2026-09-25 — Akamai’s Anthropic-deal surge fades to a 3% gain at the close

• Akamai shares closed up about 3% Friday, a fraction of the 21.3% surge they showed in premarket trading on Thursday’s $11.6 billion, seven-year Anthropic cloud-services deal. CNBC’s closing report named Akamai a key winner of the session; the deal, potentially expanding to about $20 billion, carries a warrant for Anthropic to buy up to roughly 5% of Akamai. (CNBC live updates)
• US core capital-goods orders, non-defense excluding aircraft, jumped 1.6% in August against a 0.5% forecast, after July was revised up to +0.6% from unchanged, Reuters reported Friday. Core orders were up 10.6% from a year earlier, orders for computers rose 1.5% (up 20.1% year on year) and communications equipment 0.3% (up 35.8%), and shipments of core capital goods rose 0.6% after 1.4% in July; economists said the figures point to another quarter of robust business equipment spending amid the AI infrastructure buildout, although the headline durable-goods total was flat. (Reuters, via Lufkin Daily News, U.S. Census Bureau)

2026-09-24 — A BlackRock-backed group is in talks for Stack’s Asia-Pacific data centers at up to $25 billion

• A consortium backed by BlackRock and IFM Investors is in exclusive talks to buy Stack Infrastructure’s Asia-Pacific data-center portfolio for up to $25 billion, Bloomberg reported Thursday, citing people familiar with the matter. The group includes the BlackRock-backed AI Infrastructure Partnership, is preparing due diligence and hopes to agree terms soon with Stack’s owner, Blue Owl Capital; Bloomberg reported in May that Stack was weighing a sale of its Asia operations at more than $30 billion. Reuters could not immediately verify the report, and talks could be prolonged or fail. Stack operates in Tokyo, Osaka, Sydney and Melbourne. (Economic Times, from Reuters)
• Akamai announced an $11.6 billion, seven-year commitment from Anthropic to run its CPU workloads on Akamai Cloud, potentially expanding to $20 billion total, with Akamai issuing Anthropic a warrant for up to about 5% of its common stock Akamai said capital expenditure tied to the commitment will run about $5.5 billion, with roughly $1.7 billion added to 2026 capex to secure supply-chain components including memory, and that the deal will not affect 2026 revenue guidance. (Akamai, via GlobeNewswire)
• Akamai’s 8-K filing on its $11.6 billion Anthropic compute deal shows the underlying agreement — two new project plans under an existing May 5 master services agreement — was signed September 18, 2026, five days before the September 24 public announcement, and details a warrant covering 387,051 shares of Series B Non-Voting Convertible Preferred Stock at $2,226.60 per share The warrant vests in four tranches: 40% on Anthropic’s first payment under the newest project plan, then 20% for each additional $3 billion of contractual value Anthropic commits, and remains exercisable for seven years from issuance; exercise must be settled in cash. (SEC EDGAR)

2026-09-10 — Microsoft’s own destination number: 12GW today, 38GW by 2032

• Bloomberg reported (09-10) that Microsoft plans to more than triple its global data-center footprint to more than 38 gigawatts by 2032, from roughly 12GW today, with AI-specific capacity growing from ~2GW of that current 12GW to about a third of the 38GW target — the single largest named-company destination figure this thread has logged, larger than any individual site tracked here so far. Reported, not filed: Microsoft has not confirmed the figure in any release or on a call. It reads directly against this thread’s 09-01 Reuters entry on the national capacity-committed-vs-energized gap (700GW+ of connection requests against ~10% actual current draw) — one hyperscaler alone is now guiding to a physical build nearly as large as the entire current installed US data-center base, which is the destination-side scale this thread’s watch line exists to track. (Bloomberg, via Global Data Center Hub, Seeking Alpha)

2026-09-01 — Reuters puts a hard number on the capex-committed-vs-capacity-energized gap, nationally

