Dwarkesh Patel touts a compute land‑grab and Gary Marcus warns of overreach; our listings show OpenAI and Anthropic growing, but most new AI jobs sit in big enterprises and consultancies.
The week’s split-screen: centralization dreams, reality checks
Dwarkesh Patel frames a stark direction for AI: "how Anthropic and OpenAI are on track to control most of the world’s usable FLOPs within the next few years (because they can monetize compute better and thus outbid everyone)." Gary Marcus fires back with skepticism, calling it "Anthropic’s $30 trillion fantasy" and adding, "Meanwhile, back on Planet Earth, Thomson Reuters has become the latest company to pull back on its Claude usage." Ethan Mollick pushes for urgency, noting that not every technology takes decades to transform work: "Ford went from inventing the modern assembly line to rebuilding all work around it in 3 years, cutting car making time by 88%." Simon Willison spotlights agents writing and refining large codebases, quoting Paul Dix: "If you can build a verification system and give proper direction, AI can produce a highly complex, highly sophisticated piece of software and it can continue to refine it until it just works."
Zvi Mowshowitz argues against deferring to consensus: "Modesty arguments often say that you should mostly or entirely bow to ‘expert consensus’ or the views of particular others, and who are you to disagree." He concludes with a reminder that resonates with our job-tracking mission: "Think for yourself, schmuck. Or, as I once put it: You Have The Right To Think, also the moral duty to do so."
Our role is to test these claims against hiring data. Our tracker counts 4,450 open AI roles across 174 employers, taken from their own job feeds.
Who is actually hiring
If compute centralizes, do AI jobs centralize too? Our numbers say no. The biggest single employer in our dataset is Accenture with 802 open roles and 351 opened in 30 days. Capital One has 187 open, Amgen 172, PwC 112, Waymo 97, and Databricks 91. Frontier labs are growing fast but are not the majority: OpenAI lists 155 open roles with 65 opened in 30 days, and Anthropic has 119 open with 38 opened in 30 days.
Across role families, hiring remains broad:
- Modelling and engineering: 1,757 open, 523 opened and 85 closed in 30 days, across 142 employers
- Data: 824 open, 248 opened and 40 closed in 30 days, across 123 employers
- Infrastructure: 412 open, 89 opened and 19 closed in 30 days, across 87 employers
- Research: 225 open, 42 opened and 7 closed in 30 days, across 51 employers
- Product and design: 101 open, 27 opened and 2 closed in 30 days, across 50 employers
- Evaluation and safety: 41 open, 12 opened and 2 closed in 30 days, across 19 employers
This picture supports Mollick’s argument about rapid operational change, more than a slow, top-heavy consolidation of work. The bulk of net-new postings are not just at two labs. They are spread across consulting, finance, biotech, autonomy, and platform companies.
The Anthropic pullback claim vs hiring data
Marcus writes, "The race between fantasy and reality continues," and flags one data point: "Meanwhile, back on Planet Earth, Thomson Reuters has become the latest company to pull back on its Claude usage." Our tracker does not monitor product usage decisions. We cannot confirm or refute changes in Claude adoption from our listings alone.
We can test whether hiring momentum at Anthropic looks like a pullback. It does not. Anthropic currently has 119 open roles, with 38 opened in the last 30 days. OpenAI’s comparable figures are 155 open and 65 opened in 30 days. Those are active pipelines, not a freeze. That does not settle the question of revenue or retention, but it contradicts any inference that hiring has stalled at the labs Marcus critiques.
Centralization of compute vs decentralization of jobs
Patel also raises macro stakes: "the >$10T of total AI capex we’ll see by the end of the decade" and asks whether this could spark a sovereign debt crisis. Our tracker cannot assess capex or debt sustainability. What we can say is that the job market for AI talent is not concentrating in a way that mirrors a compute land grab. With 4,450 open AI roles across 174 employers, and with modelling and engineering openings spread across 142 employers and data roles across 123, hiring today is decentralized.
