Skip to content

Compare moats

Up to three covered companies, band by band. Every call is a curated editorial judgment, never a disclosed figure — and every band carries its cited basis.

comparing Talen Energy×GitLab×GlobalFoundries× maximum of 3 — remove one to swap
Talen Energy TLN ai moat: latest change 2026-02-26 GitLab GTLB ai moat: latest change 2026-08-05 GlobalFoundries GFS ai moat: latest change 2026-02-27
Moat rating narrow

The FY2025 10-K grounds a real but bounded advantage. Susquehanna produced about 17 TWh in 2025 at an all-in cost of about $27/MWh, and up to 1,920 MW of it is contracted to AWS under a long-term, fixed-price PPA through 2042 'at anticipated premium prices', with deliveries expected to ramp to full volume no later than 2032. The moat is narrow rather than wide because, outside its contracts, the fleet sells into competitive wholesale markets, and the 10-K says that 'unlike regulated utilities, we are not assured of any rate of return on capital investments through a regulated rate structure'. It also concedes that competitors 'may have advantages over us through access to greater resources, newer generation facilities, lower costs, or more experience'. The advantage is also concentrated: Susquehanna has typically been about half of total annual generation.

source: sec.gov

narrow

The FY2026 10-K (year ended 2026-01-31) shows real expansion inside the installed base: 118% Dollar-Based Net Retention, Base Customers up to 10,682 from 9,893, customers above $100,000 of ARR up to 1,456 from 1,229, and $1.0 million ARR customers "to 155 from 123, an increase of 26%". But the filing concedes "limited barriers to entry" and names "Microsoft Corporation, which owns GitHub" as principal competitor. Retention decelerates (130% FY2024, 123% FY2025, 118% FY2026), and since DBNR is reported "on a threshold basis of 130%," the FY2024 figure is a cap and the slide is steeper than it looks. Narrow, not wide: expansion must be re-won against a bundled hyperscaler.

source: sec.gov

narrow

The FY2025 20-F gives a measured lock-in: approximately 63% of 2025 wafer shipment volume was single-sourced business - products GF says "can only be manufactured with our technology or cannot be manufactured elsewhere without significant customer redesigns" - and long-term agreements carried an aggregate remaining revenue commitment of approximately $11 billion at 31 December 2025, as "binding, multi-year, reciprocal minimum purchase and supply commitments". That advantage is bounded by admissions in the same filing: "some of our competitors may offer more advanced or differentiated technologies than we do and some have greater access to capital and substantially greater production capacity, research and development... including access to government subsidies"; China's mature-node "foundry capacity is expected to grow faster than expected demand at those nodes"; and GF "renegotiated a number of LTAs with certain of our customers, including in 2025, as a result of which some of our LTAs now have longer commitment periods... and some of our LTAs have lower pricing or volume commitments than originally negotiated". Gross margin of 24.9% in 2025 ($1,690m on $6,791m of net revenue) on essentially flat revenue is the economics of a position that is defended rather than uncontested.

source: sec.gov

Moat type cost scale

The 10-K says Talen's ability to compete 'is affected primarily by electricity prices, fuel prices, the relative cost of electric generation, and the reliability and availability of generation assets'. The edge it cites is low-cost baseload supply: Susquehanna, the seventh-largest U.S. nuclear plant, produced power at an all-in cost of about $27/MWh in 2025, under NRC licences that expire in 2042 and 2044. The 10-K says natural gas facilities 'often serve as the marginal, price-setting generating units' and that energy margin is influenced by the relationship between electricity and natural gas prices, so a low-cost nuclear unit earns the gap between market price and its own cost. The AWS adjacency is a contract rather than a fleet-wide switching cost, and network effects and IP do not apply.

source: sec.gov

switching costs

Stickiness comes from platform embedding, not IP. The 10-K rests differentiation on "our single platform with a unified data model," on "Consolidation of multiple tools into a single platform," and on deployment inside the customer's own perimeter for regulated buyers. Replacement means re-integrating a toolchain and re-certifying compliance. IP is the wrong label: only "16 issued patents and 25 pending patent applications," plus open-source licenses granting "broad permissions to use, copy, modify, and redistribute." The no-lock-in claim is the filing's own characterisation, footnoted to a Forrester study "commissioned by GitLab."

source: sec.gov

switching costs

The filing rules out scale as the source: GF concedes competitors with "substantially greater production capacity, research and development" and its four fabs held roughly 2.8 million wafers per annum of installed capacity at end-2025 against a $6.8bn revenue base. The mechanism it names instead is the cost of moving a qualified design - "[g]iven the time and costs associated with moving a single-sourced product to a competitor, clients are more likely to continue awarding us single-source contracts for such products" - and it is quantified at approximately 63% of 2025 wafer shipment volume. It is contractually reinforced by the ~$11bn of remaining LTA revenue commitment, which GF says continues to matter "for certain longer term, more durable end markets, such as automotive", where 21% of 2025 net revenue ($1,410m) now sits.