• A Reuters analysis of utility and grid data found electricity connection requests from large power users — overwhelmingly data centers — now exceed 700GW across the Midwest, Mid-Atlantic and South, roughly ten times current actual US data-center power consumption, the clearest national quantification yet of the gap this thread has tracked since the 08-03 Texas audit-and-freeze entry. Texas’s own data-center grid-connection requests grew from ~48GW (2023) to 474GW+ (matching this thread’s earlier figure); separately, ten major utilities including AEP Ohio, Southern Co. and PPL have logged roughly 270GW of data-center demand requests combined. Pennsylvania’s own count is the sharpest single data point: of 100+ proposed data centers, only 20 have applied for the permits needed to actually proceed — after Governor Josh Shapiro signed an 08-18 executive order tightening permitting for projects 25MW and up, joining Texas as a second state now gatekeeping the queue rather than taking every announcement at face value. Texas PUC Chairman Thomas Gleeson, quoted: “When you don’t know what is real, you really don’t know how to build the infrastructure for it.” (Reuters, via Yahoo Finance, Investing.com mirror)

2026-08-26 — AWS triples its GPU order; the chip destination gets a number attached

• AWS and Nvidia announced AWS will deploy 2 million additional Nvidia GPUs across 2027-2028, on top of the 1M+ GPU commitment AWS made at GTC 2026 (already run through early on demand) — plus new Vera CPU-based infrastructure, “some integrated with Rubin, others standalone” per Nvidia CFO Colette Kress. Nvidia’s own Q2 earnings release the same evening put its total order backlog above $2 trillion and raised its aggregate top-five-hyperscaler capex forecast to ~$1.3T for 2027 (from ~$800B in 2026) — see Hyperscaler Capex for the spender-side framing. For this thread’s destination framing: this is the “chips” line item of the ~$750B question getting a concrete, dated, multi-year commitment from a single named buyer. (NVIDIA — primary, TechCrunch)
• Georgia Power announced its 3.2 GW Camellia/OpenAI contract cleared PSC regulatory review at 9:00pm ET on 08-26 — one of this thread’s tracked destinations (power + land) getting its last-flagged regulatory gate resolved. OpenAI funds its own infrastructure; the deal is framed as netting ratepayers ~$950M/year starting 2029. ⚠️ Whether this was affirmative PSC approval or a lapsed objection deadline is unconfirmed — detail on Camellia. (Georgia Power, via PR Newswire — primary)

2026-08-25 — An $11.3bn Tennessee campus files, with power attached

• Backers of “Project River” filed rezoning and annexation applications for an $11.3bn Bradley County, Tennessee campus pairing a large data center with on-site natural-gas and biomass generation. Public forums are set for 09-01 and 09-12 before any vote. The bring-your-own- generation pattern routes around the interconnection queue that has been the binding constraint at most of this thread’s other sites. (Chattanoogan)
• ✏️ Correction, 2026-08-30: “pairing a large data center with” generation is more confident than the developer itself is being. Wright Brothers CEO Mitchell Simpson told the Cleveland Banner on 08-28 that there is no confirmed off-taker for the campus’s power — “We do not have somebody that is definitely the off taker for that power at this time” — and named advanced manufacturing, semiconductor production and robotics as alternative end uses. What is filed is a generation-first campus looking for a load, not a data center that brought its own power, which is the inverse of the pattern this entry read it as. The forums stand at 09-01 (18:00-20:00 ET, Cleveland PIE Center) and 09-12.

2026-08-22 — Alibaba raises $10.2bn in Hong Kong’s largest-ever follow-on, and puts 100% of it into AI

• Alibaba launched a HK$80bn ($10.2bn) share placement — 710 million ordinary shares at HK$112.70, a 3.6% discount to the last close — and says 100% of net proceeds go to its “full stack” AI capabilities: chips, infrastructure, and model development and deployment. It is the largest primary follow-on offering ever by a Hong Kong-listed company and the world’s third-largest of the year, after Alphabet and Intel. Bookrunners: CICC, HSBC, Morgan Stanley and UBS. (Bloomberg, Reuters via Investing.com)
• The financial context is what makes it a capital story rather than a product one: Alibaba’s June-quarter profit fell more than 75% to RMB 10.5bn (~$1.6bn) with a $6.6bn free-cash outflow, which the company attributes to the rising cost of AI projects and computing infrastructure. A firm burning cash on AI is funding more AI by selling equity at a discount — the public-market version of the vendor-financing loop this map tracks on the US side, with the dilution taken openly instead. (Bloomberg)
• The earliest verified report of this is the Financial Times at 08:33 UTC on 08-23 (04:33 ET). It is bucketed to 08-22 on that basis, and cross-referenced on the 08-23 page. The exact HKEX filing time was not established; if it is later pinned past 05:00 ET, this belongs on 08-23.

