Emtek's 30 Percent and Indonesia's Ad Curve

Who eats the surplus, and how long does it last? Every generative-AI announcement out of Indonesian media this year has begged that question and then walked past it. A production cost falls. An animation cycle compresses. The vendor's page names the tools. The broadcaster's page names the shows. Neither page names the balance sheet, the ad-spot rate, or the labour line the surplus is supposed to flow into. Something is moving. What is moving is less obvious than the numbers already public.

Reading the Emtek release

Google Cloud's release on Emtek's 'Studio of the Future' launch, dated 23 April 2026, reads as a product announcement. It is also, once you overlay the year's ad-market math, a capex thesis. Emtek Group, Indonesia's largest listed media conglomerate, brought its streaming platform Vidio and its production houses onto the Gemini Enterprise Agent Platform, using Veo, Imagen and Gemini for a workflow it now calls VidioGen. A separate tool, Vivi, converts long-form footage into promotional cuts. The pilot ran on New Keluarga Somat, a Ramadan animated series aired on Mentari TV and Vidio. Emtek's own numbers, filtered through the vendor's press desk: a 30 percent reduction in redevelopment time and cost, audience-share growth of 74 percent in prime time and 90 percent during Sahur slots.

Take those numbers with the pinch of salt any vendor-attributed figure earns. The counterfactual for the audience-share growth is a prior Ramadan on a different day-part calendar, against a different competitive schedule. The cost figure is a vendor's chosen slice of a production budget the outside reader cannot audit. Neither figure is fraudulent. Neither is settled.

By the time of Marketech APAC's report on the partnership's expansion, dated 30 April 2026, Emtek had rolled the platform into an internal AI Centre of Excellence covering outpainting, inpainting, scene extension, transcription, subtitling and dubbing. The vendor was named. The internal team's size was not. Neither was the licensing bill.

Reading the ad curve under the announcement

Here is the macro line the vendor omitted and every Indonesian media CFO already has open in a tab. PwC's Global Entertainment and Media Outlook 2026 to 2030 for Indonesia puts the country's total entertainment and media market at US$31.7 billion in 2025, growing to US$42.9 billion by 2030 at a compound 6.2 percent per annum. The internal split matters more than the headline. Internet advertising, on the same schedule, moves from US$10.7 billion to US$16.8 billion, at 9.4 percent CAGR. Video streaming subscriptions, by contrast, only reach US$1.7 billion by 2030. The ad line is the one that funds the transformation.

That is the number Emtek's April announcement was priced against. Not the streaming subscriber count. The ad curve. A sustained cost cut on production, held across a slate that runs against a 9.4 percent per-annum ad market, compounds into a present value that pays back the capex on VidioGen inside three or four years. If the cost cut is real. If it holds. Both conditionals matter.

Where the surplus goes

Set that against Emtek's own segment math. ContentAsia's summary of Surya Citra Media's most recent full quarterly breakout, reported before the AI programme was announced, put SCM's net revenue at Rp 1,737.9 billion in the quarter, down 1.1 percent year on year. Free-to-air was down 14 percent at SCTV and 5 percent at IVM against a strong 2024 election-driven comparison. Digital media, including Vidio, was up 15.8 percent to Rp 429.2 billion, with Vidio revenue up 30.6 percent standalone.

Those two segments tell one story between them. The linear channel is a mature cash generator whose ad line is contracting in real terms. The digital arm is growing faster than the linear arm can match, on a smaller base. VidioGen and the Google Cloud partnership are being asked to pay off in both directions: cheaper production on the linear side, faster inventory turnover on the digital side, and enough cost-per-episode reduction across both to convert the market's structural shift into an operating-margin story instead of a revenue-mix warning.

