On August 10, Anthropic, Macquarie Asset Management, and GIC announced Theseus Infrastructure, a new platform built to develop, own, and lease dedicated data centers to Anthropic under long-term agreements. Macquarie-managed funds and GIC (Singapore’s sovereign wealth fund) own the platform and fund most of the equity. Anthropic is the anchor tenant, starting in the US. Read the press release on its own and it sounds like just another infrastructure headline in a year that’s had roughly one a month.

It isn’t, not quite. Theseus is the seventh entry in a string of Anthropic capacity deals announced since November 2025, and it’s the one that finally puts a name and a dedicated ownership structure on what the other six were dancing around: Anthropic doesn’t have enough compute, and it’s been saying so, in public, since May.

Quick Summary: What Happened

Detail	Info
Announced	August 10, 2026
Parties	Anthropic, Macquarie Asset Management, GIC
What it is	A platform to develop, own, and lease dedicated data centers to Anthropic under long-term agreements
Who owns it	Macquarie-managed funds and GIC, funding most of the equity
Anthropic’s role	Anchor tenant, initial focus on the US
Anthropic’s pledge	100% of grid-upgrade costs tied to its demand, plus covering consumer electricity price increases in affected areas
Why it matters	It’s the 7th Anthropic infrastructure deal since November 2025, arriving after Jefferies checks put Claude Code uptime at 99.18% vs. 99.98% for OpenAI’s Codex
Official source	Macquarie Group announcement

Bottom line: Theseus buys Anthropic more dedicated capacity, further out, at bigger scale than a lease. It doesn’t buy Claude better uptime this quarter — that gap is a today problem, and today’s compute is already spoken for.

---

What Actually Happened

Strip away the finance-speak and Theseus is a sale-leaseback structure wearing a mythological name. Macquarie and GIC put up the capital, build or acquire the data centers, and own the resulting platform outright. Anthropic doesn’t own a single server rack in the deal — it signs long-term leases as the anchor tenant and gets purpose-built capacity in return. That’s a meaningfully different structure than, say, Anthropic’s $50 billion Fluidstack buildout announced last November, where Anthropic itself was footing the capital bill for sites in Texas and New York.

Two details in the announcement are worth sitting with, because they’re becoming an Anthropic signature rather than a one-off. First, Anthropic pledged to pay 100% of the grid-upgrade costs its data center demand creates — transmission lines, substations, the works. Second, it committed to covering electricity price increases that consumers in the surrounding area would otherwise absorb. Neither commitment is unique to Theseus. Anthropic made the same pledge in February, and it’s now the standing policy attached to every site the company opens, according to its own announcement on the practice. Whatever else is true about the capacity math, Anthropic has clearly decided that “the data center made your electric bill go up” is not a headline it wants to explain twice.

Why This Matters

Here’s the string of deals Theseus completes, in order:

What Infrastructure Deals Has Anthropic Signed in 2026?

1. November 12, 2025 — $50B Fluidstack buildout. Anthropic self-funds custom data centers in Texas and New York, per the company’s own announcement.
2. November 18, 2025 — $30B Azure/Nvidia partnership. Anthropic commits to $30 billion of Azure compute and up to a gigawatt of additional capacity; Nvidia and Microsoft invest up to $10B and $5B into Anthropic, respectively. Announced jointly by all three companies.
3. April 6, 2026 — Google/Broadcom TPU deal. 3.5 gigawatts of Google TPU capacity, supplied by Broadcom, coming online starting 2027 — on top of 1GW already contracted for 2026, per Anthropic’s announcement.
4. April 20, 2026 — Amazon compute expansion. Up to 5 gigawatts of new Trainium and Graviton capacity, tied to a $100 billion, 10-year AWS spending commitment and up to $25B invested back into Anthropic, per Anthropic’s own release.
5. May 5, 2026 — $200B Google Cloud commitment. Roughly $200 billion in Google Cloud spending over five years, first reported by The Information — over 40% of Alphabet’s entire disclosed cloud revenue backlog, from one customer.
6. May 6, 2026 — SpaceX/Colossus 1 deal. 100% of the Colossus 1 data center in Memphis — 300+ megawatts, 220,000+ Nvidia GPUs — handed to Anthropic, explicitly tied to raising Claude Pro and Max usage limits that same week.
7. August 10, 2026 — Theseus Infrastructure. Macquarie and GIC-owned data centers, leased to Anthropic long-term, with Anthropic as anchor tenant.

