Summary
- B92 has a genuine network operating surface: AS9081 is active, originates 4,096 unique IPv4 addresses through three overlapping route announcements, uses two observed upstreams and has valid route-origin authorisation for every visible prefix. Its RIPE NCC membership also carries a larger 8,192-address allocation and an IPv6 allocation, although no IPv6 route was visible in the reviewed routing snapshots.
- The economic evidence points to media, not retail access, as the paid business. B92 publicly sells television ratings, digital impressions, sponsorship and branded content; its regulatory registrations describe a national television and online media provider. No current consumer or business broadband tariff, coverage map, service-level offer, subscriber count or downstream network was found.
- Network autonomy can still earn a return without a telecom invoice. It can keep editorial, advertising and distribution operations reachable during carrier trouble, preserve stable addressing and routing policy, reduce migration friction, improve leverage with suppliers and move a portion of outage and lock-in risk back under B92's control.
- The investment case is conditional. Two upstreams are better than one, but both visible paths remain a narrow supplier set; no public exchange presence or IPv6 origination was found. B92 should therefore be judged as a media company with useful network capability, not as a hidden regional ISP, unless segmented revenue and customer evidence show otherwise.
The economic incentive comes before the router
A media business does not need to sell internet access for network autonomy to pay. It only needs the cost of being unreachable, delayed or dependent on one carrier to exceed the cost of maintaining its own routing capability. For B92, the relevant lost unit is not a megabit sold to a subscriber. It is an advertisement not delivered, a live stream interrupted, a television campaign underperformed, a portal visit lost, a distribution partner unable to receive content, or a newsroom unable to publish while events are moving fastest.
That distinction separates direct telecom revenue from internal value creation. A conventional regional ISP builds access or hosting capacity, signs customers, invoices recurring fees and tries to spread fixed network costs across a growing subscriber base. B92's visible commercial doors lead elsewhere. Its current marketing page offers rate cards for television and the b92.net site. Its legal notice identifies the company behind the portal, names digital-content and sales leadership, and lists news, sport, business, lifestyle, travel, technology, automotive, health, culture and video sections. Its privacy notice says the company deals with clients under advertising, content-distribution and other cooperation contracts. These are the contracts of a media owner.
Yet the same company has maintained internet-number capability for decades. That capability can create value in four ways. First, it can improve resilience by permitting more than one carrier path and by keeping addresses portable between those paths. Second, it can increase distribution control by letting B92 set routing policy for services it chooses to expose from its own space. Third, it can reduce switching friction because an address plan owned by the enterprise does not have to be renumbered every time a managed-connectivity contract changes. Fourth, it can transfer risk: instead of leaving every outage, routing decision and remediation timetable with one supplier, B92 accepts some engineering responsibility in exchange for more control.
The trade is not free. Routers, security, monitoring, skilled staff, transit, facilities, power and incident response cost money. Autonomy also means there is no carrier to blame for errors made inside AS9081. The question, therefore, is not whether an autonomous system sounds strategically impressive. It is whether the avoided losses, operating flexibility and bargaining benefits are worth the recurring cost and managerial attention. The public record suggests that they can be, but on an insurance model rather than a retail-telecom model.
Identity, licences and control
The relevant legal identity is unusually well anchored. Serbia's Regulatory Authority for Electronic Media lists Radio difuzno preduzece B92 drustvo sa ogranicenom odgovornoscu, Beograd as the provider of TV B92, with registration number 07528604, tax number 100049242 and an address at Ikarbus 3 Nova 19 in Zemun. The terrestrial-service registration describes a general-programme service distributed nationally through MUX 1, with an approval valid from August 2022 to August 2030. A separate cable, satellite and IPTV registration runs to July 2032. The company's own imprint gives the same address and registration context for B92 NET.
The media regulator names Astonko d.o.o. Beograd as B92's immediate 100% owner. Its published ownership chart then follows a chain through Astonko Holdings, Lake Bade Holdings, Antenna Stream TV, Antenna Serbia and Kopernikus entities in Cyprus and the Netherlands, with Srdjan Milovanovic appearing in the chain. A commercial business-information page instead compresses the owner description to Kopernikus Corporation (Cyprus) Ltd. The regulatory chart is the better source for the immediate legal structure; the difference is a reminder that group control and the name attached to the operating company are not always the same thing.
