Scale & Strategy
together with
This is Scale And Strategy, the newsletter designed to make you the smartest person at the water cooler.
Here’s what we got for you today:
- Pathway Just Challenged AI's Scaling Economics
-
Pathway Just Challenged AI's Scaling Economics
One of the first neolabs to put forward something tangible may have just shown a path to fundamentally changing the economics of AI.
On Tuesday, Pathway unveiled BDH-CQ, a 150-million parameter reasoning model, alongside benchmark results that suggest its post-transformer architecture can deliver comparable reasoning performance at a fraction of the cost and compute required by today's leading models.
The headline result came from ARC-AGI-1, one of the industry's most challenging reasoning benchmarks. BDH-CQ achieved 29.5% pass@2 accuracy, meaning it solved nearly a third of the benchmark's problems when given two attempts. More interesting than the score itself was the cost: Pathway estimates inference at just $0.0007 per task.
For comparison, OpenAI's GPT-5.6 Luna (Low), which received an 80% price reduction on July 30, scored 34.5% on the same benchmark. Despite the modest performance advantage, Luna's inference cost was still roughly 11x higher than BDH-CQ.
The significance isn't that Pathway has built a better model than OpenAI. It hasn't. The significance is that it may have found a dramatically more efficient way to generate intelligence.
BDH-CQ is small, doesn't rely on chain-of-thought reasoning, and requires less data due to improvements in memory. The result is a model that approaches the performance of much larger systems while consuming a fraction of the resources. If those advantages persist as the architecture scales, the implications are enormous.
"We need to be able to squeeze more intelligence per dollar, and for this you need to change the paradigm," Pathway CEO and co-founder Zuzanna Stamirowska told The Deep View. "This is a very deep innovation, and we wouldn't have done it if it wasn't going to be, and if it didn't have a chance to capture the market."
The company describes its breakthrough as a "PageRank moment for intelligence," drawing a parallel to the insight that transformed search from a keyword-matching problem into a system that leveraged the structure of the web itself.
That is an ambitious claim. But Pathway has assembled a team with enough credibility to warrant attention.
The company's advisors and leadership include:
• Łukasz Kaiser, co-author of the original Transformer paper that launched the modern AI era. He also independently verified Pathway's ARC-AGI-1 results.
• Alex Kurzok, former Group Product Manager for Gemini at Google DeepMind and now Chief Product Officer at Pathway.
• Jonathan Frankle, Chief AI Scientist at Databricks, who advises the company on scaling and deployment and is also an investor.
• Martín Farach-Colton, Chair of Computer Science and Engineering at NYU and an early Google engineer responsible for multiple foundational technical breakthroughs.
Pathway's ideas are not entirely new. In October 2025, the company published The Dragon Hatchling: The Missing Link Between the Transformer and Models of the Brain, a paper that quickly gained attention across the AI research community. The paper laid out the theoretical foundation for a post-transformer future. BDH, incidentally, stands for "Beautiful Dragon Hatchling," a reference borrowed from Terry Pratchett.
The release of BDH-CQ is the first meaningful attempt to validate that theory with real-world benchmark results.
There is still a long road ahead. A 150-million parameter model is nowhere near competitive with frontier systems. Pathway believes its architecture can eventually scale to 600-billion-parameter models capable of competing with the world's leading AI systems, but that remains unproven.
What is becoming increasingly clear, however, is that the industry desperately needs alternatives.
The scaling laws that have powered AI's progress over the last several years depend on ever-increasing amounts of compute, energy, and data. Those costs are becoming harder to sustain. Data center construction is exploding, power constraints are emerging globally, and model training costs continue to rise.
That is precisely why investors have poured more than $40 billion into over 40 neolabs pursuing alternatives to the transformer architecture.
Most will fail. A few may uncover entirely new approaches to intelligence.
Whether Pathway ultimately becomes the winner is almost beside the point. What matters is that the race to reinvent AI architecture is now producing tangible results.
