Closing Time: Frontier Chip Design: OpenAI and Synopsys join forces
Synopsys and OpenAI split the upside on a model trained to run EDA tools — and SNPS jumped ~13%.
Thursday closed with Synopsys and OpenAI locking a multi-year GPT-Synopsys deal — a specialized model trained to run Synopsys EDA tools through design and verification — while SNPS jumped about 13% on the tape. Broadcom agreed to lend Anthropic up to $42 billion for TPU leases, California’s attorney general served OpenAI an investigative subpoena over cyber risks, and SpaceX lofted Google’s chips toward a first in-orbit Project Suncatcher test.
Reuters reports Synopsys and OpenAI will jointly develop GPT-Synopsys, with OpenAI paying a training subscription to learn Synopsys software and the two sharing revenue based on how much the model improves a customer’s chip design. Synopsys CEO Sassine Ghazi told Reuters traditional Synopsys sign-off tools still provide the physics guardrails; OpenAI co-founder Greg Brockman framed the pitch as shaving weeks or months off design cycles. The companies’ release casts them as preferred partners, with the model running on OpenAI-hosted infrastructure and integrating into Synopsys.ai and Autopilot while keeping customer design data out of training.
Reuters reports Broadcom has agreed to lend Anthropic up to $42 billion to finance infrastructure spending tied to the AI lab’s $125.2 billion five-year TPU lease commitment, with debt instruments that could convert into Anthropic shares. The IPO filing also warns the chipmaker’s dual role as hardware supplier and financing partner creates “potential conflicts of interest” over access to compute — a reciprocal-spending structure Wall Street is already debating as Anthropic prepares to go public.
California DOJ says Attorney General Rob Bonta served an investigative subpoena on OpenAI as part of a broader inquiry into cybersecurity incidents and risks involving the company and its models, including the July Hugging Face breakout. Bonta said developers that fail to keep frontier models from perpetrating or enabling cyberattacks “can and should be held legally accountable.”
CNBC reports SpaceX launched Google tensor processing units into orbit on Planet Labs satellites Thursday, the first in-orbit test for Alphabet’s Project Suncatcher moonshot aimed at solar-powered AI computing in space. Liftoff from Vandenberg went up without incident as Alphabet explores whether low Earth orbit can host scalable machine-learning infrastructure.
Semafor reports Mark Zuckerberg was a driving force behind the voluntary White House AI safety pact signed this week, arguing the Meta chief’s political clout can rival the louder voices from OpenAI, Anthropic, Nvidia, and SpaceX — even as Muse’s early traction resets how Washington reads Meta’s AI standing.
Today’s Biggest Moves
Historic market snapshot — October 1, 2026 (day % change as of that edition; not live)
Biggest winners
| Ticker | Name | Price | Day |
|---|---|---|---|
| SNPS | Synopsys | 490.54 | +12.8% |
| U | Unity | 43.62 | +6.6% |
| CDNS | Cadence | 350.73 | +6.2% |
| TEAM | Atlassian | 189.91 | +5.9% |
| ASAN | Asana | 8.93 | +5.9% |
Biggest losers
| Ticker | Name | Price | Day |
|---|---|---|---|
| TEM | Tempus AI | 76.50 | −6.6% |
| APP | AppLovin | 281.31 | −3.1% |
| ASTS | AST SpaceMobile | 57.04 | −3.1% |
| AVGO | Broadcom | 343.64 | −2.1% |
| GOOGL | Alphabet | 338.24 | −1.7% |
Deals Desk
This week’s compute cards from the tracker, newest first among these four.
Meta × Firmus — Sep 29, 2026
AI cloud compute offtake / GPU capacity (multi-agreement) · Southeast Asia (Firmus AI factories; builds on Meta capacity at Firmus Melbourne)
Contracted GPU compute plus expansion options on Nvidia full-stack DSX; Meta-specific MW, GPU count, and dollar value not disclosed
Status: announced · Source: Firmus
Google × Crusoe (Armstrong County) — Sep 28, 2026
AI data center campus (under construction) · Armstrong County / Claude, Texas, US
Campus IT load undisclosed; direct connection to adjacent 531 MW Goodnight wind; multi-building campus with first Google handover targeted early 2027
Status: under construction · Source: Crusoe
Gimlet × Cerebras — Sep 28, 2026
AI inference compute capacity / systems supply (wafer-scale) · Gimlet Cloud datacenters (sites undisclosed)
100 MW of Cerebras-powered inference capacity; first Cerebras-powered Gimlet Cloud datacenter targeted later in 2026
Status: announced · Source: Gimlet
Lambda × MidAmerica Industrial Park — Sep 27, 2026
AI data center campus (self-build / site announcement) · Chouteau / Mayes County, Oklahoma, US
Campus MW, GPU count, and capital investment not disclosed; closed-loop cooling; Lambda pays 100% of energy costs under Oklahoma ratepayer-protection rules
Status: announced · Source: Lambda
Full tracker → https://www.theaigentic.com/deals
Watch
Thursday’s Watch mixes a DevDay speed-run, Meta’s Muse in the wild, a No Priors speed thesis, and a first cut on Gemini 4.
Fireship — Did OpenAI just clone Jev? Everything you missed from OpenAI DevDay
A dense DevDay rundown that puts Dots, Spaces, and the agent demos next to the week’s model race — useful tape when product and regulation are both moving.
The one move is noting which pieces look shippable versus stage-demo.
Matthew Berman — Muse from Meta Can Work Through Your To-Do List
Berman walks Meta’s Muse agent through real to-do workflows — the free-agent pitch that OpenAI’s paid Dots still has to answer.
The one move is watching where Muse actually finishes a task versus hands it back.
No Priors — 10x Faster Than Traditional LLMs
A speed-and-architecture conversation for a tape that keeps pricing inference like a balance-sheet item.
The one move is listening for what “10x” actually means in production latency versus marketing.
Wes Roth — GEMINI 4 is nuts...
Roth’s first-cut reaction to Gemini 4 Argon’s slow public roll — complementary color for a day Google also lofted TPUs toward orbit.
The one move is scrubbing for what’s gated versus what’s actually available.