagentur administers the same EU rule and has fielded long-running complaints against Deutsche Telekom over practices that critics describe as commercial differentiation of traffic and of the networks that host content. A scheme that requires Orange, SFR, Bouygues Telecom and Free to sort their networks by named content companies sits close to that regulatory red line (bundesnetzagentur.de).

The small-player problem: VLC, open source and a bill with no invoice behind it

VideoLAN has raised the alarm most loudly. The non-profit develops the open-source player VLC, which has logged more than 6 billion downloads, and it estimates the new levy could cost it more than EUR 1 million a year. Nothing about that number depends on monetisation: bytes are bytes, and a free project that distributes software at scale looks, to a per-GB tax, almost identical to an advertising-funded platform.

Exposure without revenue is the structural flaw. Groups that could inherit a tax bill despite no meaningful French income stream include:

1. open-source maintainers and free-software foundations distributing installers and updates;
2. universities, research consortia and public datasets serving large files to students and researchers;
3. public broadcasters and cultural institutions offering on-demand video;
4. independent game studios and small software vendors whose installers now routinely run into tens of GB;
5. hosting resellers and SMB application providers whose traffic is aggregated under a CDN's outbound total.

Because the proposal sets a cap — 15% of French revenue or 2% of global revenue — a company with no French revenue has no obvious ceiling to shelter behind, and no obvious taxable base from which to pay. Resolving that asymmetry, not the headline rate, is where the Senate's revising powers will be tested.

Who actually pays: pass-through risk and a trade fight France has already lived

Laurent Lhardit's defence is that end users will not be billed directly: platforms such as Facebook, Instagram and TikTok give their services away, so there is nothing to attach a surcharge to. Philippe Juvin, the French parliament's budget rapporteur, draws the opposite conclusion — large platforms have enough pricing power to push a new cost onto advertisers, app developers, merchants and other commercial partners rather than absorb it.

French experience supports Juvin. After France introduced its 3% digital services tax in 2019, Amazon passed part of the cost to sellers on its French marketplace by raising fees; Apple and Google later adjusted charges for French app developers and advertising customers.

There is also an external dimension that the domestic debate tends to skip. The 3% DST drew an immediate trade response: the Office of the United States Trade Representative opened a Section 301 investigation in July 2019, determined in December 2019 that the French tax was discriminatory and burdened US commerce, and published retaliatory tariff proposals on French goods — champagne and cheese among them — while the Congressional Research Service noted that France suspended collection in January 2020 pending the OECD process. The five DST investigations, including Italy's and Spain's, were formally terminated in November 2021. A French per-GB tax that names Google, Meta, Amazon, Netflix and Akamai and leans on Arcep data showing five companies moving about half of French traffic would, even if facially nationality-blind, reopen that channel with Washington (see ustr.gov; congress.gov CRS report IF11564; federalregister.gov, Notice of Action, 16 July 2020).

Why AI and data teams should read this as an infrastructure bill

For a general AI-industry readership, the amendment is interesting less as tax policy than as a bet about where network bytes come from over the next decade. A revenue-based digital tax is anchored to advertising and marketplace economics; a byte-based tax is anchored to whatever the traffic curve does — and the fastest-moving segments of that curve are precisely the ones this site covers.

Cisco's 2026 study of AI's impact on wide-area networks puts today's AI inference traffic as still negligible next to dominant categories such as video streaming, but models a much larger role for agentic workloads: enterprise network traffic rising toward roughly nine times current levels over the next decade where agents are widely adopted, against about 2.5 times if they are not; coverage of Cisco's WAN research describes consumer-driven traffic multiplying around 6.6 times by the mid-2030s. Meanwhile model distribution has become a mass-download business: Meta's Muse Spark Open, an open-weight checkpoint of approximately 816 billion parameters, is delivered as files users pull from the Hugging Face hub. Any levy that bills on bytes delivered will eventually bill on weights, datasets, embeddings and inference payloads as well — at EUR 0.055 per 1GB, a single 100GB weight pull is EUR 5.5 and a multi-hundred-gigabyte release is a five-figure tax event per distribution.

The comparison that makes this concrete is the market price of moving data. Amazon Web Services lists internet data transfer out at around USD 0.04 to USD 0.09 per GB depending on volume and route (aws.amazon.com/s3/pricing). At those benchmarks, a French tax of EUR 0.055 per GB is not a rounding error layered on delivery cost — it is comparable to, or larger than, the transport itself. Rational engineering responses follow immediately: quantised and smaller default checkpoints, region-sharded object storage, more aggressive CDN caching inside France, streaming inference replacing bulk downloads, and slower adoption of byte-heavy open-weight distribution into the French market.