Why it matters
For the past several years, AI progress has largely been purchased with larger clusters, more GPUs, more power, and more data. Pathway is betting that the next leap forward won't come from spending more, but from finding a fundamentally better architecture. If that bet proves correct, it could reshape the economics of AI as dramatically as the transformer reshaped the industry in 2017.
While data factories churn out quantity, leading AI labs need partners who co-own research goals and engineer the complex human-AI loops that push models from promising to state-of-the-art. Turing specializes in closing capability gaps through custom research acceleration.
Turing’s research-focused approach includes:
- Co-owned experimental outcomes, not just data delivery, and vendor neutrality
- Quality-by-design workflows with transparent data lineage and auditable results
- Custom RL environments and SFT/RLHF/DPO pipelines designed for your benchmarks
Partner with the research accelerator that understands what frontier AI labs actually need.
Anthropic Is Selling the Future IPO and Investors Are Starting to Ask Hard Questions.
Anthropic is heading toward what could become the largest IPO in history, but as the company prepares for a public debut as early as September, investors are no longer treating AI growth as a foregone conclusion.
According to The Wall Street Journal, Anthropic has been meeting with prospective investors in recent weeks to reinforce confidence in its business ahead of an IPO that could value the company near $1 trillion. The discussions have centered on a question that increasingly hangs over the entire AI industry: can today's AI leaders justify the enormous amount of capital being poured into them?
Investors have reportedly pressed Anthropic executives on several emerging risks, including the rise of low-cost Chinese AI models, tensions with the Trump administration, and growing public opposition to data-center construction across the United States.
Those concerns aren't unique to Anthropic. They're challenges facing the entire AI ecosystem.
The Chinese question is perhaps the most immediate. Models coming out of China continue to improve while undercutting Western competitors on price. Anthropic's response has reportedly been straightforward: focus on building the most capable models possible. The company's leadership has argued that users ultimately gravitate toward the best-performing systems, even if cheaper alternatives exist.
That may be true today. The question is whether that remains true if the performance gap continues to narrow.
Anthropic is also attempting to broaden its narrative beyond chatbots and coding assistants. In investor discussions, the company reportedly highlighted its ambitions in healthcare and biology, two areas where AI could deliver measurable real-world value and potentially soften growing public skepticism around the technology.
The timing is notable.
Public sentiment toward AI remains mixed. Businesses are rapidly adopting the technology, but concerns around energy consumption, job displacement, and the construction of massive data centers continue to grow. As AI companies consume increasing amounts of power and infrastructure, they are facing scrutiny from regulators, local communities, and policymakers.
That matters because the economics of the AI boom depend on continued expansion.
Trillions of dollars are being invested across data centers, power generation, networking infrastructure, and semiconductor manufacturing based on the assumption that demand for AI will continue to accelerate. If that assumption weakens, the ripple effects extend far beyond model developers.
Anthropic enters the public markets from a position of strength.
The company emerged as one of the biggest winners of the current AI cycle after Claude Code became a breakout success among developers. Earlier this year, Anthropic reported annualized revenue exceeding $47 billion, placing it among the fastest-growing software businesses ever created.
Recent infrastructure partnerships with companies including SpaceX and Google suggest demand remains strong. Usage of Claude products continues to grow, and periodic service outages indicate the company is still struggling to keep up with customer demand.
Yet public markets are less forgiving than private ones.
Private investors can justify almost any valuation when the future appears limitless. Public investors eventually want evidence that growth is durable, competition is manageable, and margins can withstand pressure.
Anthropic is about to become the first major test of whether the public markets are willing to place trillion-dollar valuations on AI model companies.
The outcome won't just affect Anthropic. It will influence how investors value every major AI company waiting behind it, including OpenAI.
Why it matters
For the last three years, AI companies have largely been valued on potential. Anthropic's IPO will be one of the first moments when public markets are asked to put a price on that potential. If investors embrace the story, it could unlock another wave of capital for the AI sector. If they don't, some of the assumptions underpinning the entire AI infrastructure boom may start to look a lot less certain.
Was this email forwarded to you?
That’s it for today and as always It would mean the world to us if you help us grow and share this newsletter with other operators.
Our mission is to help as many business operators as possible, and we would love for you to help us with that mission!