2026-08-17 — A named destination gets a physical shape: Pike County, Ohio, 8 IT-GW, built in phases

• The signed PORTS-Pike Technology Campus deal (Nvidia/OpenAI/SB Energy) gives this thread’s Ohio site a physical shape and a timeline: an initial 4.25 IT-GW tranche under Nvidia’s credit support, with an option on a further 3.75 IT-GW, for a full 8 IT-GW ceiling with OpenAI named as the customer for all of it. Reaching it requires SB Energy and SoftBank to build at least 10 GW of new power generation and invest at least $4.2B in regional grid infrastructure via AEP Ohio. (NVIDIA, OpenAI)
• 💡 This is the capex-to-capacity conversion the thread has been mapping since the 08-03 Texas gate entry, spelled out here as a phased build with a stated first tranche — 4.25 of 8 IT-GW committed outright, the rest optioned — rather than the single go/no-go moment Texas’s audit-and-freeze imposes elsewhere on this map.

2026-08-15 — The gate produces its first compliance case, not just its first casualty

• Texas Governor Abbott announced that Core Scientific, Vantage Data Centers and SB Energy have committed to comply with his data-center standards (pay for their own electric infrastructure rather than shifting costs to ratepayers, reuse their own water, lower Texans' electricity costs, avoid disrupting residential neighborhoods, and drop reliance on taxpayer-funded incentives) — the same PUCT/ERCOT audit-and-freeze process flagged here 08-03. Abbott’s office explicitly framed AWS’s Lusby, MD withdrawal (logged below, 08-04) as the counter-example: a project that ended rather than complied. (Texas Governor’s Office)
• 💡 The gap this thread has tracked since 08-03 now has both of its outcomes on the record: AWS’s withdrawal showed capex that didn’t convert to capacity; this is the first evidence the other direction exists too — capex clearing the gate rather than falling into the gap. Vantage’s stake here is its $25B, 1.4GW “Frontier” campus in Shackelford County (full detail on Datacenter Sites), so this is a large, named site, not a marginal one.

2026-08-04 — The capex-to-capacity gap this thread flagged 08-03 gets its first concrete casualty

• AWS formally withdrew its application for a 2,000-acre, ~500MW data center campus in Lusby, Maryland (on Constellation Energy land next to the Calvert Cliffs nuclear plant) after local opposition over transparency and a commissioner’s signed NDA — filed May, dead by August, no reason given publicly. (Southern Maryland Chronicle)
• 💡 This is the gap made concrete, not just structural: the 08-03 entry below flagged that Texas’s audit-and-freeze could make committed capex and energised capacity diverge, with “no node for that gap yet.” Separately, PJM (the mid-Atlantic/13-state grid operator, unrelated to Texas’s ERCOT) rolled out its own version the same week — a “bring-your-own-power-or-get-cut-off” rule pushing data centers to self-supply or risk curtailment starting June 2027 (full detail on the Grid & Turbines child thread). Two different grids, same instinct: hosts are now pricing in the possibility that a announced site never energises. AWS’s withdrawal is the first actual node for that gap — a capex-committed site that didn’t convert to capacity, for reasons unrelated to power availability at all.

2026-08-03 — The destination tree acquires a gatekeeper

• Texas ordered an audit-and-freeze on new data-centre grid connections, directing PUCT and ERCOT to vet every queued project before it can energise and to deny connection to any that fails. (Texas Tribune)
• 💡 Meta-thread read: this thread maps capex flowing down into compute, power and sites. Every leaf has assumed the money converts into physical capacity at whatever rate the money arrives. Texas — holding a queue of 474+ GW, ~90% data centres, over five times its own peak-demand record — has just inserted a discretionary approval step between the two. Capex committed and capacity energised can now diverge, and this map has no node for that gap yet.

2026-07-24 — Opened (meta-thread)

• Opened as the destination cut of AI capex — gathers the clear destination-projects (Stargate, Colossus, Jalapeño) and frames the gaps (the chip-spend itself, power/nuclear-for-AI deals, the specific sites) as the collection worklist.

← Backstory