Anthropic's own campuses (Fluidstack)
Anthropic / Fluidstack · Build-out
Announced Sites coming online throughout 2026
Anthropic's $50B US build-out, starting with custom Fluidstack sites in Texas and New York (towns not officially named). Separate from the rented Amazon capacity Anthropic also runs — it operates both tracks.
Figures as reported — see sources.
Why this label: Anthropic announcement; site detail via Data Center Dynamics.
Sources: Anthropic announcement via research compilation · Data Center Dynamics
Checked Oct 14, 2026
Stargate UAE — first capacity + redesign
G42 / OpenAI / Oracle / NVIDIA / Cisco / SoftBank · Build-out
Announced First 200 MW expected live 2026
The first Stargate outside the US: a 1 GW cluster inside a planned 5 GW Abu Dhabi campus; first 200 MW under construction for 2026 delivery.
In Sep 2026 Reuters reported the master plan is being redesigned toward distributed, hardened facilities — timelines and costs under review.
Figures as reported — see sources.
Why this label: Split label: the build is announced; the redesign is reported (Reuters, six sources).
Sources: Data Center Knowledge · Reuters via MENA Cyber Wire
Checked Oct 14, 2026
OpenAI's first hardware device — public reveal
OpenAI · Device
Teased Second half of 2026
OpenAI's chief global affairs officer told Axios at Davos the company is 'on track' for an unveiling in the latter part of 2026.
The form factor is officially undisclosed; court filings in the iyO case say only that it is not an in-ear device or a wearable, and the name 'io' will not be used.
Why this label: Exec statement to Axios at Davos; no firm date, form factor undisclosed.
Sources: MacRumors · 9to5Mac
Checked Oct 14, 2026
1X NEO home robot deliveries
1X Technologies · Robotics
Reported First US deliveries late 2026 (reported)
NEO, a consumer home humanoid, is on pre-order at a reported $20,000 or $499/mo, with first deliveries reported to begin late 2026 — which would make it the first humanoid sold for the home.
Figures as reported — see sources.
Why this label: Single-aggregator sourcing so far — flagged in the research pack itself.
Single-aggregator sourcing so far; re-verify against 1X's own site before featuring prominently.
Sources: OptimusK tracker
Checked Oct 14, 2026
GPT-6 Astra wider availability (API + paid tiers)
OpenAI · Model
Announced Rolling out from Oct 2026; no GA date
Astra, OpenAI's top-tier GPT-6 model, launched limited to a set of organisations; OpenAI's announcement describes planned expansion to paid tiers and API access. No GA date has been given.
Why this label: OpenAI's own GPT-6 announcement describes the planned expansion.
Sources: dev.to — rollout breakdown · explainx — on the 6.1 cancellation
Checked Oct 14, 2026
Microsoft Maia 300 deployment
Microsoft · Hardware
Announced Unveiled Oct 2026; Azure deployment ramp through 2027
Microsoft's third custom AI inference chip (2nm-class compute chiplets, 288GB HBM4, ~950W) has just been unveiled (Oct 2026); the upcoming part is fleet deployment across Azure to serve Copilot/OpenAI workloads.
Why this label: Microsoft unveiling, as covered by RCR Tech.
Sources: RCR Tech
Checked Oct 14, 2026
Stargate US — next phases
OpenAI / Oracle / SoftBank · Build-out
Announced Phases late 2026 through 2029
The multi-site US programme beyond Abilene: Shackelford County early buildings late 2026-early 2027, Lordstown OH operational 2026, Wisconsin and Michigan as 2028 stories; programme figure >9 GW by 2029 against the $500B headline.
The proposed +600 MW Abilene expansion was cancelled in Mar 2026 — see the cancellation record.
Figures as reported — see sources.
Why this label: Programme announced by OpenAI/Oracle/SoftBank; schedule detail per Epoch AI tracking.
Sources: Epoch AI schedule via research compilation · OpenAI five-sites announcement, extracted · Mint/Reuters — on the Abilene cap
Checked Oct 14, 2026
Vera Rubin volume ramp
NVIDIA · Hardware
Announced Volume ramp Q4 2026, growing through H1 2027
Rubin (R100 GPU + Vera CPU, 288GB HBM4) is in production; NVIDIA's CFO confirmed production shipments start Q3 2026 with the volume ramp through Q4 2026 and H1 2027. First allocations go to hyperscalers.
