Photo by Laura Ockel on Unsplash
500,000 tokens per second. That single performance figure — 20 times what an Nvidia H100 server delivers on the same benchmark — is what Etched revealed when it emerged from stealth on June 30, 2026. But the throughput claim isn't the most interesting part of this story. Reporting by TechCrunch, amplified by Google News on July 3, 2026, surfaced a more instructive detail: how the investor who secured an early stake got in the door. The answer involves an Arizona state board seat, a TSMC fab commitment, and a venture playbook that coastal investors structurally cannot replicate.
The thesis in one sentence: geographic proximity to semiconductor manufacturing has become a structural investment moat, and Thiel Capital's Jack Selby has been building it quietly.
What Just Happened
As of July 3, 2026, Etched has raised $800 million in total funding — including a $500 million round led by Stripes in December 2025 — reached a $5 billion valuation, and announced over $1 billion in signed customer contracts, according to TechCrunch's original reporting. The company's first product, the Sohu chip, was manufactured by TSMC at its Arizona facility using a 4nm process in early 2026.
Jack Selby, managing director at Thiel Capital and founder of Copper Sky Capital (formerly AZ-VC), secured his early position by leveraging his seat on the Arizona Commerce Authority board. That board role provided visibility into Etched before the company became a competitive bidding round — and critically, Selby could credibly offer to help route production to TSMC's Arizona fab. For a hardware startup where manufacturing access is existential, that promise carries more weight than a term sheet improvement of 50 basis points.
Copper Sky Capital is now raising a $300 million second fund after fully deploying its first $115 million fund, expanding focus from Arizona startups to all of North America. Selby's investment thesis is direct: coastal startups in California, Massachusetts, and New York are, in his framing, grossly overpriced compared to companies emerging in manufacturing-dense regions.
The Pattern — Geographic Arbitrage as the New VC Wedge
The mechanism Selby is running isn't novel in principle — local investors have always had sourcing edges. What's changed is the asset class where that edge now compounds hardest: AI hardware, where supply chain access has become as strategic as capital allocation.
As of July 3, 2026, Arizona has attracted over $200 billion in private semiconductor investment over the past five years, securing 40-plus semiconductor expansions representing $102 billion in capital and more than 15,700 direct jobs since 2020. TSMC's Arizona fab generated $514 million in profit in its first full year of mass production — and Q1 2026 earnings alone surpassed the entire 2025 total. TSMC has since committed to a total of $165 billion in Arizona investment, expanding its original $12 billion commitment to cover three additional fabs, two advanced packaging facilities, and a major R&D center. Arizona separately recorded $34 billion in new investment and 27,749 projected new jobs across semiconductors, aerospace, and AI sectors in 2025 alone.
That manufacturing density creates a sourcing moat that a Palo Alto fund cannot buy. A Sand Hill Road partner can write a $10 million check; Selby can introduce a hardware team to fab operations leadership before the term sheet closes. For a company like Etched — which needed TSMC's Arizona line running before it had any public profile — that introduction has direct, quantifiable value.
This dynamic echoes what Smart Startup Scout flagged in covering OpenAI's government stake maneuvering: the competitive moats in AI infrastructure increasingly run through physical geography and policy access, not just model capability or capital stack.
Chart: Inference throughput comparison — Etched Sohu versus Nvidia H100 and Blackwell B200 on the Llama 70B model, per Etched's own benchmarks reported at stealth exit.
Photo by Qinru Xie on Unsplash
What the $5 Billion Valuation Actually Signals
Etched's valuation requires belief in several compounding assumptions: that transformer architectures remain dominant long enough for a hardwired ASIC (application-specific integrated circuit — a chip designed to execute one class of operations rather than general computation) to recoup its R&D, that hyperscalers shift inference workloads away from Nvidia meaningfully, and that Etched can manufacture at scale fast enough to honor its contract backlog.
The bull case has genuine data behind it. As of July 3, 2026, AI chip startups raised $8.3 billion globally in 2026, including a $1 billion round for Cerebras in February and $500 million rounds for MatX, Ayar Labs, and Etched itself. The capital is chasing a real demand signal: hyperscalers — Microsoft, Meta, Amazon, and Google — collectively represent an estimated 40-50% of Nvidia's revenue but are actively developing custom inference chips to reduce their dependence. Industry analysis cited in reporting on the sector suggests inference could represent 80% of long-term AI compute demand. Nvidia currently controls 80-90% of the high-end AI chip market. That's the gap Etched is pricing itself to fill.
