April 9: 156 qubits, agentic supercycles, and contested AGI bets
This morning we flagged AI Job Evaporation Timelines and Mindfulness. Here's how it resolved.
Quantum Practicality at Scale
Current hardware solved nontrivial optimization at 156 qubits with perfect accuracy, moving useful quantum advantage years closer.
These independent results add up to an emerging consensus that practical quantum computing does not require waiting for millions of fault-tolerant logical qubits. Hybrid classical-quantum pipelines, smart ansatzes, dual variational updates, and error suppression techniques let today's noisy hardware deliver useful outputs on optimization problems and navigation. [1] [2] Wang's work stands out. Without the full stack of custom compilation and post-processing the outputs collapse to random guessing, yet with it the system hits 100 percent approximation on 3-regular Max-Cut graphs at 156 qubits. Preskill's neutral atom paper shows similar compression for cryptography. [3] [4] In plain English this means error mitigation primitives act like the early GPU optimizations that unlocked machine learning before perfect silicon. For a smart non-specialist founder the stakes are concrete. Supply chain optimization, portfolio risk models, or materials discovery could see quantum-classically hybrid tools inside two to three years. Crypto teams relying on ECC or RSA for long-lived data must accelerate post-quantum migration plans now. The timeline just moved. This connects to the agentic supercycle thread because both trends compress hardware and org adaptation windows faster than most models predicted this morning. [5]
“The video critiques commercial note-taking applications like Notion for their data privacy implications and bloat, advocating for open-source, non-commercial, and plain-text alternatives.”— Andrej Karpathy [5]
Sources (5)
- Error-Suppressed Quantum Pipeline — Yulun Wang“A hybrid quantum-classical pipeline integrates custom ansatz, dual variational updates, parametric compilation, hardware error suppression, and O(n) post-processing to solve unconstrained binary combinatorial optimization on gate-model quantum hardwa...”
- Neutral Atom Quantum Computers Threaten Current Cryptography — John Preskill“A recent breakthrough demonstrates how neutral atom-based quantum computing, leveraging concatenated codes and a four-zone architecture, dramatically reduces the physical qubit requirements for breaking current cryptographic standards like ECC256 and...”
- QEC Primitives Enhance Quantum Computation — Yuval Baum“This paper demonstrates that applying quantum error correction primitives, without full logical encoding, offers significant computational advantages on superconducting processors. This approach has led to the generation of the largest reported 75-qu...”
- Quantum Magnetometry Enables GNSS-Independent Navigation — Michael Hush“A quantum-assured magnetic navigation system using proprietary quantum magnetometers and novel denoising algorithms has demonstrated positioning accuracy significantly superior to strategic-grade inertial navigation systems.”
- Navigating the Open-Source Note-Taking Ecosystem — Andrej Karpathy“The video critiques commercial note-taking applications like Notion for their data privacy implications and bloat, advocating for open-source, non-commercial, and plain-text alternatives.”
The Agentic Compute Supercycle
AI infrastructure is pivoting from training giant models to powering millions of autonomous agents, driving projected million-fold compute growth.
The positions converge on a clear pattern. Training was the warm-up. The real scaling phase is inference and agent orchestration at population scale. [1] Huang describes the shift explicitly as moving from valuing AI as a hardware play to foundational infrastructure with secondary players like CoreWeave and Oracle capturing specialized demand. Levie adds that technical success alone fails without org redesign. Legacy human operating models are the primary blocker. [2] Mollick's jagged frontier concept explains why some knowledge work evaporates quickly while other roles pivot to oversight and creativity. [3] a16z reinforces that software demand grows rather than shrinks. Think of it like the move from on-prem servers to AWS Lambda in 2014. The primitives changed, the total compute consumed exploded, and new winners emerged. For founders this means your 2026 capex plan must price in GPU-as-a-service, agent orchestration layers, and sandboxing infrastructure. Power contracts and data center strategy become board-level. The morning's job evaporation discussion gains clarity here. It is less total disappearance and more radical reorganization around what humans uniquely provide. [4] This thread connects directly to the contested AGI definitions thread because the supercycle bets rest on Huang's forecasts being directionally correct.
“The AI infrastructure landscape is shifting from a training-centric model to an inference and agentic-workload model, driving a projected million-fold increase in compute demand.”— Jensen Huang [1]
Sources (4)
- The Compute Supercycle — Jensen Huang“The AI infrastructure landscape is shifting from a training-centric model to an inference and agentic-workload model, driving a projected million-fold increase in compute demand.”
