Nvidia revenue doubles year over year as Huang declares "compute is revenue"

Nvidia reported 96.2 billion dollars in quarterly revenue on August 27, up 18 percent from the prior quarter and 106 percent from a year earlier, with data center revenue of 89 billion dollars. CEO Jensen Huang framed the quarter as a structural shift rather than a cycle. "AI has reached its inflection point. It's doing useful work. Its tokens are productive and profitable. Now, compute is revenue." He added that where a single lab drove the buildout a year ago, multiple frontier labs are now scaling in parallel alongside a growing open-model ecosystem, and that Vera Rubin is in full production. CFO Colette Kress guided to roughly 70 percent revenue growth in fiscal 2028, well ahead of consensus, and said the company remains supply constrained. She also defended Nvidia's practice of financing its own customers, including backing OpenAI capacity at the Ports Pike site in Ohio, saying "we expect them to become the largest technology companies in history" and that the risk is limited. The read for North American vendors is the supply constraint, not the growth rate. If Nvidia is the gating item, then facility readiness stops being the critical path and shifts to being the thing that must be waiting when chips arrive, which changes how buyers value schedule certainty over unit price.

Source: Data Center Dynamics (August 27, 2026)

AWS commits to deploying two million additional Nvidia GPUs through 2028

Amazon Web Services agreed to deploy an additional two million Nvidia GPUs across its global infrastructure through 2027 and 2028, announced August 27. The mix spans Blackwell Ultra, Rubin, and Rubin Ultra. AWS will also bring Nvidia Vera CPU-based infrastructure onto its platform, extend NVLink Fusion with custom high-bandwidth memory, and build AI factories for the US government running 100,000 GPUs on secure AWS infrastructure. Nvidia's Nemotron models will be offered through AWS, Amazon Robotics will adopt Nvidia's physical AI platform, and EC2 G7 instances will use RTX Pro 4500 Blackwell Server Edition GPUs, which AWS claims deliver 4.6 times the AI inference performance of the prior G6 generation. The scale-up is notable because AWS told GTC 2026 attendees it would bring more than one million GPUs online during 2026, and then found demand outrunning that plan. Matt Garman, CEO of AWS, said customers "want the freedom to choose the best tools for their AI workloads." For the North American supply chain, two million accelerators represent a multi-year pull on power distribution, liquid cooling, and structural capacity across existing AWS regions, and much of it lands as retrofit and densification inside buildings that already exist rather than as greenfield announcements.

Source: Data Center Dynamics (August 27, 2026)

Delaware becomes the first state to write a bring your own power clean energy mandate into law

Governor Matt Meyer signed a package of data center bills into law, reported August 27. The centerpiece requires hyperscale data centers to build their own clean generation to power their own facilities over a ten-year period, with a portion of backup power also drawn from clean sources. Delmarva Power, an Exelon subsidiary, will establish a separate and higher rate class for hyperscale data centers, and operators in that class must carry the costs of new transmission, distribution, and capacity procurement wherever possible. The laws also require large facilities to curtail demand during peak periods, and strip data centers of eligibility for job-creation tax credits. "Data centers that do not pay their fair share do not belong in our communities," Meyer said at the signing. The Sierra Club called Delaware the first state in the nation to pass a “bring your own energy” policy carrying clean energy requirements. Delaware currently hosts only about 19 data centers, most around Wilmington, but Starwood has filed for a 1.2-gigawatt, 11-building campus in New Castle. Delaware joins a lengthening list of states rewriting the rules, including Ohio, North Carolina, and Virginia, alongside new rates across TVA's service territory. For vendors, the practical effect is that on-site generation, storage, and load-curtailment controls are moving from optional scope into permit conditions.

Source: Data Center Dynamics (August 27, 2026)

NPR documents 54 nondisclosure agreements signed by Louisiana officials on data center deals

Reporting published August 27 by the Gulf States Newsroom and Type Investigations identified, through public records requests, at least 54 nondisclosure agreements signed by Louisiana elected officials on data center projects since Governor Jeff Landry took office in early 2024. The reporting centers on Delta Forge 1, a 3.6-billion-dollar, 300-acre Applied Digital campus now under construction in Boyce, Rapides Parish, a town of roughly 1,000 people. Phase one draws 300 megawatts, roughly the consumption of 200,000 homes, and initial operations are possible in mid-2027. The project also carries a state sales tax exemption on equipment. Negotiations ran under the code names Project Lightning and Project Pixel, with NDAs signed by at least five state senators, the parish assessor, and every member of the England Economic and Industrial Development District, which approved the deal on April 23. Applied Digital CEO Wes Cummins acknowledged that NDAs "feel sneaky" but cited securities law obligations around material non-public information. Researcher Eric Bonds found NDAs in at least 25 of 31 Virginia projects surveyed. The countertrend matters more than the complaint. West Feliciana Parish President Kenny Havard declined NDAs on a 10 billion dollar project and published every agreement, and bills to ban data center NDAs are now moving in Michigan, Oklahoma, Kentucky, and Ohio, while Pennsylvania Governor Josh Shapiro has barred their use by agencies he oversees. Disclosure is becoming a siting variable, and vendors bidding into code-named projects should expect the schedule to carry political risk that does not show up in the RFP.