Nvidia's response signals it takes the threat seriously. The company acquired Groq's inference assets for $20 billion in December 2025 and invested $4 billion in two photonics technology companies in March 2026 — defensive moves that simultaneously validate the inference opportunity and raise the competitive bar for challengers.
In my analysis, the most revealing number here isn't the $5 billion valuation or even the 500,000 tokens-per-second claim — it's the $1 billion in signed customer contracts from a company that had zero public profile until June 30, 2026. That figure, if it converts to recognized revenue, puts a floor under the valuation that most pre-revenue hardware startups never reach. The unit-economics sniff test worth running: which customers are actually in those contracts? Hyperscaler logos validate the inference thesis at scale. Mid-market AI companies suggest a more constrained total addressable market than the headline implies.
The Founder Move for Q3 2026
The Etched story carries concrete signals for founders building in AI hardware, enterprise software with significant inference cost exposure, or any vertical where GPU spend is becoming a material operating line.
Selby's access to Etched came from board-level proximity to economic development infrastructure — not from pitch competitions or cold outreach. If your hardware or AI infrastructure startup depends on manufacturing access, identify who sits at the intersection of regional venture capital and state-level industry boards before your Series A. Texas (Samsung's Taylor fab), Ohio (Intel's announced facilities), and upstate New York (GlobalFoundries) are building gravity wells similar to Arizona's. A warm introduction from a local LP network compounds differently than an email blast to Menlo Park.
The hyperscaler shift toward custom inference chips — and the $8.3 billion in AI chip startup investment raised globally in 2026 — signals that GPU spend is a line item CFOs are actively targeting. If your product runs on expensive inference, build a defensible cost model and present it proactively. An ICP-fit conversation (identifying the exact customer profile who has the most to gain from your solution) anchored to compute spend opens doors that capability demos do not. Buyers who are already paying Nvidia prices for inference are primed to evaluate alternatives.
Etched's entire bet is on transformer architecture remaining the dominant paradigm. The company's ASIC hardwires transformer operations into silicon — if model architectures shift materially, that's a strategic liability, not just a technical one. Any founder building inference-optimized infrastructure — hardware or software — should be able to articulate a three-to-five-year thesis for why their architectural assumptions hold. Investors writing checks into AI hardware in this environment are implicitly making that same bet, and they will ask.
Frequently Asked Questions
Who is Jack Selby and what is his connection to Peter Thiel?
Jack Selby is managing director at Thiel Capital, the investment vehicle associated with PayPal co-founder and venture investor Peter Thiel. Selby also founded Copper Sky Capital (formerly AZ-VC), a regional fund that fully deployed its first $115 million fund and is currently raising a $300 million second fund with expanded North American focus. His early stake in Etched came through his Arizona Commerce Authority board seat — not through Thiel Capital's broader deal network — illustrating how regional civic infrastructure creates proprietary dealflow unavailable to coastal-only investors.
What does Etched's Sohu chip actually do and how does it compare to Nvidia H100 performance?
Sohu is an ASIC purpose-built for transformer model inference — the process of running an already-trained AI model to generate outputs. As of July 3, 2026, Etched claims a single Sohu server processes over 500,000 tokens per second on Meta's Llama 70B model, versus 23,000 tokens per second on an Nvidia H100 server and approximately 45,000 tokens per second on Nvidia's newer Blackwell B200. The tradeoff is architectural inflexibility: Sohu is hardwired for transformer workloads and cannot execute other AI model types the way a general-purpose GPU can. That's a bet, not just a design choice.
Is Etched a good investment compared to Nvidia stock in the current AI chip market?
This article does not constitute financial or investment advice. From a purely analytical perspective, Etched and Nvidia represent fundamentally different risk profiles within an investment portfolio. Nvidia is a publicly traded company with 80-90% market share in high-end AI chips and established revenue. Etched is a private company at a $5 billion valuation with $1 billion in signed contracts — not recognized revenue — competing against an incumbent that spent $20 billion acquiring Groq's inference assets in December 2025 and $4 billion on photonics investments in March 2026. Private hardware startups carry liquidity, execution, and architectural risk that public equities do not. Any investor should evaluate within the context of their own financial planning and risk tolerance, ideally with qualified professional guidance.
Disclaimer: This article is editorial commentary for informational purposes only and does not constitute financial or investment advice. All figures are sourced from publicly reported data. Research based on publicly available sources current as of July 3, 2026.