The CAPEX Gap in SAF: What Your Board Needs to See

Section 01The Trillion-Euro Question Nobody Can Answer Fast Enough

In September 2025, Shell permanently cancelled its 820,000 tonne-per-year biofuels facility in Rotterdam - the project that was supposed to be one of Europe's largest SAF and renewable diesel refineries. The write-down: $600 million to $1 billion. The reason cited was a reassessment of competitiveness after what Shell described as weak market conditions for biofuels in Europe.

Shell's retreat was not an isolated event. BP paused two biofuel projects in Germany and the United States around the same period. UBS analyst Joshua Stone noted that the delays highlighted that the advanced biofuels market was proving far more difficult than anticipated. These were not speculative bets by newcomers - these were flagship projects by two of the world's largest energy companies, backed by teams of experienced engineers and analysts.

And yet, on the demand side, the trajectory has never been clearer. IATA projects global SAF production will reach 2.4 million metric tons in 2026, which still represents only 0.8% of total jet fuel consumption. The EU's ReFuelEU Aviation mandate requires 2% SAF blending at airports from 2025, rising to 6% by 2030 and 70% by 2050. Strategy& (PwC) estimates that reaching net zero in aviation by 2050 would require approximately 325 million tons of SAF annually - demanding roughly €1,000 billion in cumulative refinery CAPEX just to build the production infrastructure.

Sources: Argus Media, Sep 2025; IATA Global Media Day, Dec 2025; Strategy& SAF Study.

The paradox is clear: the demand signal has never been stronger, yet billion-dollar CAPEX decisions are failing at an alarming rate. The question every board should be asking is not whether to invest in SAF - the mandates answer that. The question is: which pathway, at what scale, in what geography, under which policy scenario?

And that question cannot be answered with spreadsheets.

Section 02Why Board-Level SAF Decisions Are Breaking Down

A typical SAF CAPEX evaluation involves comparing multiple production pathways - HEFA (Hydroprocessed Esters and Fatty Acids), Fischer-Tropsch synthesis, Alcohol-to-Jet, Power-to-Liquid (e-SAF) - each with radically different feedstock requirements, conversion efficiencies, carbon intensity profiles, Technology Readiness Levels (TRL), and policy eligibility across jurisdictions. A single project evaluation at investment-grade quality has historically taken 4 to 6 weeks of analyst work.

During those 4 to 6 weeks, an analyst team manually extracts CAPEX, OPEX, yield rates, and emission factors from hundreds of pages of technical PDFs and vendor specifications, cross-references web sources and government datasets, converts units (barrels to liters, BTU to megajoules), draws process flow diagrams by hand, and assembles scenario comparisons in Excel with no systematic completeness validation.

The SkyNRG/ICF 2025 SAF Market Outlook highlighted that 82% of current SAF capacity relies on HEFA technology, which faces hard feedstock ceiling constraints. The 26 Mt gap between projected 2035 demand of 40 Mt and expected 2030 capacity of 18 Mt means most of the needed capacity hasn't even reached Final Investment Decision. Every month of analytical delay is a month of lost construction lead time.

Source: SkyNRG & ICF SAF Market Outlook 2025.

Section 03The Pathway Comparison Problem: Why Spreadsheets Fail at Scale

Consider what a genuine board-ready SAF CAPEX comparison actually requires. An energy developer evaluating a greenfield SAF facility must simultaneously assess HEFA (TRL 8-9, proven, but constrained by used cooking oil and waste fat availability), Fischer-Tropsch (TRL 6-7, feedstock-flexible via gasification, but higher CAPEX), Alcohol-to-Jet (TRL 5-7, can use cellulosic ethanol, but lower volumetric yields), and Power-to-Liquid or e-SAF (TRL 4-5, unlimited feedstock in theory via green hydrogen and captured CO₂, but currently the most expensive pathway).

Each pathway must then be modeled across multiple scenarios - baseline, optimistic, and pessimistic policy assumptions - in every target geography. The EU's ReFuelEU mandates differ from the UK's mandate (which nearly doubles to 3.6% in 2026), which differs from Singapore's levy-funded procurement scheme, which differs from the U.S. Inflation Reduction Act's 45Z Clean Fuel Production Tax Credit. CAPEX per installed liter, OPEX per liter of output, feedstock cost volatility, carbon intensity per MJ, and offtake pricing must all be compared with source attribution at every data point.

In practice, this means a properly scoped evaluation involves dozens of parameters across at least four pathways, three to five geographies, and three policy scenarios. That is a minimum of 47 discrete analytical scenarios - each requiring defensible, source-attributed data extraction from technical literature, academic papers, and policy documents.

No analyst team in the world can produce 47 investment-grade scenarios manually in time for a board cycle.

The real-world complexity makes the above table look deceptively simple. Each pathway intersects with feedstock supply chains, regional grid carbon intensities (which affect the lifecycle analysis of e-SAF), trade tariffs (the EU imposed anti-dumping duties on Chinese biodiesel and HVO in 2025), and the interplay between HVO and SAF production at shared facilities - where operators must dynamically allocate refinery output between the two based on shifting price signals.

Sources: Argus Media SAF Outlook, Jan 2026; Fortune Business Insights SAF Market Report, 2026.

Section 04What AI-Powered Techno-Economic Analysis Actually Changes

The gap between what boards need and what analysis teams can deliver is not a talent problem - it is a tooling problem. The analysts understand the domain. What they lack is the ability to synthesize hundreds of data sources across dozens of scenarios at the speed the market demands.

Multi-agent AI orchestration systems - where specialized agents handle discrete analytical tasks in parallel - are demonstrating a fundamentally different operating model for energy strategy. Rather than a single analyst working linearly through PDF extraction, unit conversion, cross-referencing, and scenario assembly, a multi-agent system decomposes the problem into parallel workflows.

In production deployments, this architecture has demonstrated measurable results: evaluation cycle time compressed from 4-6 weeks to 2-3 days (a 15x acceleration), data extraction completeness improved from approximately 60% (manual) to 95% (agent-based), analysis variance between junior and senior analysts reduced by 70%, and early identification of technology readiness gaps improved by 40%.

Critically, every extracted data point retains full metadata: the source document, the specific page, the technology basis, the geographic context, the year basis, the scenario assumption, and the uncertainty range. When a board member asks "where did this CAPEX figure come from?" - the system can trace it to a specific sentence in a specific document. That kind of provenance is what transforms an internal analysis into an investment-grade deliverable.

Section 05What This Means for Your Next Board Presentation

The SAF CAPEX gap is not going to close itself. The mandates are set. The demand trajectory is locked. The question is which companies will move capital at the speed the market requires - and which will be writing their own Shell-sized post-mortems.

The boards that get this right will be the ones that can simultaneously evaluate multiple production pathways with investment-grade data extraction, stress-test CAPEX assumptions across policy scenarios in every target geography in near-real-time, trace every number in an investment memo back to its primary source, and generate updated analyses within days as policy or market conditions shift - not months.

The infrastructure for this capability exists today. Multi-agent AI orchestration systems purpose-built for energy techno-economic analysis can be deployed within enterprise VPC environments, using swappable LLM providers and operating entirely on private data - meaning proprietary analysis stays proprietary. No vendor lock-in, no data exposure, no SaaS dependency. The institution's analytical IP compounds over time as every document analyzed becomes permanently searchable institutional intelligence.

The trillion-euro SAF buildout will be one of the defining capital allocation decisions of the next two decades. The companies that close the gap between the speed of analysis and the speed of the market will build the facilities. Everyone else will be reading about it in the write-down disclosures.