Source: NPR (August 27, 2026)

Georgia regulators approve 3.2 GW of new generation for OpenAI's Effingham County campus

The Georgia Public Service Commission approved a power supply contract between Georgia Power and OpenAI on August 27, covering the company's planned campus in Effingham County near Savannah. Georgia Power will develop 3.2 gigawatts of new generation, and OpenAI has committed to paying the full cost of the infrastructure required to serve the site. The generation type has not been disclosed. OpenAI will also provide up to 1 gigawatt of flexible demand response, and Georgia Power expects to begin delivering energy between 2028 and 2032. The project, known as Project Camellia, sits inside the Savannah Gateway Industrial Hub, a site of more than 2,600 zoned industrial acres. Four buildings are planned across 4.4 million square feet, with OpenAI expected to invest more than 30 billion dollars and to use a closed-loop water system that does not draw from local supply. Effingham County currently hosts no data centers, with Georgia's existing footprint concentrated around Atlanta and Augusta. "Data centers are paying more so families and small businesses can pay less," said Kim Greene, chairman, president and CEO of Georgia Power. The structure is the story. Full cost recovery from the customer plus a gigawatt of dispatchable curtailment is quickly becoming the template a commission needs to see before it will approve load of this size.

Source: Data Center Dynamics (August 27, 2026)

Four more data center buildings filed at Cedar Creek in Bastrop County, Texas

Four new applications were filed with the Texas Department of Licensing and Regulation for land at Cedar Creek in Bastrop County, outside Austin, reported August 27. Two cover 577,000-square-foot single-story buildings at 6543 FM 535, designated EDCAUS31 and EDCAUS32, each carrying 700 million dollars of investment with construction running from October 2026 to December 2028. Two more, EDCAUS13 and EDCAUS14, are planned at 6682 FM 535 at 730,000 square feet each, also at 700 million dollars apiece, running from September 2026 to March 2029. The filings do not name the developer, but the land has been tied to EdgeConneX in local reporting and the naming convention matches known EdgeConneX projects. EdgeConneX filed EDCAUS11 and EDCAUS12 at the same 6682 FM 535 address in June on identical terms. The nearby AUS01 site occupies 130 acres and is slated for 96 megawatts across 920,000 square feet by 2027, with potential to exceed 336 megawatts in later phases, and some capacity is leased to CoreWeave. Full build-out across the area could exceed a dozen buildings on more than 1,500 acres. This is worth watching because it is one of the larger sustained multi-building programs in Central Texas, and it is being executed through incremental permit filings rather than headline announcements. The vendor opportunity therefore arrives in tranches, and it rewards whoever is already qualified on the earlier phases.

Source: Data Center Dynamics (August 27, 2026)

Anthropic reportedly signs a $45 billion, 460 MW capacity agreement with Nscale in West Virginia

Anthropic has signed a 45 billion dollar AI cloud capacity agreement with Nscale, reported by Bloomberg on August 26 and covered by DCD on August 27. The arrangement leases 460 megawatts for six years at Nscale's Monarch Compute Campus in West Virginia, running Nvidia Vera Rubin chips expected online late next year. Nscale has not confirmed the contract publicly, and the figures come from people familiar with the agreement rather than from either party, so treat the number as reported rather than filed. Nscale acquired the 2,250-acre Monarch site in March 2026. The campus has potential for up to 8 gigawatts, and its first phase of 2 gigawatts is targeted for the first half of 2028, powered by Caterpillar G3500 series fast-response natural gas generator sets. Microsoft has separately signed for 1.35 gigawatts at the same site using Vera Rubin NVL72 systems. Nscale is preparing a US IPO reported to target roughly 3 billion dollars. Anthropic has now committed to renting more than 10 gigawatts of server capacity across providers including Google, AWS, CoreWeave, Akamai, and Fluidstack. West Virginia has not historically been a primary market, and a single campus of this scale will pull specialty contractors, gas handling, and switchgear capacity into a region with a thin local bench.

