👀 Mira's new model
1. Key Themes
Theme 1: Open-Weight / Customizable AI as Enterprise Strategy
The dominant investment thesis in this issue is that enterprises will increasingly reject expensive, closed frontier models in favor of open-weight models they can fine-tune on proprietary data and self-host — a direct challenge to OpenAI and Anthropic's business models.
"Thinking Machines is making a different bet than many AI labs: that enterprises ultimately care less about the smartest general-purpose model than one they can make their own."
"Organizations can fine-tune open-weight models on their own proprietary data and deploy those models on infrastructure that they control, giving users more flexibility over hosting and costs than most closed models."
Theme 2: AI Moving Off Screens Into Physical Hardware
Both the Codex Micro keypad and OpenAI's rumored smart speaker signal that AI interaction is migrating into dedicated physical devices — creating a new hardware product category.
"A physical controller for coding agents at work and a screen-free AI companion for the home both signal AI moving off our screens and into hardware."
"That was a single button to summon AI for chatting. Two years later, Codex Micro is a whole control surface for people living with agents all day."
Theme 3: AI's Environmental Footprint as a Reputational and Regulatory Risk
Big Tech's emissions and water disclosures are increasingly divergent, and transparency itself is becoming a competitive and political variable — not just a PR exercise.
"Their willingness to disclose these impacts is becoming almost as important as the impacts themselves."
"Transparency is emerging as a key response to growing opposition to AI focused on data centers' energy and water toll."
Theme 4: Agent Management as the New Workflow Paradigm
The Codex Micro story reveals that "agentmaxxers" — power users running multiple autonomous AI agents simultaneously — represent a growing, distinct user segment requiring new tooling.
"The future of work could involve managing fleets of agents, not just chatting with one assistant."
"AI usage is fragmenting as much by which model you use as by how you use the model."
2. Contrarian Perspectives
Perspective 1: Being the "Best" Model Is Not the Winning Enterprise Strategy
Against the mainstream race to top benchmarks, Thinking Machines is deliberately conceding raw capability to focus on customizability — and raised $2B at a $12B valuation before shipping a product, suggesting the market agrees.
"The company is clear that Inkling is not the strongest model available, instead focusing on how the model is customizable, which could help users get better performance with lower costs."
"Instead of trying to beat competitors like OpenAI and Anthropic on model benchmarks, Thinking Machines is currently focused on customization."
Supporting evidence: Palantir CEO Alex Karp amplified this thesis publicly, arguing that closed frontier tools are both too expensive and insufficiently clear on IP protections — giving the enterprise open-weight thesis mainstream executive validation.
Perspective 2: Open-Weight Releases May Be a Temporary, Tactical Choice — Not a Philosophical Commitment
Murati's history at OpenAI suggests her current openness stance is situational, not ideological. Thinking Machines may quietly close its models as capabilities (and risks) increase.
"Just because Inkling is open weight doesn't mean the rest of Thinking Machines' models will be. Her current strategy could mirror that case-by-case logic: Release models openly when the risks are manageable and hold them back when they aren't."
"She was at OpenAI in 2019 when the lab — founded on a promise of openness — withheld the full version of GPT-2 over misuse fears, heralding the company's retreat from fully open releases."
Perspective 3: Chinese AI Labs Are Already Embedded in Western AI Supply Chains
The detail that Thinking Machines used training data generated by Moonshot AI's Kimi K2.5 — a Chinese lab — is a quiet signal that the AI training data supply chain crosses geopolitical lines regardless of public positioning.
"Thinking Machines used data generated by existing open models — including Kimi K2.5 from Chinese lab Moonshot AI — in its final training phase."
3. Companies Identified
Thinking Machines
- Description: AI startup founded by former OpenAI CTO Mira Murati
- Why mentioned: Launched Inkling, its first open-weight foundational model, targeting enterprise customization
- Quote: "Thinking Machines raised a record $2 billion seed round at a $12 billion valuation in 2025, before it had released a model or product."
OpenAI
- Description: Leading closed AI lab, maker of ChatGPT and Codex
- Why mentioned: Launched Codex Micro hardware keypad for agent power users; also developing a smart home speaker; facing Apple lawsuit
- Quote: "OpenAI opened orders for Codex Micro — a limited-edition desktop keypad that lets agentmaxxers monitor and control their AI minions."