source: sec.gov

Leadership at parity

Talen calls itself 'a leading independent power producer', but its merchant output is undifferentiated. The 10-K says it sells into PJM capacity auctions and day-ahead and real-time markets against 'other non-utility generators, regulated utilities and their competitive subsidiaries, industrial companies, financial institutions, and other energy marketers', and that some rivals may have greater resources, newer plants or lower costs. Susquehanna's rank as the seventh-largest U.S. nuclear facility shows scale, not market leadership.

source: sec.gov

fast follower

The 10-K positions GitLab against a leader rather than as one: "Our principal competitor is Microsoft Corporation, which owns GitHub." Competition reads from the challenger side, differentiating "from GitHub through flexible deployment options that work within enterprise security and compliance requirements, LLM neutrality with self-hosted gateway support" and an open core model. On AI it follows: Duo Agent Platform went "Generally available in January 2026," the last month of the fiscal year. What keeps it a follower rather than behind is cadence, "a new version of our software every month for 172 months in a row," plus a regulated-deployment franchise a SaaS-first rival does not cover.

source: sec.gov

fast follower

The 20-F places GF inside a group of "five major foundries (including four scaled pure-play foundries) that accounted for the vast majority of worldwide foundry revenue in 2025", naming TSMC first among key competitors along with UMC and SMIC, and separately conceding competitors with "more advanced or differentiated technologies than we do" and "substantially greater production capacity, research and development, marketing and other resources". GF does not compete for the leading edge at all; its stated pillars are differentiated RF, FDX, silicon photonics, power and feature-rich CMOS plus geographic diversity. That is an established, scaled participant operating inside a pace set elsewhere, not one setting it, and not one shut out of the top tier.

source: sec.gov

Pricing power moderate

Pricing power is mixed and sits in the AWS contract. The AWS PPA is fixed-price 'at anticipated premium prices' through 2042, and the 10-K says such large-load contracts 'can lock in meaningful premiums'. The Brandon Shores and H.A. Wagner RMR agreements provide FERC-approved annual fixed-cost payments of $145 million and $35 million from June 1, 2025, each including a performance hold-back, and FERC's approval is under appeal by the Maryland Office of People's Counsel; they give revenue certainty rather than pricing power. The merchant remainder takes market prices. The 10-K says gas units 'often serve as the marginal, price-setting generating units', and the PJM auction imposed a capacity price collar with an approximate minimum and maximum of $175/MWd and $325/MWd for the 2026/2027 and 2027/2028 auctions. In the 2027/2028 auction Talen cleared 8,745 MW at $333.44/MWd.

source: sec.gov

moderate

FY2026 gross margin was 87% against 89%, with cost of revenue up $35.6 million "primarily due to an increase of $18.4 million in third party hosting costs for SaaS and cloud usage," and management expects SaaS and Duo mix to bring costs that "may adversely impact our gross margins." Strain is explicit: GitLab "may be required to reduce our prices," competitors "may offer their products and services at a lower price or for free," and it "implemented user limits on our free SaaS product." Power retained: Premium lists at $29 per user/month billed annually, and seats bundle $12 and $24 of GitLab Credits. But "A majority of our subscriptions are on a one-year period."

source: sec.gov

moderate

Gross margin was 24.9% in 2025 ($1,690m on $6,791m) versus 24.5% in 2024 ($1,651m on $6,750m) - held, not expanded, even though average shipment utilization across the global fabs rose from 77% to 86%, so most of the volume gain was absorbed rather than dropping through. The filing states outright that "the market prices for technology and services tend to fall over time, except in times of extreme supply shortage", and discloses LTAs renegotiated in 2025 to "lower pricing or volume commitments than originally negotiated". Against that, GF says wafer price "varies significantly across technology platforms" and that devices with "richer feature sets, higher performance, better yields and greater system-level integration... generally command higher wafer prices". Price-setting inside designed-in single-sourced sockets, price-taking at the competitive margin.

source: sec.gov

Summary

Talen's advantage comes from one contracted asset. It owns 90% of the 2.5 GW Susquehanna nuclear plant, which produced about 17 TWh in 2025 at about $27/MWh all-in, and it has agreed to supply the adjacent AWS data campus with up to 1,920 MW under a fixed-price PPA through 2042. The rest of its roughly 13.1 GW fleet at the 10-K date (a total that includes its 2.2 GW nuclear share) includes newly acquired H-class combined-cycle gas plants, intermediate and peaking units, the Brandon Shores and H.A. Wagner units kept running under FERC-approved RMR agreements, and about 800 MW of minority interests in three coal-fired plants. Apart from the RMR units, that fleet sells capacity and energy mainly into competitive PJM markets through capacity auctions, spot markets and futures contracts, and the 10-K notes that, 'unlike regulated utilities, we are not assured of any rate of return on capital investments through a regulated rate structure'. The 10-K's 'Talen flywheel' strategy is to repeat the AWS model with other large-load buyers. Until more of the fleet is contracted, the moat stays narrow and concentrated in Susquehanna.

