Memos from the field. Written by operators.
Our house view on where capital should move next — ecommerce operator discipline, telecom fragmentation, the AI data center buildout, and the chip supply constraints that sit underneath all of it.
Scaling an 8-figure DTC brand — the operator's playbook
What we look for in acquisition targets, and the five levers we pull post-close to compound returns.
Global ecommerce will cross $6.88 trillion in 2026 — 21.1% of total retail — but the capital opportunity isn't in the giants. It's in the thousands of 7- and 8-figure brands stuck below the threshold where strategic buyers engage. These brands have real unit economics, durable demand, and operator-starved teams. That's exactly the shape of target we invest in.
After we acquire, we pull five levers in a consistent order: (1) paid acquisition efficiency, usually underbuilt in founder-led brands; (2) conversion rate optimization across funnel and checkout; (3) logistics and fulfillment — margin hides here; (4) international expansion into LATAM and selected EU markets where our operating infrastructure already sits; and (5) call center and backend CX consolidation. Each lever compounds on the last.
Exit discipline is the final step. We don't hold forever. When the category multiple is right and the brand is operating at peak margin, we position the business for strategic sale, aggregator roll-up, or secondary transaction.
The telecom market is fragmenting — in our favor
eSIM, IoT, and network virtualization have opened hundreds of addressable niches. The $100B+ MVNO market is becoming institutional.
Fifteen years ago there were three US carriers, two spectrum auctions worth paying attention to, and one business model. Today the US MVNO market alone is approximately $46.8 billion in 2026 and the global MVNO market is estimated at roughly $79–103 billion depending on methodology. Full MVNOs alone represent ~53% of the market globally.
Three forces drove the fragmentation. First, eSIM — users can switch networks in software, which killed the physical-SIM lock-in that protected incumbent carriers. Second, network function virtualization and cloud-based core networks — a "Light MVNO" can now be launched in weeks rather than the 18–24 months it used to take. Third, IoT — industrial connectivity is booked on 10-year device life cycles, creating annuity revenue that wholesale-access math makes highly attractive.
For investors with operating depth in telecom, the result is a rare opening. Dozens of profitable niche operators exist at the low end. Strategic capital can consolidate, professionalize, and scale them. That's where we're deploying.
Why we invest in niche MVNO brands the carriers abandoned
Big carriers can't profitably serve narrow demographics. The operators who can are early, undercapitalized, and ready for a partner.
Incumbent carriers optimize for the largest middle segment of the wireless market. That leaves dozens of profitable niches — diaspora communities, international travelers, specific industrial fleets, gamers, seniors, family plans with very particular structures — on the table. The operators who built MVNO brands targeting those segments are often first-generation, bootstrapped, and under-resourced.
Our thesis: consolidate, professionalize back-office operations (billing, activation, CX), negotiate better wholesale terms through aggregated volume, layer in IoT connectivity products, and run disciplined paid acquisition. A good MVNO operator with real capital behind them is a different business than a good MVNO operator alone.
The math is unusual in a good way. MVNO unit economics improve with scale almost monotonically — wholesale costs drop with volume, CX costs drop with platform investment, and churn drops with brand equity. That's a compounding business.
AI compute is a power problem, not a chip problem
Why we're investing in land, substations, and grid-adjacent DC capacity ahead of the demand curve.
Every conversation about AI eventually becomes a conversation about GPUs. The real constraint, though, is upstream: electrical power, grid interconnection, and siting. Goldman Sachs projects US data center power demand will rise 165% by 2030. Morgan Stanley estimates a 49-gigawatt US supply shortfall by 2028. PJM interconnection queues are measured in years.
You can't conjure megawatts. You secure them years in advance through substation proximity, utility relationships, power purchase agreements, and permitting leverage. That's where disciplined capital wins — not by buying GPUs, but by controlling the real estate where the GPUs will eventually rack.