Source: Data Center Dynamics (August 27, 2026)

Rum Group discloses a lease worth up to $13.7 billion at its Maysville, Georgia campus

Rum Group, formerly Rumble, disclosed in an SEC 8-K filing a capacity lease with an unnamed US-based third-party cloud customer for GPUs and GPU services at its Maysville, Georgia site, reported August 26. The customer will purchase in three tranches under a six-year commercial agreement with a total contract value of up to 13.7 billion dollars, with the third tranche contingent on the customer approving the delivery schedule. The agreement also grants warrants for up to 50.81 million Class A shares at one cent per share, vesting against purchase volume. The Maysville campus was originally announced by Northern Data in December 2024 at 120 megawatts, since expanded to 180 megawatts, with operations targeted for the first quarter of 2027. Rum Group rebranded after acquiring 85 percent of Northern Data in June 2026 and renamed the cloud business Quake AI. Two cautions belong with the headline number. The counterparty is not disclosed, and the company reported a net loss of 79.1 million dollars in the quarter ended June 30 against 40.4 million dollars of revenue, so this is a contract announcement rather than delivered capacity. Still, 180 megawatts in northeast Georgia with a signed offtake is a real pipeline event for anyone selling into that corridor.

Source: Data Center Dynamics (August 26, 2026)

Volato Group to merge with Alignment Engine, converting an aviation software firm into an Ohio data center developer

Volato Group signed a definitive merger agreement on August 28 with AI infrastructure company Alignment Engine, in a transaction valuing the target at 500 million dollars. The deal repositions Volato from aviation AI software into AI infrastructure, high-performance computing, and data center development, built around Alignment Engine's Ohio campus. The asset is 154 megawatts of current capacity with a near-term path to 480 megawatts, plus GPU compute, networking hardware, and proprietary technology including a DPU the company calls NeuralSync. "We have a powered industrial campus and 154 megawatts available today," said Chris Ensey, CEO of Alignment Engine. Alignment Engine, founded in 2021, runs AMD hardware exclusively and plans to deploy AMD Helios rackscale systems with Instinct MI455X GPUs in Ohio during 2027. Read the balance sheet alongside the headline. Volato's quarterly report published August 10 flags substantial doubt about going concern status, with a net loss of roughly 4.7 million dollars for the six months ended June 30 and an accumulated deficit near 105.5 million dollars. The pattern is worth tracking regardless. Powered land is being reverse-merged into public shells as a financing route, and a comparable deal was recently struck between Healthy Choices Wellness Corp and Host Digital Infrastructure. Expect more of these, and expect the megawatts to be real while the timelines stay soft.

Source: Data Center Dynamics (August 28, 2026)

Reciprocating gas gensets as prime power, not standby: the Monarch campus model

Buried in this week's Anthropic and Nscale coverage is a detail that deserves more attention than the dollar figure. Power at Nscale's Monarch Compute Campus in West Virginia comes from Caterpillar G3500 series fast-response natural gas generator sets, and the campus is planned at up to 8 gigawatts with a 2 gigawatt first phase targeted for the first half of 2028. Reciprocating engines have always lived in the data center as standby assets, sized for a few hours a year, commissioned to start and carry load during an outage, and otherwise idle. Running them as the primary source for a multi-gigawatt AI campus is a different engineering problem in almost every respect. Duty cycle goes from tens of hours annually to continuous. Fuel supply shifts from an on-site diesel tank to a firm pipeline contract with its own interruption risk. Emissions permitting moves from a nominal standby exemption to a real operating limit, which is exactly the category of minor-source permit now under debate at the federal level. Maintenance becomes a scheduled production activity with its own spare parts logistics rather than a quarterly test. Heat rejection, exhaust routing, noise attenuation, and vibration isolation all get designed to residential-adjacent standards because these sites sit near communities that are increasingly organized.

Why it matters for North American builds: the appeal is speed. An engine plant can be ordered, delivered, and energized on a timeline measured in quarters while an interconnection request in a mature market can run past five years, and last week's Global Energy Monitor data showed reciprocating engine capacity tied to data center projects tripling to 45 gigawatts. That gap is now the single largest driver of site selection in this sector. For vendors, it opens scope that did not exist in the traditional utility-served model. That scope includes gas conditioning and metering, selective catalytic reduction and emissions monitoring, paralleling switchgear rated for continuous operation, black start and load-following controls, engine cooling loops kept separate from the IT thermal system, and a commissioning regime that has to satisfy both the offtaker and the air permit. The firms that will win this work are the ones that can speak fluently to a power plant operator and a data center operator in the same meeting, because on these campuses those are the same customer.

Source: Data Center Dynamics (August 27, 2026)