Control matters economically because B92's ad inventory is already sold in a group context. The 2026 television rate card is headed Kopernikus Corporation and grants volume discounts based on an advertiser's combined annual budget on Prva and B92. It directs sales contact to Prva TV at the same Zemun address. That arrangement can lower selling costs and improve the group's negotiating position with large buyers, but it also makes standalone B92 economics less transparent. Shared sales, content, facilities, distribution or corporate services can create genuine scale economies while obscuring which brand earns which margin.
The network identity is just as specific. RIPE NCC's Serbian member list identifies "B92" DOO BEOGRAD as a registry member, and the registry handle attached to AS9081 is rs.opennet. The name is a historical clue, not proof of a current product. A RIPE presentation on internet access under sanctions records that the OpenNet project connected Radio B92 to Amsterdam in 1995, when Serbian internet access was scarce and politically consequential. That origin explains why connectivity capability may sit unusually close to B92's institutional identity: distribution independence was once central to the outlet's ability to reach an audience at all.
The continuity between that history and today's corporate network should not be romanticised. Ownership, editorial direction, technology and distribution markets have all changed. But the economic lesson survives. When access to an audience is the product, communications infrastructure is not merely office overhead. It is part of the chain that turns content into cash and influence.
The paid product is attention
B92's public prices reveal the operating model more clearly than the ASN does. The 2026 television price card uses cost per rating point for a 30-second advertisement against an all-persons aged four and above target. It sets B92's base at EUR80 outside prime time and EUR120 in the 18:00-24:00 window, before seasonal factors, position factors, volume discounts and tax. This is not a flat EUR120 spot price: delivered rating determines the bill. The schedule offers combined Prva-and-B92 annual budget discounts that rise to 42% above EUR700,000, while sponsorships are negotiated separately.
The digital rate card sells a different set of units: impressions at stated cost-per-thousand rates, fixed-rotation takeovers, mobile and app placements, video prerolls and midrolls, and paid publication of promotional articles. The document lists promotional-article prices of EUR1,000 on the home page, EUR550 in business or sport and EUR450 in other categories. Its audience page claims 5.6 million average monthly users, 30 million sessions and 200 million monthly page openings across web, mobile and app; it also says 66% of readers are in Serbia and 70% of acquisition comes from direct visits and organic search. These are seller claims rather than audited audience statements, but they show exactly what the company asks buyers to value.
This product architecture has three consequences for the network thesis.
First, every extra unit of useful availability can be monetised indirectly. A portal outage during a major news or sporting event destroys high-intent inventory that cannot be stored and sold later. A stream failure can reduce campaign delivery. Delayed editorial publication can surrender search and social traffic to rivals. Stable, well-routed services therefore protect perishable revenue.
Second, the bandwidth itself is an input. B92 cannot simply charge a premium because it owns addresses or speaks BGP. Advertisers pay for audience, targeting, placement, context and measured delivery. Viewers choose content and convenience. Distribution partners care about rights, audience demand, technical quality and commercial terms. Network engineering improves those outputs, but it is not the output.
Third, the digital business is exposed to platforms even when B92 controls its own routes. Organic search, social networks, app stores, third-party video platforms, content-delivery networks and measurement vendors sit between publisher and audience. The media kit's assertion that 70% of acquisition is direct or organic is encouraging because direct traffic signals brand habit, but the same sentence also exposes search dependence. Owning an ASN does not control a search ranking, a mobile notification channel or a social referral feed.
Unit economics: two ledgers, one company
The most useful economic exercise is to keep a media ledger and a network ledger separate before asking how they interact.
On the media ledger, revenue is driven by audience volume, rating points, impressions, sponsorship, promotional content and distribution arrangements. The variable cost of serving one more cached page can be low, but creating and acquiring content, maintaining a newsroom and studio operation, selling campaigns and meeting broadcast obligations are not. Inventory is perishable. An unsold prime-time audience or an unavailable page view cannot be carried into the next month. Yield management, audience quality and advertiser concentration matter as much as raw reach.
