Hello world, it’s Tuesday, August 18th.
Nvidia has agreed to provide up to $105 billion in guarantees backing SB Energy’s PORTS-Pike campus in Ohio, a commitment that puts the chipmaker in a financial role well beyond its core hardware business. Anthropic reported a 14-fold jump in second-quarter revenue year-over-year, reinforcing the AI chip demand story, while Groq continues to raise money.
Let’s get into it. — Austin & Vik
Be sure to check out the Semi Doped podcast on YouTube or your favorite podcast player!
Nvidia Backs OpenAI Ohio Campus With $105 Billion Guarantee
Nvidia is taking on a financial role that stretches well beyond selling chips. The company has agreed to provide up to $105 billion in guarantees backing SB Energy’s PORTS-Pike campus in Ohio, which OpenAI will use as its primary AI compute facility. Nvidia is simultaneously taking a $1.5 billion equity stake in SB Energy, the SoftBank-linked data center developer building the project. CEO Jensen Huang told investors the campus could generate $600 billion in revenue over its lifetime, a projection that helps explain why Nvidia is willing to tie its balance sheet to the deal’s financing.
The arrangement gives Nvidia exclusive chip placement at the site in exchange for its credit support, per Digitimes, essentially swapping financial guarantees for guaranteed sales. That structure is what’s drawing scrutiny: Bloomberg reports Nvidia is actively calling on Wall Street to help finance the AI buildout as customer acquisition costs rise alongside chip prices. The circular-finance question, raised by Chosunbiz among others, is straightforward: Nvidia guarantees the debt that funds the infrastructure that buys Nvidia chips. The Ohio campus reportedly spans up to 10 gigawatts of capacity.
Anthropic’s 14-Fold Revenue Surge Anchors AI Chip Demand Before IPO
Anthropic disclosed to prospective investors that its second-quarter revenue jumped at least 14-fold versus the same period a year ago, with Bloomberg reporting the figure came in at more than $11.5 billion for Q2 alone. The disclosure, contained in documents seen by Bloomberg News, arrived as the company builds toward a public listing. Its annualized revenue run rate has now surpassed $65 billion, up more than sevenfold from its pace at the end of last year. That trajectory gave AI infrastructure investors something concrete to price: shares of chip and datacenter names edged higher in the US session, while EU chip stocks posted a sharper move on the news.
The revenue figures carry weight beyond Anthropic’s own valuation because they land directly against hyperscaler CapEx narratives, offering demand-side evidence that accelerator spending has a paying customer base behind it. Bloomberg’s Rachel Metz noted the figures invite direct comparison with OpenAI’s growth trajectory, framing the two as the clearest gauges of frontier-model commercialization. The IPO roadshow hasn’t formally begun, but the pre-marketing disclosure already has family offices and institutional investors recalibrating exposure to the AI infrastructure stack sitting beneath Claude’s revenue line.
Vik: We are about to witness the largest IPO in history pretty soon. So much will be known about the inner workings of Anthropic then. Looking forward!
Groq Raises $350M at Half Last Year’s Valuation, With Nvidia in the Round
Groq closed a $350 million Series A on Monday, led by Disruptive, with planned participation from Nvidia. The raise values the company at $3.5 billion. Add the $650 million from June and recent funding is $1 billion. That $3.5 billion is about half the $6.9 billion Groq commanded last September, a few months before Nvidia hired founder Jonathan Ross and other key people as part of a $20 billion licensing deal paid out to investors. Groq told TechCrunch it does not see a down round. It sees a new price for the post-licensing company.
Read the use of proceeds. The money is for “those seeking usage of medium and larger sized clusters of NVIDIA accelerated computing for training and inference.” Groq is an Nvidia Cloud Partner. It runs 13 data centers, claims more than six million developers, and wants to go from 54 megawatts to more than 200 megawatts in 2027. Alex Davis, Disruptive’s CEO and Groq’s executive chairman, said they are building “the world’s leading AI inference cloud.” The LPU story is not the fundraising story anymore.
This is the conversion of an inference-silicon challenger into another Nvidia-attached neocloud. Cerebras is the one with a named frontier model on wafers this month. Groq is raising to buy the stack it used to compete with. The open question is the same one CoreWeave already lives with: inference demand is real, and the returns on a leveraged GPU fleet are still a credit story, not a transistor story. (Bloomberg)
Vik: Crazy they are still raising money after Nvidia’s $20B acquihire!
Sector Watch
Foundry & Packaging
Sony and TSMC sign definitive 747 billion yen deal for a new Kumamoto sensor fab, with volume production targeted for 2029. (CineD)
Shinko Electric advances 22-layer glass substrate development as Samsung Electro-Mechanics faces ramp-up uncertainty on the same technology. (LEDinside)
Hanmi Semiconductor invests 130 billion won to build its largest-ever Plant 8 on a new Incheon site, expanding TC bonding capacity for AI packaging demand. (Chosunbiz)
Data Centers & Power
Trane Technologies and Eaton launch an industry-first integrated power-and-cooling reference design for AI data centers, co-developed with Nvidia, cutting installation costs and deployment time. (Business Wire)
Nebius switches its New Jersey AI data center from gas turbines to Bloom Energy fuel cells, citing faster permitting; Bloom also secures a $34M deal with LS Electric to power US data centers. (The Korea Times)
Blue Energy and GE Vernova Hitachi advance a 2.5 GW gas-plus-nuclear plant in Texas designed to supply up to 1.5 GW to co-located data centers. (Data Center Dynamics)
Compute
Groq raises $350M Series A at a $3.5B valuation, pivoting from chip ambitions to Nvidia-powered AI cloud infrastructure targeting 200+ MW of compute capacity. (Bloomberg.com)
Crusoe is in early IPO talks, according to Axios, as the data center developer seeks to capitalize on surging AI infrastructure demand. (Axios)
UCloud brings Hygon Tianxi AI accelerators to its public cloud platform, expanding domestic Chinese AI compute options. (TechNode)
Optics & Networking
Fabrinet reported record Q4 and full-year fiscal 2026 revenue, with annual sales rising 36% year-over-year. (Fabrinet)
Huawei open-sources the AscendNPU IR at the core of its BiSheng compiler, adding Triton and multi-language support for the Ascend 950. (Pandaily)
Ulvac shares plunge their most in over seven years after the chip vacuum-pump supplier issues an annual outlook that misses analyst expectations. (Bloomberg Tech)
Edge
Kulicke and Soffa appoints Dr. Raj Talluri as President and CEO, bringing a semiconductor product leadership background to the advanced bonding equipment maker. (Kulicke and Soffa)
Google confirms MediaTek will supply the modem for Pixel 11, calling the new chip faster and more power-efficient than its predecessor. (9to5Google)
L&T data center unit will host Nvidia hardware for Together AI, marking an early hyperscale-grade AI cloud deployment in India. (Data Center Dynamics)