The Dogfooding Gap

The distance between what companies advocate and what they actually do is widening under competitive pressure. When SUSE executives accidentally revealed they pay for proprietary software despite promoting open source, they exposed a deeper pattern: principles bend when they conflict with operational needs.

This gap appears across multiple domains. Meta is using complex financial structures to keep a $27 billion data center project off its balance sheet, prompting auditor warnings. Venture capital firms are funding direct competitors in the AI race, abandoning a longstanding taboo about portfolio conflicts. Even security practices show similar patterns, with social engineering attacks on payroll systems succeeding because business processes prioritize speed over verification.

The common thread is scale forcing compromise. Open source tools may not handle enterprise collaboration at SUSE's size. Meta's capital requirements exceed what traditional accounting would allow while maintaining investor optics. VCs cannot afford to miss the AI wave by backing only one horse. These are rational responses to competitive dynamics, but they create authenticity problems.

The second-order effect matters more than the hypocrisy itself. When the gap between stated values and actual practice becomes standard, it signals that market forces have overpowered institutional principles. That recalibration changes what customers and employees can reasonably expect from corporate commitments.

Deep Dive

When Your Own Products Cannot Solve Your Problems

The revelation that SUSE uses Microsoft Teams and Red Hat relies on Gmail matters less as hypocrisy than as market signal. These companies sell enterprise Linux and advocate for digital sovereignty, yet they pay competitors for collaboration tools. This tells you where open source still has gaps at scale, and those gaps represent either opportunity or fundamental limitation.

Microsoft proved decades ago that dogfooding drives product improvement. When it migrated Hotmail from FreeBSD to Windows Server despite significant cost and complexity, the effort identified weaknesses in its own stack and forced improvements that made the product competitive for similar workloads. The company later did the same moving LinkedIn from CentOS to Azure Linux. These migrations had no direct revenue impact but delivered strategic value through reduced vendor dependence and improved engineering knowledge.

FOSS vendors face the same dynamics but respond differently. SUSE has 2,500 employees, Red Hat has 19,000, and Canonical has 1,200. At that scale, collaboration infrastructure matters operationally. Yet rather than migrate to open alternatives like Nextcloud, Grommunio, or Open-Xchange, they pay Microsoft and Google. The decision reveals that available FOSS groupware either cannot handle their operational requirements or the migration cost exceeds the strategic benefit they would gain.

For founders, this creates a clear market map. If companies that exist to promote open source will not use open source collaboration tools, the product-market fit problem runs deep. Either the tools genuinely cannot scale to enterprise needs, or the switching costs and feature gaps make migration economically irrational. Both scenarios point to opportunity for builders who can solve the actual enterprise collaboration problem rather than building what the ideology suggests should exist. The gap between what FOSS vendors advocate and what they actually run represents the difference between aspiration and operational reality.

Portfolio Conflicts Stop Mattering in Winner-Take-All Markets

Major venture firms are now backing both OpenAI and Anthropic, abandoning a longstanding principle against funding direct competitors. This is not sloppiness. It reflects a rational response when the cost of missing a category exceeds the cost of managing portfolio conflicts. The AI race has recalibrated what matters in fund construction.

Traditional VC practice avoided backing competitors for clear reasons. Portfolio companies share information with their investors, creating potential for leaks. Board members face conflicts when competing companies need different strategic advice. Most importantly, backing both sides in a race reduces your upside concentration. If one winner will capture most of the value, spreading bets across competitors means you own less of the winner.

Those economics shift in true winner-take-all dynamics with massive total addressable markets. If AI foundation models create trillions in value and you can own pieces of multiple credible contenders, the expected value of diversification exceeds the cost of conflicts. Missing the category entirely by backing the wrong horse becomes the unacceptable outcome. This is particularly true when capital requirements run into billions and follow-on ownership gets diluted anyway. Better to have board seats at both OpenAI and Anthropic than full ownership of neither.

For founders, this changes fundraising dynamics in hot categories. Investors will no longer offer traditional exclusivity or commitment to back you over competitors. Expect term sheets with weaker pro-rata rights and more flexible redemption terms. The VCs are explicitly hedging their bets, which means your strategic value to them diminishes relative to other portfolio companies where they have concentration.

For other VCs, the pattern is spreading beyond AI. When categories hit genuine winner-take-all dynamics with huge markets, expect the same portfolio construction shift. The taboo against backing competitors only holds when maintaining it improves returns. Once it becomes a handicap, the practice evolves.

Signal Shots

Waymo Targets Million-Ride Week : Waymo expects to hit 1 million paid weekly robotaxi rides in the US by end of 2026, up from 400,000 per week across six cities today. The milestone matters because it pushes autonomous vehicles from pilot to infrastructure scale, where unit economics and regulatory frameworks face real-world stress testing. Watch whether the growth comes from geographic expansion or density increases in existing markets. The latter would signal genuine product-market fit rather than just capital deployment across new cities.

Private Equity's SaaS Exposure Meets AI Risk : Private equity and private credit groups face potential losses on SaaS investments as AI threatens to make digital services obsolete. PE activity concentrated heavily in SaaS over the past decade, creating concentrated exposure to automation risk. Watch which specific SaaS categories show pricing pressure or churn acceleration first. Point solutions with narrow workflows are most vulnerable to AI replacement, while platforms with strong network effects and integrated workflows have better defensive positions.