That’s not a company hedging its bets across vendors for negotiating leverage. That’s a company that keeps running out of room. Look at deal #6 again: Anthropic didn’t frame the SpaceX arrangement as expansion for its own sake. It framed it as the fix for usage limits that were already too tight — an admission, in the announcement itself, that existing capacity wasn’t cutting it for paying subscribers.

That admission wasn’t a one-off, either. Around the same week as the SpaceX deal, Anthropic said outright that “growth at this pace places an inevitable strain on our infrastructure,” and that the resulting reliability and performance hit was landing on free, Pro, Max, and Team users alike, worst during peak hours — reporting covered by Search Engine Journal. That’s Anthropic’s own language, not a competitor’s dig.

The Uptime Gap, in One Table

The clearest external evidence of that strain is a set of channel checks from Jefferies analyst Brent Thill. His research showed a real gap between Anthropic’s flagship coding product and OpenAI’s:

	Claude Code	OpenAI Codex
Uptime (Jefferies checks)	99.18%	99.98%
What that gap means	~71 hours of downtime/degradation per year at that rate	~1.75 hours per year

Thill attributed the gap to “token constraints, compute limitations and technical debt,” as reported by Yahoo Finance in coverage of Anthropic’s broader compute spending. Eighty basis points doesn’t sound like much until you convert it to hours a paying enterprise customer can’t run its agent pipeline. For teams that have wired Claude Code into production workflows, that’s not an abstraction — it’s a support ticket, or a missed deadline.

This isn’t the first time capacity pressure has visibly bent Claude’s product decisions, either. We covered the five weeks in March and April when Anthropic quietly dropped Claude’s default reasoning effort from high to medium, then reversed it without explanation once power users noticed the quality drop. Different mechanism than an uptime gap, same root cause: demand outrunning the compute available to serve it at full strength.

What Are Your Options Now

If you’re running Claude Code or Claude in production and the uptime gap is a real business risk, a few things are worth doing this month rather than waiting on Theseus to change anything:

• Build in a fallback path. If your agent pipeline has a hard dependency on Claude and no tested fallback to another model provider, that’s the actual exposure — not which vendor announced the bigger data center this week.
• Ask your Anthropic account team which capacity pool you’re on. Enterprise and Team plans have historically weathered peak-hour throttling better than Pro and free tiers. Confirm that in writing rather than assuming it.
• Watch the SpaceX capacity rollout, not the Theseus headline, for near-term relief. Colossus 1’s 300MW was explicitly earmarked to loosen today’s usage limits. Theseus facilities are new-build data centers — those take quarters to years to come online, not weeks.
• Track the Jefferies-style uptime numbers going forward, not just Anthropic’s own status page. Third-party channel checks caught this gap before Anthropic’s public messaging did.

The Bigger Picture

Zoom out and the pattern across Theseus, Fluidstack, Azure, Google-Broadcom, Amazon, the $200B Google Cloud number, and SpaceX is that Anthropic is diversifying its compute supply chain about as aggressively as any private company ever has — six hyperscaler-and-sovereign-capital partners in nine months, spanning Nvidia, Google TPU, AWS Trainium, and now dedicated real estate. That’s the correct strategic response to a real problem. Nobody serious thinks Anthropic is spending this money for the press releases.

But strategy and this-quarter reliability are two different clocks. Every one of these deals — Theseus included — measures its timeline in data-center construction schedules: site selection, permitting, power interconnects, buildout. Even the SpaceX deal, the fastest of the bunch because it repurposed an existing facility, took from May’s announcement to actually show up in loosened usage limits. Theseus facilities, being new-build, are further out than that. None of this is a knock on the deal itself. It’s a reminder that “we signed a landmark infrastructure agreement” and “your Tuesday afternoon Claude Code session won’t time out” are running on different calendars, and buyers evaluating Anthropic’s expanding enterprise footprint should price that gap in now rather than discover it during a deadline.

There’s also a valuation angle worth naming. Anthropic’s infrastructure spending has scaled roughly in step with its valuation — both went from “large” to “unprecedented” inside the same six months. That’s not a coincidence; compute commitments and funding rounds have been announced in the same weeks more than once this year. Buyers should read Theseus as part of that financing story as much as an engineering one: it’s Anthropic securing the capacity its revenue growth requires, financed in a way that keeps the capital off Anthropic’s own balance sheet.

