Teahose.
SIGN IN
NEW HERE — WHAT TEAHOSE DOES
We read the entire AI & tech firehose — so you don't have to.
PODPodcastsAll-In, No Priors, Acquired…
NEWNewslettersStratechery, Newcomer…
PAPPapersPhysical AI research
PHProduct Huntdaily launches
VCInvestor ScoutSequoia, a16z, Benchmark…
CLAUDE DISTILLS →
7 reads, 30 sec each — free, 6 AM ET.
+ a live graph of the companies, people & themes underneath.
HOME/AXIOS AI+/🔄 Routing roulette
NEWS
// NEWSLETTER ISSUE
AXIOS AI+

🔄 Routing roulette

DATE August 25, 2026SOURCE AXIOS AI+PARTICIPANTS AXIOS AI+
AI AgentsCompetitive StrategyEnterprise SaaSLarge Language ModelsOpen SourceProduct & Design
In this episode
AnthropicOpenAINvidiaGoogleMetaStripeHugging FacePalantirOpenRouterFactoryCiscoOuraLuffuTrustedRouterSupermicroMatan GrinbergJoseph PerlaJames Park
// SUMMARY

1. Key Themes

Theme 1: AI Model Routing Is Emerging as a Major Threat to Frontier Lab Margins

Routing — directing each query to the most efficient model rather than always using the most powerful one — is rapidly maturing into an industry-reshaping force. The business logic is simple: why pay frontier prices for simple tasks?

"Routing turns models from destinations in their own right into interchangeable commodities, which could threaten margins for frontier AI models."

The scale is already significant: "Stripe agreed to buy unicorn OpenRouter for more than $8 billion" and "OpenRouter had 8 million users and 400 models available on the platform as of May."

Theme 2: Open-Source and Open-Weight Models Are Winning on Usage

Despite the marketing dominance of OpenAI, Google, and Anthropic, actual enterprise usage is skewing heavily toward open models — a leading indicator of where enterprise AI spend may shift.

"The five most popular models used on OpenRouter are all open weight or open source. OpenAI and Google each have only one model in the top 10, and Anthropic has none."

Theme 3: AI Data Sovereignty Is a Real and Growing Enterprise Priority

Enterprise reluctance to share proprietary data with frontier labs is fueling demand for routing platforms that offer privacy controls — a trend that incumbents like Palantir are actively capitalizing on.

"I see people excited about us and moving to us because they don't like what Anthropic is becoming." — Joseph Perla, founder of TrustedRouter

Palantir recently hosted a sovereignty boot camp in NYC packed with speakers, including one who said, "Are we going to throw our years of knowledge to a frontier AI model? Absolutely not."

Theme 4: AI-Enabled Elder Care Hardware Is an Emerging Consumer Market

The Luffu launch signals a maturing of AI-powered wearables beyond fitness into family-connected health monitoring — targeting the large and underserved "sandwich generation" of caregivers.

"The project aims to address the challenges faced by caregivers, including those in the sandwich generation with aging parents and kids at home."

"The purpose of it is to keep people healthy, safe, independent, but also connected to the people who care about them." — James Park, Luffu CEO

Theme 5: Infrastructure Demand Remains Robust — Incumbents Are Betting Big

Cisco's expansion of its Nvidia-backed AI infrastructure offering into high-density compute signals that major players see no near-term slowdown in AI hardware demand.

"Cisco is leaning into its AI infrastructure business, betting that compute demand is not slowing anytime soon."

2. Contrarian Perspectives

Routing Won't Kill Frontier Labs — It May Actually Expand Their Reach

The consensus fear is that routing commoditizes frontier models. The contrarian read from operators on the ground: routing platforms may actually surface frontier models to more users and workloads than they'd reach otherwise.

"It's not to say you won't ever use commercial models, but it's like you want to have the options to use models that are specific to you." — Akshay Krishnaswamy, Chief Architect, Palantir

Demand for intelligence is such that there can be multiple winners, because these platforms also "put their models in front of more users and more workloads." — Matan Grinberg, CEO, Factory AI

Falling Token Prices Reflect Competitive Desperation, Not Efficiency Wins

Average token costs have "plummeted since a peak in mid-May due to price cuts from frontier AI labs looking to gain back share from their competitors." This framing suggests labs are reacting defensively to routing and open-source pressure — not leading on cost innovation — which is a structurally weaker position.

Meta Building a Routing Competitor Is a Cost Control Play, Not a Commerce Play

While Stripe acquiring OpenRouter is clearly a payments/commerce integration strategy, Meta's reported routing tool "Switchboard" is described primarily as an internal cost-cutting measure — suggesting even hyperscalers find frontier model costs unsustainable at scale.

"Meta is reportedly working on an OpenRouter competitor called Switchboard, which the company may use to lower its own AI costs."

3. Companies Identified

OpenRouter

  • Description: AI model routing platform with 8 million users and 400 models
  • Why mentioned: Acquired by Stripe for $8B+; central example of routing boom
  • Quote: "Stripe agreed to buy unicorn OpenRouter for more than $8 billion"

Stripe

  • Description: Payments infrastructure giant
  • Why mentioned: Made a major strategic bet on AI routing by acquiring OpenRouter
  • Quote: "Stripe agreed to buy unicorn OpenRouter for more than $8 billion, the company announced last week."

TrustedRouter

  • Description: Open-source AI routing company with end-to-end encryption
  • Why mentioned: Case study for enterprise AI sovereignty features; founder quoted
  • Quote: "TrustedRouter is encrypted end to end to make compute 'confidential.'"