Starlink Creates Jurisdiction Vacuum : Research shows Starlink sometimes moves data faster than terrestrial networks using satellite-to-satellite laser links, but regulators struggle with legal frameworks. APNIC notes that Starlink users change national jurisdiction mid-flight and the service lands traffic in unexpected countries based on geopolitical constraints. Watch how governments balance sovereignty concerns against rural connectivity benefits. The regulatory gap matters because it sets precedent for how nations handle infrastructure that exists partially outside territorial control.

SpaceX Escapes Labor Relations Oversight : The National Labor Relations Board classified SpaceX as a common carrier under the Railway Labor Act, exempting it from standard labor law and making unionization significantly harder. The decision matters because it extends airline-style labor frameworks to spaceflight based on thin evidence about mail contracts and future passenger services. Watch whether other space companies seek similar classification and whether courts uphold the precedent. The Railway Labor Act's extensive dispute resolution process effectively blocks strikes, changing the bargaining dynamics for an entire emerging industry.

Chrome Extensions Harvest 37 Million Users' History : Security researchers identified 287 Chrome extensions selling browsing history data to data brokers like Similarweb, affecting 37.4 million installations. The extensions request broad permissions without justification while privacy policies obscure actual data practices. Watch whether Google enforces its Limited Use policy more aggressively or if the exception clauses remain exploitable. The research documents that web analytics firms continue harvesting data through extensions despite years of public scrutiny, suggesting enforcement gaps matter more than policy existence.

Scanning the Wire

Modal Labs Seeks $2.5B Valuation for AI Inference : The four-year-old startup is in talks with General Catalyst to lead a new funding round, capitalizing on surging demand for GPU compute infrastructure as AI deployment shifts from training to inference workloads. (TechCrunch)

Trane Acquires LiquidStack as GPU Cooling Becomes Strategic : The datacenter cooling market is consolidating as AI hardware requirements drive demand for liquid cooling solutions, with traditional HVAC companies now viewing specialized GPU cooling startups as critical infrastructure plays. (The Register)

Grab Acquires Stash Financial for $425M Despite Revenue Miss : The Singapore rideshare company is expanding into US fintech while forecasting 2026 revenue of $4.04B to $4.1B, below analyst expectations of $4.13B, signaling slower growth momentum in its core business. (Reuters)

Adyen Stock Falls 15% on Conservative Growth Forecast : The Dutch payments processor reported H2 2025 net revenue up 17% to €1.27B but guided 2026 growth at 20% to 22%, missing the 22.8% estimate and suggesting payment processing volume growth is decelerating. (CNBC)

Cisco Raises Hardware Prices to Cover Memory Costs : The networking giant is passing through increased memory component costs to customers and reports that larger bills are not changing enterprise buying patterns, suggesting inelastic demand for campus infrastructure refresh cycles. (The Register)

Microsoft Patches Six Zero-Days Exploited in the Wild : February's Patch Tuesday includes fixes for a half-dozen vulnerabilities that attackers exploited before Microsoft released patches, highlighting the persistent gap between vulnerability discovery and remediation deployment. (The Register)

DeepSeek Expands Context Window to Over 1 Million Tokens : The Chinese AI startup quietly upgraded its flagship model's context window from 128K tokens to over 1 million, allowing the system to process and remember significantly more information in single conversations without documented changes to underlying architecture. (South China Morning Post)

Meta Launches 'Dear Algo' for Threads Feed Customization : The new feature lets users post AI prompts publicly to temporarily adjust what content appears in their Threads feed, representing Meta's latest attempt to give users algorithmic control while maintaining engagement optimization. (TechCrunch)

Apple's Siri Overhaul Delayed Beyond March Timeline : The AI-powered Siri revamp originally expected in iOS 26.4 will now roll out incrementally through May or potentially delay until iOS 27 in September, extending the gap between announcement and delivery of Apple's answer to ChatGPT integration. (TechCrunch)

Archive.today CAPTCHA Executes DDoS Against Critics : The archiving service's CAPTCHA page launched a distributed denial-of-service attack against a blog that attempted to uncover the founder's identity in 2023, prompting Wikipedia to consider banning links to the site over malicious code concerns. (Ars Technica)

Outlier

The Context Window Arms Race Goes Parabolic : DeepSeek quietly expanded its context window from 128K to over 1 million tokens without architectural fanfare, marking another step in a capability race that is outpacing actual use cases. Most applications struggle to utilize even 32K tokens effectively, yet labs keep pushing boundaries as if infinite context is the goal. This signals two things: context window size has become a proxy metric for model capability in the absence of other clear differentiation, and labs are optimizing for benchmark dominance rather than practical deployment needs. When competitive dynamics reward capabilities that users cannot yet exploit, it hints at a maturity phase where marketing specifications matter more than operational improvements. The gap between what models can theoretically handle and what applications actually need keeps widening.

The companies that build the tools still buy someone else's. The funds that preach focus now hedge everything. Somewhere in that gap between what we say and what we do lives the actual market.