Held to their word.

We believe that, in 2025, we may see the first AI agents “join the workforce” and materially change the output of companies.

The stated window has passed. “Materially change the output of companies” has no agreed measure; we have not yet reviewed enough independent evidence to assign delivered, partially or missed.

“I think we’ll be there in three to six months—where AI is writing 90 percent of the code.”

Window has passed. Share of code written by AI is reported differently by different companies and is rarely independently audited; status awaits a review of independent measurements.

Stargate “intends to invest $500 billion over the next four years building new AI infrastructure for OpenAI in the United States. We will begin deploying $100 billion immediately.”

Multi-year commitment still within its stated window. Announced site and capacity figures are company claims; spending to date has not been independently totalled here.

“In FY 2025, Microsoft is on track to invest approximately $80 billion to build out AI-enabled datacenters … More than half of this total investment will be in the United States.”

Not yet verified. Microsoft’s capital expenditure including finance leases totalled about $88.2 billion in FY2025 (sum of the four quarterly figures its CFO reported). That covers all capital spending, not only AI-enabled data centres, and Microsoft does not break out the AI share, so we cannot yet confirm the $80 billion figure or the US share. The deadline has passed; that alone does not make the promise broken.

Microsoft intends “to invest more than $35 billion in 14 countries within three years to build trusted and secure AI and cloud datacenter infrastructure.”

Within its stated window. Country-level spending is not consistently disclosed.

“In 2025 alone, Microsoft is on a path to train 2.5 million American students, workers, and community members with the AI skills to land new jobs.”

Window has passed. Any delivery figure would be self-reported; we have not yet located a dated company report or independent count.

“We look forward to bringing Carbon to developers, businesses, and consumers soon” — making it available “as soon as possible” after gathering feedback on guardrails.

Access currently limited to cyber defenders in the Fairwind Program. We will mark delivered when a general API or product release is documented.

Microsoft will honour the common AI-training opt-out signal for publishers beginning 1 February 2027.

Paraphrased from wire reporting of Microsoft’s blog post. Google committed to “sometime in 2027”, OpenAI to “the coming months”.

OpenAI said it will roll out support for the common AI-training opt-out signal “in the coming months”.

Review deadline is our editorial assumption, not OpenAI’s.

Grok 5 will be out before the end of this year and it will be crushingly good.

Grok 5 was not released in 2025. xAI said on 6 Jan 2026 that Grok 5 was “currently in training”; as of 1 Nov 2026 xAI’s API release notes list Grok 4.7 (21 Sep 2026) as the newest model and no Grok 5.

With Grok 5, which should be out before the end of this year, we will be incorporating the entire corpus of SpaceX data.

Not released as of 1 Nov 2026; xAI has announced no release date. Earlier targets (end of 2025, Q1 2026) passed.

The @xAI Grok 2.5 model, which was our best model last year, is now open source. Grok 3 will be made open source in about 6 months.

Grok 2.5 weights were released (xai-org/grok-2). Musk reaffirmed Grok 3 would be open-sourced in Feb 2026, but as of 1 Nov 2026 xAI’s Hugging Face organisation lists only grok-1 and grok-2; we found no official Grok 3 weights.

The @xAI goal is 50 million in units of H100 equivalent-AI compute (but much better power-efficiency) online within 5 years.

Deadline not reached. For scale, Epoch AI estimated Colossus 2 at roughly 1.1 million H100-equivalents in September 2026 (as reported by AI Prediction Index).

The Colossus 2 supercomputer for @Grok is now operational. First Gigawatt training cluster in the world. Upgrades to 1.5GW in April.

Epoch AI’s satellite analysis (Jan 2026) found about 350 MW of cooling capacity, disputing the 1 GW claim; Epoch’s September 2026 estimate of about 946 MW of IT power (reported by AI Prediction Index) means the 1.5 GW April upgrade did not happen on time.

Version 0.1 early beta of Grokipedia will be published in 2 weeks

Delivered 8 days late. Musk set the two-week target on 5 Oct 2025, postponed the launch on 21 Oct, and Grokipedia v0.1 went live on 27 Oct 2025 with about 885,000 articles.