Factory AI

  • Description: AI routing company
  • Why mentioned: CEO offers bullish counter-narrative on routing's impact on frontier labs
  • Quote: "Demand for intelligence is such that there can be multiple winners, because these platforms also 'put their models in front of more users and more workloads.'"

Meta

  • Description: Social media and AI hyperscaler
  • Why mentioned: Reportedly building "Switchboard," an internal OpenRouter competitor, to cut its own AI costs
  • Quote: "Meta is reportedly working on an OpenRouter competitor called Switchboard, which the company may use to lower its own AI costs."

Palantir

  • Description: Data analytics and AI platform company
  • Why mentioned: Actively marketing AI sovereignty; hosted a sovereignty boot camp for enterprise executives
  • Quote: "Palantir recently hosted a sovereignty boot camp in NYC packed with speakers."

Luffu

  • Description: AI health wearable startup founded by Fitbit co-founders
  • Why mentioned: Launching the Luffu Link, an AI-powered family health monitoring band targeting elder care and caregivers
  • Quote: "The project aims to address the challenges faced by caregivers, including those in the sandwich generation with aging parents and kids at home."

Cisco

  • Description: Enterprise networking and infrastructure company
  • Why mentioned: Expanding its Nvidia-backed "Secure AI Factory" with Supermicro servers, betting on sustained compute demand
  • Quote: "Cisco is leaning into its AI infrastructure business, betting that compute demand is not slowing anytime soon."

Supermicro

  • Description: High-performance server manufacturer
  • Why mentioned: Partnering with Cisco to add liquid- and air-cooled AI compute systems to its infrastructure stack
  • Quote: "Cisco is working with Supermicro to add high-density, liquid- and air-cooled computing systems to its Secure AI Factory with Nvidia."

Hugging Face

  • Description: Open-source AI model repository and platform
  • Why mentioned: Reportedly in acquisition talks at a $13 billion valuation
  • Quote: "Hugging Face is reportedly in talks to be acquired for $13 billion."

Oura

  • Description: Smart ring health wearable maker
  • Why mentioned: Seeking up to $3 billion in a U.S. IPO — a benchmark for AI-adjacent health hardware market appetite
  • Quote: "Smart ring maker Oura is looking to raise up to $3 billion in its U.S. IPO."

4. People Identified

Joseph Perla

  • Description: Founder, TrustedRouter
  • Why mentioned: Building an enterprise-focused, privacy-first AI routing alternative; vocal critic of frontier lab data practices
  • Quote: "I see people excited about us and moving to us because they don't like what Anthropic is becoming."

Akshay Krishnaswamy

  • Description: Chief Architect, Palantir
  • Why mentioned: Articulates enterprise nuance — routing as optionality, not frontier replacement
  • Quote: "It's not to say you won't ever use commercial models, but it's like you want to have the options to use models that are specific to you."

Matan Grinberg

  • Description: CEO, Factory AI
  • Why mentioned: Offers a positive-sum framing: routing expands, not cannibalizes, frontier model exposure
  • Quote: "Demand for intelligence is such that there can be multiple winners, because these platforms also 'put their models in front of more users and more workloads.'"

James Park

  • Description: CEO of Luffu; former CEO of Fitbit
  • Why mentioned: Serial hardware entrepreneur applying AI to elder care and family health monitoring
  • Quote: "The purpose of it is to keep people healthy, safe, independent, but also connected to the people who care about them."

5. Operating Insights

Set Router Priorities Explicitly — Cost, Speed, Trust, and Performance Are All Configurable

For operators deploying AI at scale, routing is not just a cost play — it's a governance tool. The article notes companies can already define their own routing priorities:

"Companies can set their own priorities for the router, including cost, speed, performance and which models or providers they trust with their data."

Operators should build this decision logic now, before vendor lock-in solidifies.

Routing Onboarding Friction Is Near Zero — The Window to Test Is Now

The barrier to experimenting with routing is extremely low, making this a near-zero-cost capability test for any engineering team.

"Routing companies say that it's extremely quick and easy to get started. (In a demo, I was able to use a DeepSeek model in under 30 seconds using TrustedRouter.)"

Hardware-Plus-Subscription Bundling Works for AI Consumer Devices

Luffu's pricing model — a $299 device paired with a mandatory $20/month family subscription covering up to four people — offers a replicable playbook for AI hardware startups seeking recurring revenue without charging separately for connectivity.

"The company won't charge separately for cellular service, but the band requires the company's $20-per-month family subscription plan (covers up to four people)."

6. Overlooked Insights

Frontier Labs Have Quietly Built Their Own Model Pickers — Acknowledging the Routing Trend

This detail is easy to miss: the frontier labs themselves are now offering tiered intelligence selectors, which is a tacit admission that always using the most powerful (and expensive) model is neither necessary nor desired by users.

"Even the frontier AI labs have launched model pickers that allow users to select which model or level of intelligence they want to use for a certain task."

This is a defensive product move that validates the routing thesis from within.

Hugging Face's Potential $13B Acquisition Signals That Open-Model Infrastructure Has Standalone Strategic Value

Mentioned only briefly in the "Training Data" section, a $13B acquisition price for Hugging Face — the primary hub for open-weight models — would make it one of the largest AI infrastructure exits to date, and would directly reinforce the theme that open models are gaining structural leverage over proprietary ones.

"Hugging Face is reportedly in talks to be acquired for $13 billion."