That surplus does not accrue evenly. The producer who worked the New Keluarga Somat rebuild spends the vendor's stated 30 percent less time and cash on it than the season before demanded, on a figure attributable to a specific series and not to a slate-wide average. Her hourly rate has not moved. The slate she covers over the next fortnight is wider than it was two Ramadans ago. Her supervisor's 2027 headcount plan is smaller than the counterfactual would have suggested, by an amount that shows up in an operating expense line rather than in a press release. The Malaysian Press Institute's December 2024 report on Indonesian journalist job cuts, citing the Alliance of Independent Journalists, framed the pressure as a digital shift and a business-model change and stopped short of naming AI. Eighteen months on, the shift has a vendor logo underneath it. The line to watch in AJI's next census is not layoffs. It is scope creep on the surviving roles.

Where the model does not hold

The disconfirming reading is fresher than the deal itself. MIT NANDA's The GenAI Divide: State of AI in Business 2025, published in July 2025 and covered widely since, reported that 95 percent of enterprise generative-AI initiatives it studied showed no measurable P&L return. The methodology combined more than 300 disclosed initiatives, 52 organisational interviews and 153 senior leaders surveyed at conferences between January and June 2025. The 5 percent that did clear were disproportionately externally built, workflow-specific, tied to persistent context, and owned by an identifiable business unit. Gartner ran a parallel finding a year earlier, forecasting that at least 30 percent of enterprise generative-AI projects would be abandoned before end-2025 for reasons of cost, risk and unclear value.

Two things follow. Emtek's deployment sits inside the 5 percent shape those studies describe: an externally built platform, a workflow the operator can name, an internal owner in the AI Centre of Excellence, and a measurable output on a defined content slate. That is the profile of a project that will not turn up on next year's abandonment list. The same numbers are the reason a Jakarta CFO reading the April release should discount the peer pressure to sign the same deal. The base rate for enterprise generative-AI ROI is unforgiving. Emtek's numbers, if they hold, describe an outlier. A second-tier Indonesian broadcaster copying the deal a quarter later without the same content pipeline, editorial oversight or ad-market pricing power will land inside the 95 percent, not the 5.

The rest of the region's ad line

Widen the frame. Sensor Tower's State of Digital Advertising in Southeast Asia 2026 put Indonesia's Q1 quarterly digital ad spend at roughly US$156 million against a regional cadence in which mobile-first creative and local-language relevance are the two variables the platforms keep asking for. Regional platform ad revenue is now the middle of every super-app's pitch. Sea Group's Q1 and Q2 2026 releases, and Grab's guidance around advertising uplift alongside its transaction take, both describe an ad line scaling faster than the underlying commerce or ride line. Whoever owns the content that fills those inventories owns a slice of the upside.

Emtek's April signal, at the level of macro allocation, was a claim that the content-production side of the ad economy is the profitable long-term position, on the working assumption that inference costs stay roughly where they are and that the licensing terms with Google Cloud do not compress the operator's margin as the platform's own pricing power grows. Neither assumption is fixed. The inference-cost line has been falling for three years and could reprice in either direction. Vendor concentration is the risk a media conglomerate hedges by keeping the AI Centre's contract templates portable, and only some of them are.

What is left on the desk

Ramadan 2027 falls in the third week of February. In the Emtek Studios building in Jakarta, an editor is scrubbing frame by frame through an animated sequence she rebuilt in a shorter cycle than the last one asked of her, checking the outpainted background where a Gemini model filled a corner her animator ran out of time to draw. The scene will air on Vidio, on Mentari TV, and on three regional channels a rights partner has licensed the block from. On her supervisor's laptop, a rate card sits in one tab. Next year's slate is in another. The Google Cloud invoice is a third. The second draft of the AI Centre of Excellence's staffing plan is a fourth. The corridor lights are the ones that stay on all night during a rebuild week. Down the hall, a colleague is still there too.

Tarry Singh is the founder and CEO of Real AI, an enterprise AI advisory and deployment firm working with global enterprises on production agent systems, model risk, and AI sovereignty strategy. He also leads Earthscan, an Energy AI startup, and is a founding contributor to the EU-funded HCAIM and PANORAIMA programmes for responsible AI education across European universities. He writes at tarrysingh.com.