On the network ledger, revenue would normally be the monthly recurring charge for access, transit resale, hosting, address rental, managed security or colocation. Gross margin depends on traffic commitments, access-loop costs, support intensity, equipment utilisation and customer churn. Capital must be committed before all capacity is sold. A small regional provider can look attractive while ports and support staff have spare capacity, then suffer a step-change in cost when it needs another upstream commit, router, facility, fibre path or support shift.
B92 publishes strong evidence for the first ledger and little for the second. Its public pages expose television and web prices. They do not expose a current internet-access price, symmetric-business package, service area, installation charge, uptime promise, address add-on, hosting plan or support desk for connectivity customers. Public routing observers show no downstream networks. This does not prove that B92 earns zero telecom revenue: private circuits, legacy accounts, intercompany charges or wholesale arrangements may never appear on a retail website. It does mean that a claim of material direct telecom income requires evidence that the public record does not currently supply.
A commercial Serbian company-information profile provides a useful, though secondary, financial series attributed to company filings. It reports total revenue of RSD1.600 billion in 2023, RSD2.241 billion in 2024 and RSD2.272 billion in 2025. Net profit moves from RSD4.4 million to RSD371.8 million and then to RSD217.2 million. It reports 2025 EBITDA of RSD332.4 million and 169 employees.
Those figures imply a 2025 EBITDA margin of about 14.6% and a net margin of 9.6%. Revenue rose only 1.4% in 2025, while reported expenses rose 10.9% and net profit fell 41.6%. The 2024 net margin of 16.6% was therefore not a stable run rate; 2023's margin was only 0.3%. Revenue per employee was roughly RSD13.45 million in 2025. Applying the same page's reported average gross monthly pay of RSD232,091 to 169 staff gives an annual wage proxy near RSD471 million, or about 20.7% of revenue. That is not a full payroll account and may mix average and point-in-time measures, but it puts labour in the right order of magnitude.
Using the National Bank of Serbia's 2025 average exchange-rate series, approximately RSD117.18 to the euro, reported revenue converts to about EUR19.4 million and EBITDA to about EUR2.84 million. The Reuters Institute's 2026 Serbia market review cites 2025 advertising spend of EUR296 million, of which television drew 45% and online 32%. B92's total reported revenue would be about 6.6% of that market-wide advertising figure, but the comparison is only a scale marker: B92 revenue may include content distribution and other items, while market spend and publisher-recognised revenue are not identical measures.
The conclusion is still useful. This is a medium-sized media balance sheet on which a modest network operation can be strategic without needing to become a separate growth engine. A few hundred thousand euros of well-targeted infrastructure, staff time and transit can be rational if it protects a much larger revenue base. Conversely, an attempt to build a consumer ISP would require an access network, acquisition budget, billing, installation, customer care and regulatory obligations that are not visible in the current cost structure or product surface.
What AS9081 proves
Public routing data establishes more than a dormant registration. BGP.Tools describes AS9081 as active, registered in April 2002 to rs.opennet, and originating three IPv4 routes. It sees two upstreams: AS25467, Akton d.o.o., and AS8400, Telekom Srbija. Hurricane Electric's BGP view independently reports the same two peers, three originated IPv4 routes, no IPv6 route, 4,096 unique IPv4 addresses and valid route-origin authorisation for all three announcements. IPinfo's AS9081 profile also sees active addresses and traceroutes entering the network through the same providers.
The route count needs care because the announcements overlap. The visible prefixes are 212.102.128.0/20, 212.102.128.0/21 and 212.102.132.0/24. The /21 and /24 sit inside the /20, so adding 4,096, 2,048 and 256 would double-count addresses. The unique originated footprint is the /20's 4,096 addresses. More-specific announcements can be used for traffic engineering, failover or policy distinctions; their presence does not create extra address inventory.
The registry allocation is larger. A current allocation compilation based on RIPE statistics lists 212.102.128.0/19, or 8,192 IPv4 addresses, under rs.opennet, with an allocation date in 1998. It also lists 2a0b:3b80::/29 from 2017. The difference between an 8,192-address allocation and a 4,096-address originated footprint may reflect held space, internal assignments, a different origin arrangement, reserved capacity or stale records in one of the observation layers. It should not be converted into a utilisation rate without address-level evidence. What can be said is narrower: B92 controls more registered IPv4 space than AS9081 visibly originates, and the reviewed collectors did not see AS9081 originate IPv6.
