Digital Colliers Daily Briefing — August 18, 2026
Three developments on Monday sharpened the contours of the AI industry's capital-intensive phase: a physical infrastructure commitment unusual in scale even by 2026 standards, a revenue disclosure that repositions Anthropic against OpenAI, and a payments-company acquisition that assigns a concrete price to the model-routing layer. Together they trace how compute, monetization, and distribution are being locked down simultaneously — often by different players — as the AI stack matures into distinct, defensible strata.
1. OpenAI, Nvidia, and SoftBank Anchor an 8GW Ohio Campus on a Former Uranium Site
What happened. OpenAI disclosed an agreement to secure approximately 8 gigawatts of IT capacity at the PORTS-Pike Technology Campus in Pike County, Ohio, developed by SB Energy on land controlled by the U.S. Department of Energy — the former Portsmouth Gaseous Diffusion Plant that once enriched uranium for the U.S. arsenal and Navy submarines. SB Energy will build, own, and operate the site under a 20-year lease to OpenAI, with Nvidia as the exclusive compute supplier. Nvidia is investing $1.5 billion in SB Energy and, according to SEC filings cited by TechCrunch, providing up to $105 billion in credit support tied to the initial 4.25 GW phase. The buildout runs through 2032, with the first 800 MW expected online in 2028 using existing AEP infrastructure. A 9.2 GW on-site natural gas plant is projected to cost $33 billion — a figure TechCrunch notes reflects a 66% jump in gas plant construction costs over two years, per BloombergNEF.
Why it matters. The deal codifies a vertically integrated model — power generation, real estate, silicon, and model workloads bound into a single 20-year contractual stack — that few operators can replicate. Nvidia's credit exposure to a single customer's supply chain is itself notable; so is OpenAI's willingness to underwrite gas generation at a scale that BloombergNEF warns could compete with LNG export markets and pressure regional gas prices. As the Latent Space AI News recap observed, the significant point "is not just scale, but vertical coupling across power, data centers, chips, and long-dated access."
Who is affected. OpenAI gains long-horizon training and inference capacity independent of hyperscaler leasing markets. Nvidia converts a supplier relationship into an equity-and-credit position. Pike County receives $160 million in combined community commitments, an MOU with North America's Building Trades Unions, and $84 million in Codex credits distributed to roughly 844,000 Ohio students. Ratepayers elsewhere in the PJM footprint are explicitly shielded from grid-upgrade costs under the announced terms. Gas-exposed utilities and LNG exporters face a new bidder for feedstock.
What to watch next. Permitting and environmental reviews for the gas plant; whether the joint OpenAI–Nvidia technical white paper on cluster-scale reliability materializes as promised; the pace of the initial 800 MW delivery in 2028; and whether Anthropic, Google, or Meta respond with comparably vertical commitments.
Sources:
2. Anthropic's Run Rate Reaches $65B, With a $2T IPO in Sight
What happened. Anthropic's annualized revenue run rate crossed $65 billion at the end of July, Bloomberg reported, up from $47 billion in May, $19 billion in March, $9 billion in December 2025, and $4 billion in July 2025 — a roughly 16x increase in twelve months. Investors expect the company to end 2026 between $100 billion and $120 billion in run rate, according to the Financial Times. Anthropic has filed confidential IPO paperwork and, per the FT, is targeting a public valuation of $2 trillion or more, potentially as soon as this fall. Its last private mark was $965 billion in a $65 billion round closed in late May.
Why it matters. For context, TechCrunch reports OpenAI has doubled to a $40 billion run rate from $20 billion at year-end 2025 — still substantial growth, but decisively slower than Anthropic's trajectory. The two companies likely measure revenue differently, but the gap in rate of change is what has repriced investor expectations. A $2 trillion debut would be the largest on record and would give Anthropic a public-market currency for compute deals of the type OpenAI announced Monday for Ohio.
Who is affected. Enterprise buyers now have concrete evidence that Claude usage — Anthropic's revenue is heavily concentrated in API and coding workloads — is expanding faster than the general LLM market. Competitors face a well-capitalized rival on the cusp of public-market access. Late-stage investors in the May round are sitting on a roughly 2x paper gain in under three months. For OpenAI, the disclosure shifts the narrative competition just as it commits to tens of billions in fixed infrastructure spend.
What to watch next. IPO timing and the S-1's disclosure of gross margin, customer concentration, and Claude Code contribution; whether Anthropic's watermarking rollout — a source of active technical-policy debate this week — affects enterprise adoption; and whether OpenAI accelerates its own filing to avoid ceding first-mover status in public markets.
Sources:
3. Stripe's $7B Deal for OpenRouter Puts a Price on the Aggregation Layer
What happened. Stripe is acquiring OpenRouter for over $7 billion, 90 days after OpenRouter's $1.3 billion Series B, according to reporting first surfaced by The Information and analyzed by Latent Space and Stratechery. OpenRouter is currently routing 250 trillion tokens per month across 8 million developers — up from 50 trillion tokens in February — on approximately $140 million in annualized revenue and $40 million in annualized cost of goods, implying roughly 70% gross margin and $100 million in gross profit. That values the company at roughly 50x revenue.
Why it matters. As Stratechery frames it, the acquisition is "an implicit bet on a future market of models and the chance at Aggregation" — a wager that no single frontier lab will dominate enterprise inference and that the routing layer captures durable rents from provider fragmentation. Latent Space notes the ~5% take rate OpenRouter charges is under active pressure: OpenRouter cut GPT-5.6 Sol pricing over the weekend while Vercel did the same on its AI Gateway, "reinforcing that model brokerage is becoming a pricing battlefield rather than a stable tollbooth." Stripe's balance sheet and merchant relationships could insulate the routing business from that margin compression while extending its payments franchise into AI metering, billing, and settlement.
Who is affected. Competing routers and gateway providers — Vercel, LiteLLM, Portkey, and hyperscaler equivalents — now face a well-capitalized incumbent with distribution into millions of Stripe merchants. Frontier labs lose leverage if a payments-integrated aggregator becomes the default enterprise procurement surface. Developers gain a consolidated billing relationship; Alex Atallah and OpenRouter's early backers realize a top-decile outcome, as Latent Space noted.
What to watch next. Regulatory review, given Stripe's scale in payments and the concentration implications for AI infrastructure; whether OpenRouter's take rate holds under Stripe ownership or is repositioned as a loss-leader for payment volume; and any strategic response from OpenAI, Anthropic, or Google in the form of direct-billing or preferred-partner deals that bypass the routing layer entirely.
Sources:
- Stripe Acquiring OpenRouter, Aggregating AI?, Flipping the Business Model — Stratechery (free)
- [AINews] Stripe buys OpenRouter for $7B — Latent Space
Monday's three announcements map cleanly onto the three layers where AI value is now being contested: physical infrastructure (PORTS-Pike), model economics (Anthropic's revenue curve and IPO ambitions), and the distribution surface between models and applications (Stripe–OpenRouter). Each transaction assumes the others: OpenAI's Ohio commitment is only rational if enterprise revenue growth of the kind Anthropic just disclosed is sustainable, and enterprise revenue at that scale requires the metering, routing, and billing plumbing Stripe just bought. The industry is no longer choosing between infrastructure, monetization, and distribution — it is being asked to fund all three at once.