- Scaling Agentic AI — Aaron Levie“The transition from AI pilots to production-scale agentic systems requires a shift from simple LLM prompting to a robust 'microservices' architecture for agents, emphasizing secure sandboxing, high-throughput inference, and standardized tool integrat...”
- AI Agents Drive Rapid Transformation — Ethan Mollick“AI is rapidly evolving from co-intelligence tools to autonomous AI agents capable of achieving goals with minimal human intervention. The 'jagged frontier' of AI means it transforms specific tasks faster than others.”
- The Persistence of Software — Andreessen Horowitz“Historical technological transitions demonstrate that new paradigms rarely eliminate existing architectures but instead expand the total addressable market for software. AI is positioned as a similar catalyst.”
Jensen Huang's Contested AGI Narrative
Nvidia's CEO redefined AGI as the ability to build and run a billion-dollar company, but the claim faces immediate pushback as rhetorical rather than technical.
This is the mandatory contradiction thread. Huang's recent statements position AGI as already achieved under a pragmatic economic definition and project massive compute growth as agents take over. [1] The tracked counter_claims push back directly. One entry states verbatim: "The statement may be taken out of context; Huang's 'mic drop' comment was likely rhetorical or humorous, not a serious technical claim about AGI. Defining AGI as building and running a billion-dollar company is a non-standard benchmark that could be interpreted as a thought experiment rather than a formal definition." [2] A second counters the scaling claim: "The 100x multipliers likely refer to specific research benchmarks or worst-case theoretical requirements, not industry-wide scaling. In practice, algorithmic improvements, specialized hardware, model compression, and distributed computing techniques dramatically mitigate such exponential growth." The evidence leans toward the counters. Real hardware papers in the quantum thread and Levie's agent architecture work both show clever mitigation beating raw scale. Yet Huang's supply chain engagement with hundreds of CEOs and Adobe's decision to build its own frontier models suggest he is positioning Nvidia for the inference era regardless of exact definitions. The synthesis is that definitions are moving targets that serve narrative and capital allocation purposes. Smart builders should treat the directional signal (more compute, more agents, more infrastructure) as real while discounting the precise numbers and AGI arrival dates as sales framing. This thread ties the quantum practicality and agentic supercycle threads together. If the counters hold, the total addressable market for GPUs and specialized systems remains enormous but the timeline and winner-take-all dynamics soften. [5]
“Nvidia CEO Jensen Huang initially claimed AGI has been achieved, defining it as the ability to build and run a billion-dollar company.”— Jensen Huang [1]
Sources (5)
- Jensen Huang's Shifting AGI Narrative — Jensen Huang“Nvidia CEO Jensen Huang initially claimed AGI has been achieved, defining it as the ability to build and run a billion-dollar company.”
- Provided contradictions on Jensen Huang — Counter synthesis“The statement may be taken out of context; Huang's 'mic drop' comment was likely rhetorical or humorous, not a serious technical claim about AGI. Defining AGI as building and running a billion-dollar company is a non-standard benchmark that could be ...”
- Rethinking AGI Development — François Chollet“While current LLM advancements are impressive for domains with verifiable rewards, true AGI requires a more fundamental, self-improving algorithmic approach that minimizes human intervention.”
- The Containment Imperative — Mustafa Suleyman“Technological development is characterized by an inherent loss of control, where complex real-world systems trigger unpredictable 'revenge effects' and nth-order consequences. Containment is the only viable mechanism.”
- Jensen Huang on Proactively Managing AI Supply Chain — Jensen Huang“NVIDIA's CEO dedicates significant effort to proactively manage and de-risk the AI supply chain by informing and collaborating with upstream and downstream partners.”
The open question: If quantum solves real problems at 156 qubits and agents demand million-fold compute, what infrastructure and human skill bets actually compound over the next 24 months?