Three other details improve confidence in the operating surface. The routes match trusted internet-routing registry information in BGP.Tools. All visible origins are RPKI-valid rather than invalid, reducing the chance that networks enforcing route-origin validation will reject them. And current traceroute observations reach responsive B92 addresses in Zemun. None of these proves uptime or capacity, but together they are inconsistent with the idea of a purely ceremonial ASN.
The narrowness is equally important. Two observed upstreams are a compact transit set. No public Internet exchange entry for AS9081 was found in PeeringDB during this research, and neither major routing observer showed IPv6 peers or downstream autonomous systems. IPinfo found only a small number of hosted domains on sampled addresses, a method that misses services behind content-delivery networks, shared infrastructure or non-web uses. The right evidence grade is therefore strong for current routing existence and weak for external customer scale.
Where the network can create internal value
The first value pool is service continuity. Provider-independent routing lets B92 announce the same address space through more than one upstream. If one transit path fails cleanly and the other path, edge equipment and physical route remain healthy, traffic can converge to the surviving provider without changing public addresses. This is materially different from having two ordinary business-internet lines where services must be renumbered or hidden behind a failover appliance. It does not guarantee continuity: both links can share ducts, facilities, power or upstream dependencies. But it creates the possibility of policy-based resilience.
The second pool is migration control. Stable public addresses can anchor mail, authoritative DNS, remote access, security allowlists, contribution feeds and partner integrations. Renumbering those services is often less expensive in hardware than in coordination. Partners must update rules, certificates and documentation; mistakes emerge over weeks; old dependencies linger. Address autonomy turns a carrier change from an identity change into a route change. That lowers switching cost and makes a threat to move providers more credible.
The third pool is distribution performance. A broadcaster and portal move several distinct traffic classes: newsroom access, publishing, ad technology, live and on-demand media, corporate services, partner transfer, monitoring and audience traffic. More-specific announcements can steer selected address blocks differently, and two providers can offer different paths into Serbian and international networks. B92 does not need to become a broad peering network to gain from route diversity. It can use routing policy to isolate important services, protect management traffic or test provider performance.
The fourth pool is incident control. When an enterprise owns neither its address space nor its routing policy, remediation depends on the carrier's queue and interpretation. With its own ASN, B92 can withdraw a route, change preference, apply filters and coordinate directly with both providers. The trade is that it must maintain competent engineers, route filters, RPKI records, abuse contacts, credentials and change controls. Autonomy shifts risk; it does not erase it.
The fifth pool is strategic option value. IPv4 space is scarce and transferable under RIPE policy, subject to registry rules and transaction conditions. RIPE NCC explains that it can authorise transfers and that partial IPv4 blocks can be transferred, with a 24-month holding restriction after receipt or restructuring. A commercial broker said early-to-mid-2026 large /16 blocks traded around USD10-15 per address and smaller /20-/24 blocks at more than twice that rate. Applying any broker quote mechanically to B92 would be a mistake: block quality, routing history, legal title, demand, taxes and the cost of disrupting operations all matter. Still, thousands of clean IPv4 addresses carry an opportunity cost. Keeping them for resilience is a capital-allocation decision even if historical accounting shows little or no book value.
Why direct telecom revenue remains unproven
There are traces of an ISP past. RATEL's 2007 provider list included a B92 ISP branch, and the regulator's current searchable register still returns a historical record for B92 AD BEOGRAD OGRANAK B92.ISP at an older address. The current register explicitly warns that black and grey text distinguish valid and non-valid permissions, but the text extraction did not expose the row's colour. More importantly, the record retains the old AD legal form and address while current media and company records use the DOO identity and Ikarbus address. It is evidence of historical participation, not sufficient evidence of a current retail authorisation or active service.