This does not falsify Patel’s compute thesis. It does show that, so far, the labor side is not marching in lockstep with the capital and compute story. Even if a few players control training compute, the demand for practitioners who deploy, integrate, govern, and productize AI appears to live across many firms, not only the frontier labs.
Adoption speed: the Ford analogy meets the listings
Mollick’s contention about fast reorganization finds support in our openings. Accenture’s 351 roles opened in 30 days and PwC’s 47 hint at clients pushing quickly to stand up AI programs. Modelling and engineering saw 523 openings posted in 30 days, and data roles 248, numbers consistent with teams being rebuilt around new tooling.
Where Mollick’s view meets a fric tion is governance. Marcus highlights themes from Bill Gates, writing, "Excellent new essay by Bill Gates, echoing many of the themes I have been writing about here and in my 2024 book Taming Silicon Valley:" which include the need for plans and frameworks. Our tracker shows just 41 open roles in evaluation and safety across 19 employers. That is a small slice relative to engineering and data. If society needs the kind of systematic plan Marcus and Gates emphasize, hiring for evaluation and safety is not yet scaled to match.
Open models and coding agents widen the field
Willison’s two posts point at widening access. On model availability: "Another open weights model from Qwen." On what agents can do, via Paul Dix: "If you can build a verification system and give proper direction, AI can produce a highly complex, highly sophisticated piece of software and it can continue to refine it until it just works." Our distribution of roles supports a broad base of implementation work. Infrastructure roles number 412 open across 87 employers, and research 225 open across 51. That looks like many organizations building on top of both closed and open systems, rather than a world where only two labs hire.
What the job data cannot tell us
Noah Smith’s city-level observation that "That Market Street is gone now." reflects shifts tied to remote work and local economic cycles. Our listings do not track geography at that resolution here, so we cannot test that claim. Likewise, we cannot verify usage pullbacks at specific vendors, revenue trajectories, or national debt risks from capex. We can only measure who is hiring for what.
For context, our headline jobs-created indicator is counted from job listings and calibrated to the World Economic Forum's Future of Jobs Report 2025 (11M AI roles created, 9M displaced, by 2030). That calibration anchors the signal, but the granular findings above come directly from employer feeds today.
Bottom line
- Patel’s claim that two labs will dominate compute is not mirrored in current hiring. OpenAI and Anthropic are expanding, but the majority of open roles are at large enterprises and consultancies. Our data does not support a concentration of AI jobs into two firms.
- Marcus’s warning signs about pullbacks are not visible in lab hiring. Anthropic and OpenAI both show active openings in the last 30 days. His broader concern about the gap between ambition and reality is consistent with our finding that evaluation and safety roles remain small relative to engineering.
- Mollick’s argument for rapid organizational change aligns with robust openings in modelling, engineering, data, and the surge at integrators like Accenture and PwC.
- Willison’s focus on open weights and coding agents fits a hiring map spread across 174 employers, suggesting a wide implementation frontier.
Zvi Mowshowitz writes, "Think for yourself, schmuck. Or, as I once put it: You Have The Right To Think, also the moral duty to do so." Our read of the listings obliges that stance: the jobs market shows vigorous, distributed adoption of AI with strong lab growth, not yet the labor centralization or slowdown some anticipate. The compute may centralize. The work, for now, does not.
What we read
Every quote above is taken verbatim from one of these posts.
- Dwarkesh Patel: Dylan Patel – Anthropic & OpenAI will have most of the world’s compute
- Gary Marcus: Anthropic’s $30 trillion fantasy, Excellent new Bill Gates essay on the urgency of having a coherent AI
- Zvi Mowshowitz: Against Modesty’s Bailey, On Writing #3
- Noah Smith: The death of Market Street
- Ethan Mollick: I hear the story about how it took 30 years to gain productivity from , I know it is a lot for instructors who never asked for AI to change th
- Simon Willison: Quoting Paul Dix, Qwen3.8-Flash-Next
- Emily M. Bender: Very proud that The AI Con is a finalist for the 2026 Washington State, Help! I was listening to a podcast episode recently where the host and