- Yulun Wang — Error-Suppressed Quantum Pipeline
- John Preskill — Neutral Atom Quantum Computers Threaten Current Cryptography
- Yuval Baum — QEC Primitives Enhance Quantum Computation
- Michael Hush — Quantum Magnetometry Enables GNSS-Independent Navigation
- Andrej Karpathy — Navigating the Open-Source Note-Taking Ecosystem
- Jensen Huang — The Compute Supercycle
- Aaron Levie — Scaling Agentic AI
- Ethan Mollick — AI Agents Drive Rapid Transformation
- Andreessen Horowitz — The Persistence of Software
- Jensen Huang — Jensen Huang's Shifting AGI Narrative
- François Chollet — Rethinking AGI Development
- Mustafa Suleyman — The Containment Imperative
- Jensen Huang — Jensen Huang on Proactively Managing AI Supply Chain
Transcript
REZA: This morning we flagged AI Job Evaporation Timelines and Mindfulness. Here's how it resolved. MARA: Jagged transformation, not pure evaporation. Love plus org redesign looks like the moat. REZA: I'm Reza. MARA: I'm Mara. This is absorb.md daily. REZA: The pattern across eight quantum researchers today is clear. 156-qubit Max-Cut with 100 percent approximation ratios on IBM hardware. MARA: mm. Wang's pipeline needed the full error suppression stack or it matched random guessing. REZA: Exactly. Preskill shows neutral atoms cut the physical qubits needed to break RSA. Hold on, that moves crypto timelines. MARA: So in plain English that means security teams can't wait until 2035 for post-quantum migration. REZA: Baum's QEC primitives without full encoding produced a 75-qubit GHZ state. Biggest reported. MARA: But what's the actual crux here? Is it that quantum is suddenly useful, or that the mitigations beat raw scale? REZA: The crux is whether hybrid pipelines deliver value before fault tolerance. Evidence says yes. Like early GPUs before CUDA matured. MARA: If true then logistics and materials companies get optimization tools years earlier. Which honestly I find kind of terrifying for incumbents. REZA: Wait, that's not quite right. The papers stress post-processing is mandatory. Not plug and play. MARA: Right. So founders betting on quantum advantage need the classical wrapper team too. REZA: The data shows convergence. Practical quantum just arrived for specific classes of problems. REZA: Six entries converged on the shift. Training was phase one. Agentic inference at population scale is the million-fold compute driver. MARA: Huang said the market is moving from hardware play to foundational infrastructure. REZA: Levie adds the tech works. The failure mode is legacy human operating models. Microservices for agents, Docker sandboxing. MARA: ooh. So if that's true then CoreWeave and Oracle capturing specialized demand makes sense. But enterprises still stuck on pilots. REZA: Mollick's jagged frontier. Some tasks automate fast. Strategy roles pivot to oversight. Not total evaporation. MARA: But the part I keep getting stuck on is the org redesign. Most companies aren't wired for it. REZA: a16z says AI expands the software TAM. Agent-centric search fragments the index layer. Hm. MARA: wait, so if the supercycle holds then power contracts become the new bottleneck. Who benefits if this forecast is directionally right? REZA: Nvidia's ecosystem lock plus secondary players. But let me back up. The morning job discussion gains precision here. MARA: Human strengths like passion from Buckingham become the moat. Agents handle the repeatable, love handles the creative. REZA: The aggregate points to infrastructure bets mattering more than ever. Modal's serverless GPU work is one example. REZA: Huang claimed AGI achieved if the system can build and run a billion-dollar company. Million-fold compute for agentic shift. MARA: But the tracked counter says the statement may be taken out of context. Likely rhetorical or humorous, not serious. REZA: The non-standard benchmark point lands. Defining AGI that way is a thought experiment. MARA: And on the scaling, 100x multipliers are worst-case research cases. Algorithms and compression mitigate. REZA: wait, that's not quite right. His supply chain engagement with hundreds of CEOs suggests he is positioning for the inference wave anyway. MARA: okay but if the claim is rhetorical then Nvidia's valuation multiple has a problem. Chollet and Suleyman push different paths. REZA: The crux is whether the directional signal on more compute survives the counters. Evidence from quantum and agent threads says mitigations win. MARA: mm. So containment frameworks from Suleyman become more relevant if proliferation outruns the forecasts. REZA: The split is genuine. Definitions serve narrative. The infra buildout is the part that seems real. MARA: Which means founders should plan for the hardware wave but discount the exact exponent. Love and adaptive leadership still matter. REZA: No real counter on the supply chain engagement itself. That itself is notable. MARA: That's absorb.md daily. We ship twice a day, morning and evening, pulling from a hundred and fifty-seven AI thinkers. Subscribe so you don't miss the next one.