No current B92 connectivity catalogue was found. The public portal routes potential customers to advertising, applications, content and television, not broadband. Searchable routing views show no downstream network. The observed address footprint hosts some reachable services, but addresses used for an enterprise's own operations can look identical to hosting inventory from outside. Even the RIPE member page only proves that B92 is an LIR; banks, universities, retailers and public bodies also operate LIRs without selling internet access.
A defensible direct-revenue case would need at least one of the following: identifiable connectivity customers, a current rate card, active access or hosting contracts, telecom revenue in the accounts, downstream routes, customer address assignments, a live support and installation process, or wholesale interconnection receipts. None was located. That absence should shape the valuation. Revenue should be set to zero in a base-case network model, with any undisclosed telecom income treated as upside rather than assumed cash flow.
The same discipline applies to address leasing. Unoriginated registered space can tempt an observer to infer rental income. Public routing data does not show B92-originated IPv6 and shows only half of the /19 as unique IPv4 space originated by AS9081, but it does not reveal whether the other half is unused, assigned to customers, originated elsewhere, reserved, filtered or covered by a stale registration. Leasing claims need route, contract and registry evidence.
Upstream dependence: diversity is not independence
AS9081's two upstreams are economically meaningful because they reduce single-supplier exposure. Telekom Srbija is the large national operator; Akton is another regional connectivity provider. Having both creates a procurement benchmark and a technical alternative. It can improve repair leverage, let B92 compare latency and loss, and make maintenance less disruptive if links and edge equipment are truly diverse.
But two names in a routing table do not prove two independent failure domains. Both services could enter the building through the same duct, terminate in the same room, depend on the same power or cross a shared metro facility. One provider may itself buy capacity from or peer heavily with the other. Traffic can also become asymmetric: outbound packets may prefer one carrier while replies enter through another, complicating fault diagnosis and security policy. Resilience depends on physical maps, contract terms, router design and tested failover, none of which is public.
The lack of visible IPv6 origination is another dependency signal. B92 has a large IPv6 allocation in registry statistics, yet the major public collectors reviewed here saw zero IPv6 routes from AS9081. An IPv4-only edge remains reachable to most users today, especially through third-party delivery, but it leaves the company dependent on translation and dual-stack services elsewhere as internet traffic evolves. The issue is not address scarcity: a /29 is vast. It is execution, equipment, security policy, content-delivery integration and operational confidence.
No public exchange presence was found. That means there is no evidence that B92 directly exchanges traffic with Serbian access networks, global content networks or route servers at an exchange. It may use private arrangements that are not listed, and its upstreams can deliver perfectly adequate reach. Economically, however, transit-only dependence limits the levers B92 can pull. Direct peering can lower some unit costs or improve paths when traffic volume justifies ports, transport, equipment and staff. At B92's apparent scale, buying that benefit through Telekom Srbija, Akton or a content-delivery provider may be cheaper than building a broader interconnection footprint.
The 2026 RIPE NCC fee illustrates what is and is not expensive. RIPE charges EUR1,800 per LIR account annually, plus applicable resource charges such as EUR50 per ASN assignment. That is trivial against B92's reported revenue. The expensive components are the ones the registry cannot supply: diverse circuits, suitable edge routers, security appliances, monitoring, facilities, engineering availability and the opportunity cost of management attention. A decision to retain AS9081 should not turn on membership dues. It should turn on whether the organisation uses the capability well.
Costs and capital: the hidden denominator
The public financial series does not separate newsroom, television, digital, rights, transmission and network costs. It nevertheless provides boundaries. Reported 2025 expenses of RSD2.012 billion consumed 88.6% of revenue, and profit fell sharply even though revenue edged higher. That sensitivity matters because network autonomy is largely a fixed-cost commitment. Staff and equipment must be available before an incident; the company cannot hire a routing specialist after a path fails and expect an immediate recovery.
Labour is likely the largest recurring network cost at this scale, even if engineers are shared with digital operations. A resilient rota needs more than one person who understands BGP, routing security, DNS, firewalls and provider escalation. If expertise is concentrated in one employee or contractor, the ASN can create the appearance of control while leaving a key-person dependency. Training, documentation and tested access matter more than the visible count of prefixes.