Why this label: NVIDIA CFO on the record in earnings calls, as quoted by Wccftech.
Sources: Wccftech, quoting NVIDIA earnings calls
Checked Oct 14, 2026
Mistral Large 4 open weights
Mistral AI · Model
Announced 27 Oct 2026 (reports vary 27-31 Oct)
The 1.05T-parameter flagship ('Le Chonk', ~49B active, 1M context) shipped as an API preview on 6 Oct 2026; the downloadable open weights land after a ~3-week security red-team period. Licence terms are not yet published.
Why this label: Company API preview shipped 6 Oct 2026; weights date from launch coverage.
Check the licence terms on release day — they were not published in advance.
Sources: AI News — launch coverage · CoderCops analysis
Checked Oct 14, 2026
Anthropic IPO
Anthropic · Ecosystem
Reported November 2026 (reported)
Confidential S-1 filed 1 Jun 2026 (confirmed); Reuters/FT have reviewed the draft; WSJ reports advisers moved the listing to November 2026, after the US midterms. No public S-1 on EDGAR as of 13 Oct 2026 — that filing is the next hard milestone.
Reported valuation targets (up to ~$2T) are reported, not set.
Figures as reported — see sources.
Why this label: WSJ/Reuters/FT reporting on a confirmed confidential filing; the company has not announced a date.
Sources: AITechSafar timeline · DecodeTheFuture S-1 tracker
Checked Oct 14, 2026
Anthropic compute scale-up: Trainium + TPU gigawatts
Anthropic / Amazon / Google / Broadcom · Build-out
Announced ~1 GW Trainium by end 2026; ~3.5 GW TPU from 2027 (reported)
Anthropic's April 2026 Amazon agreement covers up to 5 GW of AWS capacity (> $100B ten-year commitment, ~1 GW of Trainium2/3 by end-2026); Broadcom's 8-K separately describes ~3.5 GW of TPU capacity for Anthropic from 2027.
This is the physical precondition for whatever Claude ships in 2027.
Figures as reported — see sources.
Why this label: Split label: Trainium track announced (Amazon agreement); the TPU figure is reported from Broadcom's SEC 8-K.
Sources: Anthropic statement via Wccftech · 8-K detail via MEXC analysis
Checked Oct 14, 2026
Colossus II — next phase
xAI (SpaceXAI) · Build-out
Announced Next phase expected end-2026 (company S-1)
SpaceX's S-1 expects Colossus II's next phase to add at least 220,000 additional GB300 processors and over 400 additional megawatts (company forward-looking statement). Musk's gigawatt totals are his claims and are labelled as such.
Also announced as a plan: Terafab, the Tesla-SpaceX chip fab — no timeline published.
Why this label: Split label: the phase is announced in SpaceX's S-1; the gigawatt totals are teased exec claims.
Gigawatt totals are Musk's claims, labelled as such — the S-1 figures are the sourced floor.
Sources: TSN — S-1 roadmap review
Checked Oct 14, 2026
DeepSeek V4.1-Pro
DeepSeek · Model
Announced No date given No date given
DeepSeek has said a V4.1-Pro is coming; it has given no date and no specs. (V4.1-Flash shipped 10 Sep 2026.) Anything beyond that sentence — including everything labelled 'R2' — is unverified.
Why this label: Company statement that V4.1-Pro is coming; no date or specs published.
Sources: DecodeTheFuture status tracker, checked against DeepSeek's official docs
Checked Oct 14, 2026
Gemini 4 family expansion (Pro / Flash tiers beyond Argon)
Google DeepMind · Model
Reported Before end of 2026 (reported); no dates No date given
Gemini 4 Argon already shipped (30 Sep 2026) — what remains upcoming is the rest of the family. Reporting says DeepMind is racing to close a measured gap to Claude Opus 5.5 before 2026 ends; no tier dates exist.
Figures as reported — see sources.
Why this label: The Information reporting via trade coverage; Google has not confirmed dates.
An unconfirmed 'Gemini 4 Pro' arena checkpoint is Rumour Corner material only.
Sources: Tech Insider · explainx
Checked Oct 14, 2026
Grok 5
xAI (SpaceXAI) · Model
Teased Company claim: 'before the end of this year' (Musk, 4 Aug 2026) — a claim, not a date No date given
Confirmed in training since Jan 2026 (company statement, repeated in SpaceX's S-1, which says it is training at Colossus II). No official release date, specs or benchmarks. Circulating spec figures are unconfirmed — see the Rumour Corner.