Our Take

We don’t think Theseus, or any single deal on this list, is the wrong move. Given the growth Anthropic is managing — run-rate revenue climbing from $9 billion at the end of 2025 past $30 billion by spring, on its own account — under-investing in capacity would be the bigger error. A sale-leaseback structure that keeps a hyperscaler-scale buildout off Anthropic’s own books while still guaranteeing it dedicated, purpose-built capacity is a reasonable way to fund that growth.

What we’d push back on is any read of this announcement, or the ones before it, as an uptime fix. It isn’t one, and Anthropic hasn’t claimed it is — that’s worth crediting them for. The company that told Jefferies-tracked analysts and its own users, in plain language, that peak-hour reliability is degraded under current demand also isn’t the company promising Theseus solves that by next quarter. It’s solving 2027 and 2028’s capacity problem. The 99.18% is a 2026 problem, and 2026’s compute is already committed to somebody. If you’re deciding whether to lean harder into Claude Code this year, evaluate it on the reliability you’re getting today, not the reliability seven infrastructure deals are building toward.

Frequently Asked Questions

What is Theseus Infrastructure?

A platform announced August 10, 2026, formed by Anthropic, Macquarie Asset Management, and GIC to develop, own, and lease dedicated data centers to Anthropic under long-term agreements. Macquarie-managed funds and GIC own the platform and fund most of the equity for each project; Anthropic serves as anchor tenant, with an initial focus on US sites.

Does Theseus fix Claude’s reliability or outage problems?

Not in the near term. Theseus adds new-build data center capacity, which takes quarters to years to come online through site selection, permitting, and power interconnects. It targets Anthropic’s medium- and long-term capacity needs, not the uptime gap Jefferies analyst Brent Thill’s checks identified this year.

How much has Claude Code’s uptime actually lagged behind competitors?

According to Jefferies analyst Brent Thill’s channel checks reported by Yahoo Finance, Claude Code ran at 99.18% uptime versus 99.98% for OpenAI’s Codex — a gap Thill attributed to token constraints, compute limitations, and technical debt.

Why does Anthropic keep signing new infrastructure deals?

Anthropic’s own revenue growth — from roughly $9 billion in annualized revenue at the end of 2025 to more than $30 billion by spring 2026 — has outpaced the compute capacity available to serve it, particularly during peak hours. Anthropic has said publicly that this growth “places an inevitable strain” on its infrastructure, per reporting from Search Engine Journal, and each new deal is aimed at closing that gap.

What did Anthropic pledge to communities hosting its Theseus data centers?

Anthropic committed to paying 100% of grid-upgrade costs tied to its data center demand and to covering electricity price increases that local consumers would otherwise face. This extends a policy Anthropic first announced in February 2026 to every site it opens, Theseus included.

How is Theseus different from the SpaceX or Fluidstack deals?

Fluidstack (November 2025) and the SpaceX Colossus 1 deal (May 2026) both put Anthropic in direct control of existing or self-funded capacity — SpaceX handed over an already-built facility, and Fluidstack has Anthropic paying for construction directly. Theseus is a sale-leaseback structure: Macquarie and GIC own the platform and fund the buildout, and Anthropic leases the resulting capacity as anchor tenant, keeping the capital spend off Anthropic’s own balance sheet.

Should enterprise buyers wait for Theseus capacity before scaling up Claude usage?

No — Theseus facilities are new construction and won’t affect capacity for a while. If reliability during peak hours is a concern today, confirm with your Anthropic account team which capacity tier your plan sits on, build a tested fallback path for critical agent workflows, and track third-party uptime checks rather than waiting on infrastructure announcements to translate into better service.

---

Last updated: August 12, 2026. Sources: Macquarie Group: Theseus Infrastructure announcement · Anthropic: Covering electricity price increases · Anthropic: $50B American infrastructure investment · Anthropic: Microsoft/Nvidia strategic partnership · Anthropic: Google/Broadcom compute partnership · Anthropic: Amazon compute expansion · The Information: $200B Google Cloud commitment · Anthropic: Higher usage limits and SpaceX deal · Yahoo Finance: Jefferies uptime analysis · Search Engine Journal: Anthropic’s infrastructure strain.

Related reading: Claude Code Self-Hosted: The Fix for Rogue Agents? · Anthropic Bets $1.5B on Claude in Banks and Hospitals · Claude’s Hidden Performance Cut: What Users Found · Google’s $40B Anthropic Bet: What Changes for Claude · Anthropic Tops $900B: What It Means for Claude Users