Transit and physical diversity come next. Commit charges can be low for a media company relative to video traffic if public delivery is outsourced to a content-delivery network, but true path diversity may require separately routed fibre, entrance facilities and power. Paying two suppliers for links that share civil infrastructure buys invoice diversity, not resilience. B92's Zemun location, studio facilities and transmission arrangements determine how expensive genuine separation would be.
Capital equipment is lumpy. Edge routers must hold the required routes and policies with headroom; firewalls and denial-of-service protection must handle attack traffic; optics, spare power supplies and out-of-band management need lifecycle replacement. The cost is not only purchase price. Maintenance, licensing, software updates, spares and planned migration absorb staff time. If B92's audience traffic is largely delivered by third parties, a smaller edge may suffice, but that makes the autonomy proposition more about corporate and origin-service continuity than bulk distribution.
Content and broadcast obligations compete for the same capital. Programme rights, studio equipment, playout, terrestrial transmission, digital product development and talent all can have a clearer link to audience than another router or route. The network case must therefore be made as loss avoidance and option value. It should answer a counterfactual: how much revenue, trust and operating time would B92 lose under a carrier-controlled design, and how often would that loss occur?
Liquidity deserves attention. The commercial profile reports current assets of RSD1.389 billion against current liabilities of RSD1.530 billion in 2025, a current ratio of about 0.91. This improved dramatically from the same source's 2023 figures but remained below one. That does not prove distress; media companies can carry large receivables and supplier balances with predictable collections. It does mean speculative network expansion should face a high hurdle. Resilience spending that protects core cash generation is easier to defend than access-network expansion that requires working capital before customer scale appears.
Customer and market dependence
B92 has at least four customer groups, each with a different dependency pattern.
Advertisers and media buyers purchase measurable attention. The combined Prva-B92 discount schedule indicates that large annual budgets matter and that group negotiation can be important. Volume discounts reaching 42% show how far realised pricing can move from headline CPP. A small number of major accounts can therefore influence yield, seasonality and receivables. The network protects delivery, but it cannot cure advertiser concentration or weak audience demand.
Distribution platforms carry the television channel. Terrestrial MUX 1 gives national free-to-air reach, while cable, satellite and IPTV extend convenience and measurement. A 2022 carriage dispute with SBB exposed the commercial tension. SBB publicly alleged that the owner of Prva and B92 sought a 1,000% fee increase; B92 and the other channels disputed the framing, and B92 later announced an agreement under which the channels remained on SBB. The competing claims should not be treated as verified price data. The episode still proves that distribution contracts can become a bargaining contest in which audience reach and platform access are mutual dependencies.
Viewers and readers pay mainly with attention rather than subscriptions. Reuters Institute's 2026 survey reports B92 TV at 17% weekly offline news use and B92 online at 12% among respondents. The same country review says television captured 45% of Serbian ad spend and online 32%, with online growing much faster. B92 straddles both pools, which is an advantage if it can sell campaigns across screens. It is also a source of execution risk: television and digital require different products, measurement, content cadence and technology.
Finally, content partners, rights holders and technology vendors supply inputs. B92's digital rate card says international content comes from sources including BBC and Deutsche Welle. Video, measurement, consent, advertising and app delivery depend on vendors that can change terms or suffer incidents. An owned ASN controls only a portion of this chain. A robust continuity plan must identify which third-party failure can still take the service down even when both transit paths are healthy.
SME service continuity without an SME broadband product
The topic of small-business continuity is relevant, but not because B92 has proved itself an SME ISP. It is relevant because many of the buyers and partners around a national media platform are smaller businesses whose campaign timing, content distribution and public visibility depend on B92 being available.
A retailer buying a launch-day campaign, a travel company promoting seasonal inventory or a local business placing sponsored content cannot recover the same value from impressions delivered after the commercial moment passes. For such customers, B92's network resilience reduces a form of supplier continuity risk. The service being protected is advertising and media delivery, not necessarily internet access.
If legacy connectivity or hosting customers remain, their economics would be different. They would expect fault response, address stability, security and perhaps local-language support. A small customer may value a known Belgrade team more than the lowest global cloud price, especially when an outage stops point-of-sale, mail or a public website. But that proposition carries support costs that do not scale like media impressions. Each unusual router, mail issue or on-site fault can consume expert hours. Without current tariffs, customer counts and service descriptions, it would be irresponsible to infer that B92 still earns that local-support premium.