Anthropic
- Description: AI safety-focused lab, maker of Claude models
- Why mentioned: Two signals: JPMorgan CEO flagged risks around its Mythos model; company is scheduling investor meetings ahead of a potential IPO
- Quote: "Anthropic is scheduling investor meetings as it weighs an IPO."
- Description: Dominant AI chip manufacturer
- Why mentioned: Key infrastructure partner and investor in Thinking Machines; Inkling was trained on Nvidia's latest infrastructure
- Quote: "Thinking Machines trained the model on Nvidia's latest AI infrastructure, underscoring the company's partnership with the chip giant."
- Description: Big Tech AI and cloud company
- Why mentioned: Reportedly signed a multibillion-dollar cloud deal with Thinking Machines; also released new environmental impact report
- Quote: "Thinking Machines also reportedly signed a multibillion-dollar Google Cloud deal."
Moonshot AI (Kimi)
- Description: Chinese AI lab
- Why mentioned: Its Kimi K2.5 open model was used as a data source in Thinking Machines' training pipeline
- Quote: "Thinking Machines used data generated by existing open models — including Kimi K2.5 from Chinese lab Moonshot AI — in its final training phase."
- Description: Boutique hardware company specializing in customizable mechanical keyboards
- Why mentioned: Collaboration partner with OpenAI on the Codex Micro keypad
- Quote: "Codex Micro is a collaboration with Work Louder, a boutique hardware company known for customizable mechanical keyboards and shortcut controllers for developers and designers."
Palantir
- Description: Data analytics and enterprise software company
- Why mentioned: CEO Alex Karp went viral making the enterprise case against closed frontier AI models
- Quote: "Frontier AI tools from closed model providers are too expensive and don't offer enough clarity on IP protections."
- Description: Big Tech cloud and AI companies
- Why mentioned: Both released environmental reports showing rising emissions and water use amid AI infrastructure expansion
- Quote: "New environmental reports from Google, Amazon and Microsoft show emissions and water use continuing to rise as AI infrastructure expands."
4. People Identified
Mira Murati
- Description: Former CTO of OpenAI; founder and CEO of Thinking Machines
- Why mentioned: Central figure — her startup just shipped its first model, Inkling, making a deliberate bet on enterprise customizability over raw capability
- Quote: "Thinking Machines is making a different bet than many AI labs: that enterprises ultimately care less about the smartest general-purpose model than one they can make their own."
Alex Karp
- Description: CEO of Palantir
- Why mentioned: Went viral on CNBC articulating the enterprise case for open-weight AI over expensive closed frontier models
- Quote: "Frontier AI tools from closed model providers are too expensive and don't offer enough clarity on IP protections."
- Description: CEO of JPMorgan Chase
- Why mentioned: Publicly flagged risks associated with Anthropic's Mythos model
- Quote: "JPMorgan CEO Jamie Dimon said risks associated with Anthropic's Mythos model are a 'real issue.'"
5. Operating Insights
Insight 1: Enterprise AI Buyers Should Demand Open Weights and Infrastructure Control
The article makes a clear case that open-weight models deliver a structurally better cost and flexibility profile for enterprises — fine-tune on proprietary data, deploy on your own infrastructure, avoid IP ambiguity.
"Organizations can fine-tune open-weight models on their own proprietary data and deploy those models on infrastructure that they control, giving users more flexibility over hosting and costs than most closed models."
Insight 2: Agent Fleet Management Requires New UX Primitives
As agent usage scales, operators need to invest in monitoring, approval flows, and status dashboards — the Codex Micro is an early signal of what "agent ops" tooling will need to deliver. The approval UX is also a live safety design problem.
"There's a button on the keyboard to approve an agent's access. That seems like an easy way to accidentally give an agent access or approve a task that you didn't mean to approve."
6. Overlooked Insights
Insight 1: Anthropic IPO Signals a Potential Liquidity Moment for the Entire AI Sector
Buried in the "Training Data" roundup, Anthropic scheduling investor meetings ahead of a potential IPO is a significant market structure event that could set public valuation benchmarks for the entire frontier AI lab category — yet it received only a one-line mention.
"Anthropic is scheduling investor meetings as it weighs an IPO."
Insight 2: OpenAI's Smart Speaker Faces a Concrete Legal Threat to Its Hardware Timeline
The planned 2026-announce / 2027-ship consumer speaker could be derailed by Apple's active lawsuit — a risk that gets one sentence but could materially affect OpenAI's consumer hardware ambitions.
"That timeline could hit roadblocks over Apple's recent lawsuit against OpenAI alleging stolen hardware secrets."