Consumer content is exposed on the same arithmetic. If Grand Theft Auto 6 (GTA 6) ships a download as large as the 210GB figure cited in coverage of the game, a single copy transferred to a French user would attract about EUR 11.55 in tax at the proposed rate — and high-definition video, game installers and large software updates routinely move tens of GB and hundreds of GB. Video platforms, cloud storage, game publishers and software distribution networks would all face a cost line tied to usage rather than to price.

What still has to happen before this is law

1. Senate review. The upper chamber is dominated by the Republican Party (Les Républicains), which is not sympathetic to a Socialist-sponsored levy, and French legislative practice gives the Senate the last word in the event of disagreement — a materially adverse setting.
2. Budget committee and plenary readings of the 2027 finance bill, where the amendment may be narrowed, delayed or traded for other revenue measures.
3. Article 49.3. The government can force a finance bill through the National Assembly without a vote and, in doing so, decide which amendments survive into the final text. The tool was used to adopt the 2026 budget in January 2026 after talks stalled, which demonstrates both its effectiveness and its political cost, since it exposes the government to a no-confidence motion.
4. Legal review. Compatibility with EU open-internet law and state-aid/single-market rules, plus GDPR constraints on traffic measurement, would likely be litigated even if the tax is enacted.
5. Trade follow-up. US and European industry reactions would shape the practical scope, as the 3% DST experience showed.

FAQ: the six questions this story raises

Is France collecting this tax now?

No. On 30 October 2026 the lower house adopted an amendment to the 2027 budget by 11 votes in favour, 10 votes against and 14 abstentions. The French government and the budget rapporteur opposed it, and the Senate has not examined it.

Who is legally liable for the payment?

Content providers delivering data to French users, at EUR 0.055 per 1GB, capped at 15% of French turnover or 2% of global turnover. Ordinary internet users are not taxpayers under the text, and neither are the ISPs, though Orange, SFR, Bouygues Telecom and Free would have to measure and report volumes to the French tax authorities.

Why does the vote count matter?

Because it shows how little deliberation the measure received relative to its reach. Of the National Assembly's 577 members, 35 voted and only 21 distinguished support from opposition; 11 ayes were enough to approve a rule that could alter the business model of global platforms.

Could a free or non-commercial service be taxed?

Yes — that is VideoLAN's objection. The VLC developer, with more than 6 billion downloads behind it, estimates a bill of more than EUR 1 million a year, with no advertising revenue to offset it.

Will users see higher prices anyway?

That is the central dispute. Laurent Lhardit says no, because Facebook, Instagram and TikTok sell free services; Philippe Juvin says platforms can shift the cost to advertisers, developers and merchants. France's own 3% digital services tax from 2019 was passed through by Amazon to marketplace sellers, and later echoed in Apple and Google fee changes.

Does this touch AI workloads?

If enacted as drafted, yes: the tax base is bytes, and AI is becoming one of the byte-heaviest categories of internet traffic. Open-weight releases are downloaded as files of tens of GB and hundreds of GB, and traffic forecasts point to fast growth in agentic and inference workloads — meaning a per-GB levy would function partly as a tax on model and data distribution.

Bottom line for the AI and data sector

France has spent two decades pioneering revenue-based digital levies — the 3% DST it introduced in 2019, the pass-through fights that followed, the Section 301 confrontation with the United States that came with them. This amendment is the first serious attempt in a major economy to move the tax base from revenue to volume. Whether or not it survives the Senate, the direction of travel is what should be priced in: internet traffic itself, rather than the commercial value extracted from it, becoming a taxable event.

Three things to watch through the 2027 budget cycle: whether the rapporteur's measurement objections are answered with a self-declaration model, which would shift compliance onto the smallest taxpayers; whether the "digital data producer" definition is widened to include CDN and cloud customers, which would determine if VideoLAN-type projects and AI model hosting are genuinely in scope; and whether Brussels and Washington treat the levy as a new trade measure, given that Google, Meta, Amazon, Netflix and Akamai — five companies, about half of French inbound traffic — sit squarely inside the Arcep data the sponsor relies on. For operators of AI infrastructure and data products, the practical takeaway is narrower but useful: under a EUR 0.055 per 1GB regime, byte efficiency stops being an engineering nicety and becomes a tax strategy.