There is a useful strategic boundary here. B92 can use its network competence to improve continuity for its media customers without promising to operate their networks. Managed connectivity transfers customer outage risk onto the provider; resilient media delivery reduces risk within B92's own contracted service. The second is closer to the evidence and demands less organisational sprawl.
Competition and substitutes
In media, B92 competes for time and advertising budget against public broadcaster RTS, national commercial stations Pink, Happy and Prva, cable news channels including N1 and Nova, digital publishers such as Blic, Telegraf, Kurir, Mondo and Danas, and social or video platforms. The 2026 Digital News Report places B92 behind several of these brands in weekly news use. Prva is both a sister sales partner and a competitor for viewing minutes. Group bundling can retain advertiser budgets, but it does not remove rivalry between channels for programming and audience.
The substitute for B92's autonomous network is not another media company. It is a managed connectivity architecture. B92 could buy redundant business circuits, host public services with a Serbian data-centre provider or global cloud, put audience delivery behind a content-delivery network, and let a carrier or managed-service provider own routing. Such an arrangement converts engineering capital and payroll into vendor fees. It can provide better attack absorption and geographic reach than a compact in-house edge.
Managed service is most attractive when B92's own traffic is modest, applications are already cloud-based, staff depth is thin and provider contracts include meaningful remedies. It is less attractive when stable addresses are embedded in many partner rules, latency to local services matters, migration would be disruptive, or supplier concentration creates negotiating risk. The ideal answer may be hybrid: retain the ASN and address space for identity, critical origins and optionality while outsourcing bulk delivery and some security to specialists.
Telekom Srbija and Akton are suppliers as well as substitutes for some in-house capability. That dual role creates discipline. If they can deliver diverse, well-managed paths at a cost below B92's internal alternative, purchasing makes sense. If either supplier uses address dependence or migration friction to weaken terms, B92's own resources improve its hand. The value of autonomy can therefore appear in a lower carrier bill rather than in a network revenue line.
Risk transfer has a limit
Owning the route changes who bears which risk. Carrier-managed addressing transfers technical responsibility outward but concentrates bargaining power and recovery timing with the provider. B92-owned routing transfers control inward but leaves B92 liable for configuration, security and staffing. Neither choice eliminates shared physical infrastructure, upstream congestion, denial-of-service attacks, software defects or human error.
Cyber risk is amplified by the company's public role. A media outlet is a visible target, and an address range with long-lived services accumulates reputation and scanning attention. RPKI-valid origins are a positive control against some route leaks and hijacks, but they do not stop application attacks, credential theft, malicious traffic from a valid origin or accidental route withdrawal. Abuse contacts and registry credentials must remain current through ownership and staffing changes.
Regulatory risk spans two sectors. The television service depends on media approvals and content obligations. Historical ISP records sit under the electronic-communications regulator. If B92 were to revive retail access, it would need to confirm current authorisations, consumer obligations, security requirements and complaint handling rather than relying on a legacy registration. Media and telecom permissions are not interchangeable.
Commercial trust is another risk. The Reuters Institute describes Serbia's media market as highly polarised and notes that mainstream commercial television is closely associated with political elites. B92's historical brand, current ownership structure and editorial positioning can affect who watches, who advertises and how distributors negotiate. Network resilience preserves reach, but it cannot preserve trust if audiences leave.
Financial risk is visible in margin volatility. A business that moved from near-breakeven in 2023 to a strong 2024 and a lower 2025 profit should avoid treating one good year as permanent capacity to fund unrelated expansion. The safer rule is to link network spending to measurable avoided downtime, supplier savings, security reduction or address value. A retail venture should stand on its own customer economics.
Unofficial market signals: useful, noisy and sometimes contradictory
Market signals outside contracts and regulators can test the story, but they should not be promoted into facts.