Why this label: Company statement + S-1 confirm training; the date is an exec claim only.
Treat every Grok 5 date and spec as a claim until xAI publishes them.
Sources: TSN — S-1 and statements review · ScriptByAI — Grok timeline
Checked Oct 14, 2026
Meta Hyperion — 5 GW expansion phases
Meta · Build-out
Announced Phases through the late 2020s; no single completion date published No date given
Meta announced 13 Jul 2026 that Hyperion (Richland Parish, Louisiana) expands to 5 GW of compute capacity with investment over $50B, up from the original $10B/2 GW plan.
Entergy's supporting power build-out (~$12.9B as filed) is under Louisiana regulatory review — the power is the schedule risk.
Figures as reported — see sources.
Why this label: Meta announcement (13 Jul 2026) via Reuters; power build-out per regulatory filings coverage.
Sources: Reuters via SRN News · Construction Review
Checked Oct 14, 2026
Boston Dynamics Atlas — commercial launch
Boston Dynamics (Hyundai) · Robotics
Reported Commercial window reported 2026-2028
The all-electric Atlas is in Hyundai pilots; aggregator trackers put commercial availability in a 2026-2028 window with an initial fleet serving Hyundai plants. No dated commitment from Boston Dynamics itself.
Figures as reported — see sources.
Why this label: Aggregator trackers only; Boston Dynamics has made no dated commitment.
Sources: OptimusK tracker
Checked Oct 14, 2026
AMD Instinct MI455X / Helios production ramp
AMD · Hardware
Reported Samples + low-volume H2 2026; production ramp Q2 2027 (reported)
AMD says Helios racks (72x MI400-series, EPYC 'Venice') are on track for 2H 2026; SemiAnalysis reports manufacturing delays push the true production ramp to Q2 2027. Both are printed: AMD's line as the announcement, SemiAnalysis as the reported reality-check.
Figures as reported — see sources.
Why this label: AMD's on-track statement is announced; the Q2 2027 ramp is SemiAnalysis reporting.
Sources: The Next Platform
Checked Oct 14, 2026
Google TPU 8t / TPU 8i
Google · Hardware
Announced Late 2026-2027 (sources differ)
Google's 8th-gen split: TPU 8t for training (9,600-chip superpods, 121 FP4 exaFLOPS as announced) and TPU 8i for inference, on TSMC 2nm. One report says both launch on Google Cloud later in 2026; another targets late-2027 availability — the range is printed, not resolved.
Figures as reported — see sources.
Why this label: Previewed by Google at Cloud Next Apr 2026; availability window differs by source.
Sources: MLQ · Substack analysis
Checked Oct 14, 2026
AMD Instinct MI500 series
AMD · Hardware
Announced 2027 (roadmap)
AMD has publicly indicated an MI500 series in 2027 (next-gen HBM, copper + optical interconnect), with MI600 'under development' for 2028. Roadmap-level only; no specs confirmed.
Why this label: AMD roadmap statements as covered by BigGo Finance.
Sources: BigGo Finance — AMD roadmap coverage
Checked Oct 14, 2026
Microsoft Fairwater II + Mount Pleasant expansion
Microsoft · Build-out
Announced Second facility 2027 or later
Fairwater I (Mount Pleasant, WI) opened early 2026. Microsoft committed a further $4B (over $7B total in Wisconsin) for a second, similar-scale facility, and the village approved 15 additional data centres in Jan 2026 — permits stage, not construction.
Local politics is a live risk.
Figures as reported — see sources.
Why this label: Microsoft commitment as reported by Reuters coverage; approvals per MLQ.
Sources: AInvest / Reuters coverage · MLQ
Checked Oct 14, 2026
OpenAI IPO
OpenAI · Ecosystem
Reported 2027 (reported)
Confidential draft filed Jun 2026; Altman has ruled out 2026; CFO Sarah Friar has reportedly told staff the company will be public in 2027 'if not sooner'. No public S-1 — until one appears, date, price and valuation are not facts.
Figures as reported — see sources.
Why this label: Reporting on a confidential draft filing; Altman has ruled out 2026 on the record.
