The Brief Brief
the week in AI, briefly. then briefly again.
Artificial Intelligence — briefly, then briefly again · tbb.ceo
listen tothe daily 0:00 –:––
/daily ·28 SEPT 2026 ·MONDAY ·2 MIN READ ·6 STORIES

The $1.2 Trillion Race

Big Tech is on pace for infrastructure spending that dwarfs the railroad boom; open models are eating enterprise mindshare; and the most prominent AI lab CEO had a quiet Sunday dinner at the White House.

01 / The Day

MONDAY 28 SEPT 2026, ranked

06

Goldman Sachs: Big Tech to spend $1.2 trillion on AI infrastructure by 2027

Goldman Sachs revised its Big Tech AI infrastructure forecast upward to $1.2 trillion by 2027 — a 50% year-over-year increase the bank compares to 19th-century railroad construction as a share of GDP.

  • $1.2T in 2027 is roughly 50% above prior Wall Street consensus estimates
  • Historical parallel: US railroad construction 1870-1890 consumed a similar fraction of GDP
  • Spend covers data centers, chips, networking, and power — not model training alone
Why it mattersWhen Goldman uses the railroad analogy, institutional investors follow — this forecast will move capital allocation across every sector that depends on compute.
01 the-decoder.com↗

Open models hit 56% of Vercel tokens; enterprise mentions up 6x year-on-year

AlphaSense data tracking US corporate earnings calls shows open-model mentions rose sixfold in August and September versus the same period in 2025, while inference platform Vercel confirmed open models now account for 56% of its token traffic — predominantly from Chinese providers.

  • Sixfold increase in enterprise earnings mentions signals budget allocation is shifting, not just experimentation
  • 56% Vercel token share means open models have cleared the reliability bar for production deployments
  • Chinese providers are driving the majority of the shift away from closed frontier labs
Why it mattersEnterprise compute spending is the moat for closed-model labs — a sustained shift to open models compresses per-token economics at exactly the moment infrastructure costs are rising.
02 techmeme.com↗

Trump personally invited Dario Amodei to a private White House dinner

Sources told Techmeme that President Trump hosted Anthropic CEO Dario Amodei at a private White House dinner on Sunday — the first known one-on-one meeting between the two, framed as a thawing of relations after a year of Pentagon procurement friction.

  • Pentagon blacklisting ruling last week cited Claude safety refusals as a government procurement liability
  • Dinner was a personal invitation from Trump, not a bilateral CEO summit — unusual for a company under national security review
  • Anthropic founders are simultaneously pursuing a voting-control structure ahead of a potential IPO
Why it mattersThe dinner suggests the White House wants Anthropic inside the tent rather than outside it — the question is whether that comes with conditions on safety guardrails.
03 techmeme.com↗

Cloudflare CEO: AI bots could outnumber humans 1,000-to-1; proposes reviving HTTP 402

Cloudflare CEO Matthew Prince said AI crawler traffic is on a trajectory to reach 1,000 times human web traffic, and proposed reviving the dormant HTTP 402 Payment Required status code as a mechanism to charge bots for content access rather than block them.

  • At 1,000:1 bot-to-human ratio, most web infrastructure would effectively serve crawlers rather than readers
  • HTTP 402 has existed since 1991 but was never operationally defined; Prince sees AI crawlers as its first real use case
  • Cloudflare has already implemented unilateral blocking of major AI crawlers at customer request
Why it mattersMonetising AI crawlers rather than blocking them could create the web's first structured business model for LLM training data access — and give content publishers leverage they have not had in any prior negotiation.
04 techmeme.com↗

OpenAI: 80-90% of research already targets GPT-7 and beyond

OpenAI's Head of Applied Research told The Decoder that between 80% and 90% of current research effort targets post-GPT-6 Astra models, while noting that user adoption has become the primary constraint on impact — not capability.

  • GPT-6 Astra shipped September 3; most research capacity has already moved to the next generation
  • Adoption cited as primary bottleneck — implies the current model family is not yet used at its capability ceiling
  • At 80-90%, only a small residual team is iterating on currently deployed models
Why it mattersThe signal matters for enterprises deeply integrating GPT-6 Astra — its successor may arrive before many integration projects complete.
05 the-decoder.com↗

AI agents propose 55% of methods in model R&D, humans still decide 85% of outcomes

An analysis of 769 research task logs found AI agents proposed up to 55% of experimental methods in model development workflows, yet humans retained final decision-making authority in over 85% of outcomes — and one-third of tasks attempted would not have been tried without AI assistance.

  • 55% method proposals from AI is a structural change in research methodology, not incremental productivity gain
  • 85% human final decisions suggests AI is augmenting scientific judgment rather than replacing it
  • One-third of tasks were AI-unlocked — the volume of research attempts is expanding, not just the speed
Why it mattersHow labs structure human-AI collaboration in model development will determine whether safety oversight scales with research velocity — this data shows it currently does, just barely.
06 the-decoder.com↗
Be subscriber #014 the weekly ten, every friday by email · no spam · unsubscribe anytime
Past days

The Daily Archive

/MON 28 SEPTThe $1.2 Trillion Race/SUN 27 SEPTAgent paper trail grows; labs wrestle with governance/SAT 26 SEPTMuse, the pause, and the pipeline/THU 24 SEPTRogue agents, Gemini 4, and a real enzyme/WED 23 SEPTThe Race to Cheaper Frontier: GPT-6 Sol, Opus 5.5, and a Safety Reckoning/TUE 22 SEPTThe Control Question/MON 21 SEPTFable grounded, bilateral circuits formed, and memory doubles/SAT 19 SEPTTwo AIs in the Room/FRI 18 SEPTThe machines are doing the research now/THU 17 SEPTGovernance Under Fire/WED 16 SEPTAGI Claims, Valuations, and the Agentic Safety Bill/TUE 15 SEPTThe pushback arrives/MON 14 SEPTThe Standards Reckoning/SUN 13 SEPTPacing Day/FRI 11 SEPTThe Slowdown Question/THU 10 SEPTSafety's Reckoning Hits Congress/WED 09 SEPTThe Week AI Did Math/TUE 08 SEPTRound Numbers/MON 07 SEPTAstra pays off, Pachocki sounds the alarm/SUN 06 SEPTThe audit weekend/SAT 05 SEPTSaturday's AI Brief