First, the company's own digital media kit claims very large reach: 5.6 million average monthly users and 200 million page openings across platforms. Serbia's population is about 6.7 million in the Reuters Institute review, so the user figure almost certainly includes repeat users across measurement boundaries, people outside Serbia or a methodology that does not equal unique residents. The document itself says 66% of readers are in Serbia. The numbers may be entirely compatible, but they need an audience-measurement definition before they can support revenue forecasts.
Second, third-party web estimators show that b92.net remains a prominent site, while the Reuters survey gives it meaningful but not leading news reach. This combination is plausible because B92 covers sport, lifestyle, technology, travel and entertainment as well as hard news. It also warns against using a news-use percentage as a total-site traffic measure.
Third, public forum discussions in Serbia contain persistent criticism of B92's editorial direction, click-driven presentation and repeated television content. Other comments still describe using the portal for sport, technology or specific programmes. These posts are self-selected, frequently political and impossible to turn into a representative satisfaction score. Their economic relevance is narrower: brand repositioning may have changed the composition and loyalty of the audience, and management should compare claimed direct traffic with cohort retention rather than assuming historical reputation remains intact.
Fourth, the 2022 SBB argument showed that distribution value is contested even for free-to-air channels. SBB's allegation of a tenfold fee demand and B92's claim that the channels were important to the platform represented negotiating positions. The eventual agreement shows mutual dependence, not which side's valuation was correct. It also illustrates why B92 may value independent internet distribution even though an ASN cannot replace national terrestrial or cable reach.
Finally, AS9081's footprint looks more like an enterprise or content network than a retail access provider. Four thousand originated IPv4 addresses are meaningful, but two observed peers, no downstream autonomous systems, no visible IPv6 and no public exchange profile are modest by carrier standards. This is an inference from routing shape, not proof of traffic or customer type.
Facts that would change the judgment
The conclusion would move toward a genuine regional-ISP thesis if audited or contract-level evidence showed recurring third-party connectivity revenue, active subscriber counts, access coverage, hosting inventory, customer address assignments, material downstream traffic or a staffed telecom support operation. Even a small but profitable business-services book could justify keeping engineering capability beyond media needs, especially if it shared fixed costs with B92's digital estate.
The conclusion would move more strongly toward internal insurance if traffic data showed that most bytes and addresses support B92, Prva or other group services; if transit costs were small against protected advertising revenue; if tested failover materially reduced downtime; or if address portability regularly improved carrier bids. In that case, segment revenue could remain zero while economic return stayed positive.
The judgment would deteriorate if the two upstreams share a physical path, failover is untested, route and registry access depends on one person, the unoriginated /19 space is encumbered, address reputation is poor, or critical services have already moved to providers that make AS9081 redundant. A network can persist on the public internet long after its business rationale has weakened.
Several disclosures would therefore be decisive: revenue by television, digital, distribution and telecom service; annual transit and facility cost; network and broadcast capital expenditure; address utilisation; monthly traffic by service; content-delivery spend; outage minutes and revenue impact; upstream service commitments; physical path maps; IPv6 migration milestones; and the proportion of critical services that can actually fail over. These are not exotic metrics. They are the bridge between a technically real network and an economically defensible one.
Verdict: preserve the option, resist the category error
B92 should not be valued as a regional ISP on the evidence available. Its current paid surface is television and digital media, its regulatory identity is that of a national broadcaster and publisher, and public routing data does not reveal a customer cone. Direct telecom revenue is possible but unverified.
Nor should AS9081 be dismissed as a historical curiosity. It is active, dual-homed, RPKI-valid and backed by scarce address resources. The company has a specific historical reason to understand that control of distribution can matter when ordinary channels fail. Today the threat is more likely to be a carrier incident, cyberattack, platform dependency, contractual dispute or migration problem than a 1990s communications blockade. The economic principle is the same: reach is part of the product.
The most defensible strategy is therefore to preserve network autonomy where it protects high-value media operations, measure that protection rigorously and outsource commodity scale where external providers are better. Stable addresses, two independently engineered paths, tested failover, current routing security and a credible IPv6 plan can create returns through avoided loss and lower switching costs. Expanding into retail access without proven demand would add installation, support, billing, regulation and capital intensity to a business already exposed to volatile media margins.
B92's network is valuable because it gives the company choices. It becomes a telecom business only when customers pay for those choices directly.