the week in AI, briefly. then briefly again.
Artificial Intelligence — briefly, then briefly again · tbb.ceo
listen tothe daily 0:00 –:––
/daily ·20 NOV 2026 ·FRIDAY ·3 MIN READ ·6 STORIES

ANT's First Day, Microsoft's Seven, and $10B in Revenue

Anthropic's shares close 17% above the $83 IPO price in a volatile Nasdaq debut, Microsoft unveils seven in-house AI models at Ignite to cut its OpenAI bill, and OpenAI reports a record $10 billion quarterly revenue.

01 / The Day

FRIDAY 20 NOV 2026, ranked

06

Anthropic ANT Closes at $97.17 on Nasdaq Debut — Up 17% but off $115 High

Anthropic's shares opened at $111.50, briefly hit $115.10, then closed at $97.17 — 17% above the $83 IPO price but 15% below the intraday peak. The debut raised a record $81.3B, surpassing SpaceX's $75B prior US offering record. Nasdaq fell 2.25% the same day, underscoring the volatility of a trillion-dollar pure-play AI debut.

  • Closed at $97.17 (+17% vs $83 IPO price); opened at $111.50, peak $115.10 intraday
  • IPO raised $81.3B, surpassing SpaceX's $75B prior record for a US offering
  • Nasdaq fell 2.25% the same day; Anthropic outperformed the broader tech selloff
Why it mattersWhether retail demand held up after the open — it did not — sets the expectation for how volatile pure-play AI names will trade once the IPO pop fades.

Microsoft Launches Seven In-House AI Models at Ignite, Led by MAI Code 2

Microsoft unveiled seven MAI models at Ignite 2026, anchored by MAI Code 2 and MAI Code 2 Flash — which Microsoft claims deliver 96% of Claude Opus 5.5 performance at roughly 5% of the cost. The lineup also includes MAI Cyber 2 Flash for security operations and MAI Work Flash for productivity. All pricing uses a half-off introductory rate that holds until May 2027.

  • MAI Code 2: vendor-reported 96% of Claude Opus 5.5 on selected benchmarks at ~5% of the cost
  • MAI Cyber 2 Flash: 92.3% on CyberGym-E2E versus 91.5% for GPT-6 Astra
  • Introductory pricing halves per-token costs through May 17, 2027; standard rates apply after
Why it mattersMicrosoft building its own models is a direct hedge against OpenAI dependency — the relationship that has defined enterprise AI access since 2020.

OpenAI Reports $10 Billion in Quarterly Revenue Amid Heavy Losses

OpenAI disclosed quarterly revenue of $10 billion, confirming that enterprise and consumer demand has scaled faster than any prior forecast. Losses remain substantial as infrastructure and staffing costs grow alongside revenue. The figure puts OpenAI on track to exceed $40 billion annually ahead of its own IPO.

  • $10B quarterly revenue; losses still substantial — exact figure not publicly disclosed
  • Revenue growth driven by ChatGPT Enterprise, API, and healthcare segment partnerships
  • On track for $40B+ annual run rate before its own public market debut
Why it mattersThe revenue print is the clearest signal yet that hyperscale AI consumption is real and not just enterprise-pilot-stage — which matters for how investors valued the Anthropic debut.

DeepMind's Hassabis Steps Back; Kavukcuoglu Takes Over Gemini Model Development

Google DeepMind CEO Demis Hassabis stepped back from day-to-day model development responsibilities, with Koray Kavukcuoglu assuming leadership over the Gemini team. The reshuffle follows the fastest model-release cadence in DeepMind's history — four Gemini variants shipped in roughly seven weeks. Jeff Dean, who transitioned to an external role earlier in 2026, has since started an independent lab.

  • Kavukcuoglu was VP of Research at DeepMind; previously led AlphaGo-era reinforcement learning work
  • Jeff Dean departed DeepMind earlier in 2026 and has since launched an independent lab
  • Four Gemini model launches in ~7 weeks precede the leadership change
Why it mattersHassabis stepping back from daily operations during peak model competition typically signals either a strategic repositioning or a push toward longer-horizon AGI research rather than quarterly benchmark racing.

OpenAI Astra for Law Tops Global Legal AI Benchmark

OpenAI's Astra for Law scored highest in a global legal AI benchmark evaluation, outperforming specialized models from Thomson Reuters, Harvey, and Microsoft on contract analysis, regulatory research, and precedent recall tasks. The model is included with OpenAI's legal pricing tier at $200 per seat per month.

  • Outperforms Harvey, Thomson Reuters CoCounsel, and Microsoft Azure Legal on overall benchmark score
  • Strongest gains on cross-jurisdictional regulatory tasks and contract clause extraction
  • Included with OpenAI legal tier at $200/seat/month
Why it mattersIf a general-purpose model from an AI lab tops vertical legal benchmarks, it accelerates displacement of bespoke legal-AI tooling — compressing the runway for specialized providers.

US Approves Export of 879,000 NVIDIA and AMD AI Chips to Chinese Companies

The US Commerce Department approved export licenses for 879,000 AI accelerator chips from NVIDIA and AMD to Chinese buyers, in a selective relaxation of the export controls that have restricted high-performance AI hardware sales to China since 2022. Full-specification H100, H200, B200, and AMD MI300X chips remain restricted.

  • Covers NVIDIA H20 and AMD Instinct variants configured below the BIS control threshold
  • Chinese recipients include cloud operators and domestic AI labs, per Commerce filing
  • Top-spec H100/H200/B200 and AMD MI300X remain restricted under existing BIS rules
Why it mattersSelective approvals let Washington preserve leverage on the most capable hardware while partially softening the economic pressure on US chipmakers whose shares have been hurt by the China access overhang.
Be subscriber #028 weekly AI email interest list
Past days

The Daily Archive