Sources: DecodeTheFuture
Checked Oct 14, 2026
OpenAI device — customer shipments
OpenAI · Device
Announced Not before end of Feb 2027
In a sworn court filing (Feb 2026), OpenAI's VP/GM Peter Welinder stated the first device will not ship to customers before the end of February 2027, and that no packaging or marketing materials exist yet — the firmest date in the device story, because it is a legal filing.
Why this label: Sworn court filing by OpenAI's VP/GM Peter Welinder, Feb 2026.
Sources: MacRumors
Checked Oct 14, 2026
Figure AI compute build (Nscale)
Figure AI / Nscale · Build-out
Announced First GPU deployments H2 2027
Figure committed $3.5B to Nscale compute (announced 3 Sep 2026) to train future versions of Helix, its humanoid control model — first GPU deployments in H2 2027 at Barstow, Texas, up to 100,000 Vera Rubin GPUs at full build.
A robotics company signing a hyperscaler-scale compute deal is itself the story.
Figures as reported — see sources.
Why this label: Figure/Nscale announcement (3 Sep 2026) via eWeek.
Sources: eWeek
Checked Oct 14, 2026
Rubin Ultra (NVL576)
NVIDIA · Hardware
Announced H2 2027
NVIDIA's published roadmap: 576-GPU racks, ~15 exaFLOPS FP4, HBM4e, NVLink 7, ~600 kW per rack. Announced at GTC 2025 and reaffirmed since — a roadmap commitment, not a shipping product.
Figures as reported — see sources.
Why this label: NVIDIA's published roadmap (GTC 2025, reaffirmed).
Sources: ARN — GTC roadmap coverage
Checked Oct 14, 2026
Tesla Optimus — public sales
Tesla · Robotics
Reported End of 2027 (conditional)
Musk told BlackRock's Larry Fink at Davos (Jan 2026) Tesla expects to sell Optimus to the public by end of 2027, only when confident in very high reliability and safety — his words make the date conditional. Internal factory deployment through 2026 is the nearer milestone.
Figures as reported — see sources.
Why this label: Musk's Davos statement as reported; the date is explicitly quality-gated.
Sources: TechRepublic · Morningstar / MarketWatch
Checked Oct 14, 2026
Kimi K4
Moonshot AI · Model
Reported No window reported No date given
Bloomberg reports Moonshot is seeking additional Nvidia Blackwell chips specifically to train its next-generation Kimi K4 (K3 shipped Jul 2026). Chip-sourcing reporting, not a product announcement — no date, no specs.
Figures as reported — see sources.
Why this label: Bloomberg reporting on chip sourcing for K4 training; no company announcement.
Sources: Bloomberg via MLQ
Checked Oct 14, 2026
Muse Spark open-weights release
Meta (Superintelligence Labs) · Model
Announced No date given No date given
Meta says an open-weights Muse Spark release is coming; it has not shipped. Muse is Meta's closed frontier line (Spark 1.3 current) that replaced Llama as the flagship strategy — an open Muse would be Meta's first open release of the new era.
Why this label: Meta statement that an open-weights Muse Spark release is coming.
Sources: Mungomash — Llama/Muse version tracker
Checked Oct 14, 2026
Project Genie — wider availability
Google DeepMind · Platform
Teased Expansion planned; no date No date given
Project Genie (the Genie 3 world-model prototype) launched 29 Jan 2026 to US AI Ultra subscribers only. DeepMind says it intends to expand access and keep improving the prototype; no wider-availability date has been given.
Why this label: DeepMind has confirmed the intent to expand access; no date given.
Sources: Quantum Zeitgeist
Checked Oct 14, 2026
Feynman architecture
NVIDIA · Hardware
Announced 2028
The post-Rubin generation on NVIDIA's public roadmap: 3D-stacked design, custom HBM, paired with the 'Rosa' CPU. Named and dated by NVIDIA; essentially no confirmed specs yet — and this page says exactly that.
Why this label: Named and dated on NVIDIA's public roadmap; no confirmed specs.
Sources: ARN — GTC roadmap coverage · Cyfuture roadmap summary
Checked Oct 14, 2026
Orbital AI compute
SpaceXAI · Build-out
Reported As early as 2028 (S-1); Musk says 2027 — sources conflict
SpaceX's S-1 describes plans to begin deploying orbital AI compute satellites as early as 2028; Musk has publicly said 2027. The two statements conflict — both are printed, and the item stays in Horizon until hardware exists.