Our data center strategy is front-run, not chase. We secure power and land optionality in markets where hyperscaler site scouts will be looking in 2027–2029, and we structure capital deployment against phased build economics rather than speculative pre-sales.
Cost pressures in data centers — what investors are missing
Capex is compressing on three axes at once: power, land, and construction. Underwriting has to reflect it.
Per-megawatt DC build costs have moved materially in the last three years. Power inflation, utility interconnection fees, and construction labor have all repriced upward. At the same time, AI workloads demand much higher power density per rack than traditional cloud, which changes cooling, electrical distribution, and facility capex unit economics.
The investors who underwrote data center deals using 2020-era assumptions are finding themselves upside down on projections. Our diligence reflects current cost curves, phases capex to match stabilized tenant cash yield, and preserves optionality on power — which is the real lever.
The chip bottleneck, explained
GPU lead times, CoWoS packaging, and HBM memory — the three constraints that dictate how fast AI capacity actually comes online.
GPU orders for high-end training silicon currently run 36–52 weeks from order to rack. That's not primarily about wafer fab capacity — leading-edge fabs are running. The constraint is downstream: CoWoS advanced packaging capacity at TSMC is booked through 2027, and HBM memory supply from SK hynix, Micron, and Samsung is allocated on multi-quarter horizons.
For DC investors, the chip bottleneck sets the pace at which our tenants can actually absorb capacity. It also justifies front-running land and power — by the time packaging and memory catch up with demand, the hyperscalers looking for racks will be looking in places that already have the power plan approved.
AT&T and the tower investment thesis
Industry-aligned analysis of AT&T's network strategy and what it signals for tower asset holders.
AT&T's long-running FirstNet buildout and fiber-convergence strategy illustrate why tier-one carriers continue to densify their radio access networks even as 5G hits maturity. Every new spectrum band and every fixed-wireless push lands on additional tower sites, and every additional tenant on an existing tower is near-pure margin. That math is the foundation of the tower-as-asset-class thesis.
For Retrospx as a tower investor, AT&T's capital plan, its FirstNet obligations, and its fiber-to-the-home expansion all signal durable demand for site leases in the markets we operate in. We underwrite accordingly.
This is industry-aligned analysis based on publicly available information. Any reference to AT&T is for illustrative purposes and does not represent a specific contractual relationship.
T-Mobile, FWA, and the MVNO opening
How T-Mobile's fixed wireless access buildout and wholesale posture reshape the economics for niche operators on their network.
T-Mobile's US network buildout and aggressive fixed-wireless access strategy have changed the wholesale market. Excess network capacity on rural and suburban sites, combined with FWA adoption, has created new leverage for MVNOs negotiating wholesale access — particularly for IoT and niche cellular plans that don't compete directly with T-Mobile's consumer retail.
For investors building MVNO portfolios, this matters. Wholesale terms negotiated three years ago are not wholesale terms available today. Our telecom strategy reflects current carrier posture on capacity, renegotiation cycles, and IoT-specific wholesale frameworks.
This is industry-aligned analysis based on publicly available information. Any reference to T-Mobile is for illustrative purposes and does not represent a specific contractual relationship.
Infrastructure convergence — commerce, telecom, and compute are one stack
Why a diversified platform across these four verticals outperforms any single-vertical fund.
Ecommerce sits on top of telecom. Telecom sits on top of towers. Towers sit next to data centers. Data centers carry the compute that personalizes ecommerce. The four verticals don't just coexist — they feed each other with signals, customers, infrastructure, and margin.
A pure-play ecommerce fund misses the upstream connectivity economics. A pure-play tower fund misses the downstream commerce data. A pure-play data center fund misses the edge-compute distribution. We operate across the full stack by design — not as a portfolio theory flex, but because the operating synergies compound.
This is the thesis Retrospx was built on. Every insight we publish reflects it.
Want to discuss a specific thesis?
Reach the partners directly. We respond to founders, operators, and strategic partners in our verticals.