GitLab sells a single-platform DevSecOps toolchain defended by workflow lock-in rather than intellectual property: a unified data model plus deployment flexibility, including single-tenant "Dedicated for Government with FedRAMP compliance," makes it hard to remove where data residency is contractual. The installed base expands: 118% Dollar-Based Net Retention, 10,682 Base Customers, 1,456 above $100,000 of ARR, 155 above $1.0 million (up 26%), over 70% of ARR from public sector and enterprise. The filing supplies the counterweight: "limited barriers to entry," Microsoft/GitHub as principal competitor, 16 issued patents. Retention (130 to 123 to 118, the 130 a cap), gross margin (89% to 87%) and a January 2026 AI launch point to compression at the commodity end. No source measures market share, so every share band is "unknown".

GlobalFoundries is a pure-play specialty foundry that deliberately does not contest the leading edge. Its FY2025 20-F frames the business around six differentiated technology lines - feature-rich CMOS, ultra-low-power CMOS (FinFET and FDX), silicon photonics, RF, power, and, new in 2025, processor IP via the MIPS acquisition - manufactured at four fabs in Malta NY, Burlington VT, Dresden and Singapore, which GF describes as making it "the only scaled foundry with a global manufacturing footprint spanning the United States, Europe, and Asia operating for more than a decade". The defence that shows up in the numbers is design lock-in rather than cost: approximately 63% of 2025 wafer shipment volume was single-sourced, and ~$11bn of long-term agreement revenue commitment remained at year-end. What limits it is visible in the same document - competitors with more capital, more capacity and, in some cases, more advanced technology; mature-node capacity in China expected to outgrow demand at 28nm and larger; TSMC, Samsung, Intel, Texas Instruments and UMC all building in the United States, Europe or Singapore, which erodes the geographic-diversity argument GF calls "one of our key advantages". Pricing reflects that balance: 2025 gross margin of 24.9% against 24.5% in 2024 even as shipment utilization rose from 77% to 86%, with some LTAs renegotiated to lower pricing. AI exposure is real but indirect - Communications, Infrastructure & Data Center was 11% of 2025 net revenue, and GF's AI content is optical interconnect, RF, sensing, power and edge processor IP rather than accelerator logic.

Chain position

Supplies nuclear power from Susquehanna to Amazon's adjacent AWS data campus, and sells capacity and energy into the PJM grid.

GitLab sits at the software-tooling layer, not compute or models. It consumes third-party LLM capacity — marketing "LLM neutrality and support for self-hosted AI gateways, including air-gapped environments" — so it captures no model-layer economics and bears inference as COGS, visible in the $18.4 million rise in hosting costs. Its distinctive position is the regulated perimeter.

Pure-play specialty foundry. Inputs are silicon and specialised SOI wafers bought under multiple long-term agreements with principal suppliers, "the largest of which is Soitec", plus externally sourced EDA tools and third-party IP. Output is finished semiconductor wafers, approximately 89% of 2025 net revenue, sold to fabless designers and IDMs and delivered through the GlobalSolutions ecosystem of IP providers, EDA vendors, OSAT partners and design-service firms; the remaining ~11% is photomasks, sourcing and post-fab services, and IP licence and royalty income. Manufacturing sits at four sites - Malta NY, Burlington VT, Dresden and Singapore - with roughly 239,000 square metres of clean room and about 2.8 million wafers per annum installed capacity at end-2025. Position in the AI chain is one step removed from accelerators: Communications, Infrastructure & Data Center was 11% of 2025 net revenue ($745m), with content in optical interconnect, RF front-ends, timing and power, alongside edge-AI processor IP from the MIPS acquisition.

Products (share / barrier)
  • Enterprise Agile Planning add-on Unknown · Low source: sec.gov
  • GitLab Dedicated (including Dedicated for Government) Unknown · Deep source: sec.gov
  • GitLab DevSecOps Platform (Free, Premium, and Ultimate tiers) Unknown · Moderate source: sec.gov
  • GitLab Duo Agent Platform (with GitLab Credits) Unknown · Low source: about.gitlab.com
  • Self-Managed GitLab (on-premises and hybrid cloud deployment) Unknown · Deep source: sec.gov
Long-horizon vote +0.06 at weight 0.20 · swarm neutral

Editorial prior, not backtested.

see exactly how it voted →

+0.06 at weight 0.20 · swarm neutral

Editorial prior, not backtested.

see exactly how it voted →

+0.06 at weight 0.20 · swarm neutral

Editorial prior, not backtested.

see exactly how it voted →