Anthropic has 'expressed interest' in orbital capacity, which is not a commitment.
Figures as reported — see sources.
Why this label: S-1 language as reviewed; the 2027 figure is an exec claim that conflicts with it.
The two source statements conflict; both are shown, neither is resolved silently.
Sources: TSN — S-1 roadmap review
Checked Oct 14, 2026
Figure 03 / BotQ scale-up
Figure AI · Robotics
Announced 100,000 units over 4 years (by ~2029) — company target
Figure's BotQ factory reached 1 robot/hour in Apr 2026 with 350+ Figure 03 units delivered; the company's stated trajectory is 12,000 units/yr on the first line scaling to 100,000 robots over four years, sold largely as Robotics-as-a-Service.
Why this label: Company-stated production trajectory; targets are projections, labelled as such.
Production targets are company projections, not commitments.
Sources: Inside BotQ (ai2.work) · NextBigFuture
Checked Oct 14, 2026
AI-designed drugs reach the clinic
Isomorphic Labs (Alphabet) · Frontier
In progress First-in-human trials targeted by end of 2026 (company statement) No date given
The first real-world test of AI drug discovery: Isomorphic (17 programmes, oncology/immunology, with Lilly/Novartis/J&J) targets first-in-human trials by end of 2026 — Hassabis's on-record Davos statement, itself a one-year slip.
As of 30 Sep 2026 the company had not announced an IND filing or trial start, so the date is live but its slipping risk is documented.
Figures as reported — see sources.
Why this label: Hassabis on record at Davos (Jan 2026); status checked against Sep 2026 reporting.
Dates are company targets; this programme has already slipped once.
Sources: Reuters · BiopharmaTrend status check
Checked Oct 14, 2026
Brain-computer interfaces (AI-decoded)
Neuralink · Frontier
In progress In progress — roadmap targets are company claims No date given
Per Neuralink's published roadmap (company claims — labelled as such): Blindsight first human trial targeted 2026, multi-region/multi-device implants 2027, scale to 25,000 electrodes and AI-integration work 2028.
21 patients implanted as of recent reporting; high-volume production and automated surgery are the 2026 programme.
Figures as reported — see sources.
Why this label: Neuralink's published roadmap (2025 presentation) — company targets, not commitments.
Every date here is a company target — this company's timelines have slipped before.
Sources: BigGo — Neuralink roadmap coverage
Checked Oct 14, 2026
Continual learning
Academia + labs (no owner) · Frontier
In progress In progress — first continual frontier models estimated 2028-2030 No date given
The blocker has a name — the 'Memento Constraint': models that learn continuously without catastrophic forgetting. A May 2026 research map finds forgetting rates of 40-80% under standard fine-tuning and nothing production-ready; first genuinely continual frontier models are
estimated 2028-2030. This is the frontier most likely to change what a 'model release' even means — and it is not close.
Figures as reported — see sources.
Why this label: May 2026 research map; estimates are the map's, labelled as such.
Sources: Continual Learning Research Map, May 2026
Checked Oct 14, 2026
Embodied AI / robot foundation models
Google DeepMind; Figure AI; Tesla · Frontier
In progress In progress — no ship date No date given
The race to one model that perceives, plans and actuates across robot bodies: DeepMind's Gemini Robotics line (ER-1.6 shipped Apr 2026) and Figure's Helix are the visible programmes, consolidating video, 3D and control into single physical foundation models.
Progress is real but gated by hardware production, not just models.
Why this label: Sourced programmes in progress; no ship dates by nature.
Sources: DeepMind model history · eWeek on Helix compute
Checked Oct 14, 2026
World models
Google DeepMind; World Labs (to AMD) · Frontier
In progress In progress — no ship date No date given
Genie 3 generates real-time, interactive, physics-consistent worlds from prompts, and DeepMind presents world models as robot-training infrastructure, not just media.
World Labs is being acquired by AMD for ~$8.2B in stock (announced 28 Sep 2026, expected to close by end of 2026); the frontier to watch is persistent, simulation-grade worlds used to train robots.
Figures as reported — see sources.
Why this label: Sourced programmes in progress; frontiers carry no ship dates by nature.
Sources: Lexblog — DeepMind's Raia Hadsell on Genie 3 · RecaTools — on the AMD-World Labs deal
Checked Oct 14, 2026
