Still open
114 forecasts are still open. One is past its date; its evidence is not yet decisive.
Next to be checked
the model says · how likely · checked byA top AI lab or cloud giant trains, not just runs, a huge model on GB200 NVL72 at production scale.72%checked 1 time · evidence not yet decisive · next check 26 Oct 2026
unresolved · evidence not yet decisive
Condition not established: "Mega-scale model training on GB200 NVL72 at production scale" is measured at the cluster: a training run, pre- or post-training, for a frontier-model programme on a GB200 NVL72 cluster of at least 10,000 GPUs. Per-job GPU counts are rarely reported; cluster scale is.
last looked
first recorded · before the live record began
as written in the record
At least one frontier lab or hyperscaler is running mega-scale model TRAINING (not just inference) on GB200 NVL72 at production scale, per SemiAnalysis / credible reporting
Google open-sources the native TorchTPU backend used by vLLM or SGLang.85%
still open · check by
last looked
first recorded · live
as written in the record
Google has publicly open-sourced the native TorchTPU backend used by vLLM or SGLang
ERCOT files its Batch Zero audit verification report by December 10, 2026.60%
still open · check by
last looked
first recorded · live
as written in the record
ERCOT files its Batch Zero audit verification report by December 10, 2026
At least one solid-state transformer built for datacenters completes UL certification.55%
still open · check by
first recorded · before the live record began
as written in the record
At least one datacenter-targeted Solid-State Transformer (SST) has completed UL certification for datacenter deployment (per SemiAnalysis / credible reporting)
SemiAnalysis's H100 one-year rental index in Dec 2026 holds at or above its March 2026 level of about $2.35 an hour.68%
still open · check by
first recorded · before the live record began
as written in the record
The SemiAnalysis public H100 1-year rental contract price index for Dec 2026 is at or above its March 2026 level of ~$2.35/hr/GPU
The six biggest coding assistants reach a combined $100B yearly revenue pace, on SemiAnalysis's model.55%
still open · check by
first recorded · before the live record began
as written in the record
The six largest coding-assistant players' combined ARR reaches ≥$100B (per SemiAnalysis Tokenomics model)
xAI raises at least $10B by selling new shares in a fresh funding round.80%
still open · check by
first recorded · before the live record began
as written in the record
xAI closes a new primary equity funding round of ≥$10B (per credible reporting / filings)
Anthropic's annual revenue run-rate reaches at least $100B.85%
still open · check by
first recorded · before the live record began
The odds were revised: 50% on → 85% on .
as written in the record
Anthropic's annualized revenue run-rate (ARR) reaches ≥$100B (per SemiAnalysis / credible reporting)
AWS open-sources a native PyTorch backend for Trainium with Aten eager mode and torch.compile.85%
still open · check by
first recorded · before the live record began
as written in the record
AWS has open-sourced a native PyTorch backend for Trainium (Aten eager-mode + torch.compile), per SemiAnalysis / AWS
CoreWeave earns the top-tier ClusterMAX Platinum rating in SemiAnalysis's next ClusterMAX release.78%
still open · check by
first recorded · before the live record began
as written in the record
CoreWeave is rated ClusterMAX Platinum (top tier) in the next ClusterMAX release (per SemiAnalysis)
Cerebras Systems goes public, with its shares trading on a public exchange.80%
still open · check by
last looked
first recorded · before the live record began
as written in the record
Cerebras Systems completes an initial public offering (its shares begin trading on a public exchange), per SemiAnalysis / credible reporting
Nvidia ships a 512×800G-class Spectrum-X Ethernet switch with co-packaged optics.60%
still open · check by
first recorded · before the live record began
The odds were revised: 58% on → 60% on .
as written in the record
Nvidia has begun shipping / made generally available a co-packaged-optics (CPO) Spectrum-X Ethernet switch (512×800G-class)
Samsung qualifies HBM4 and ships it in volume to Nvidia or AMD.50%
still open · check by
last looked
first recorded · before the live record began
The odds were revised: 40% on → 50% on .
as written in the record
Samsung has qualified and is shipping HBM4 in volume to a leading merchant AI-accelerator customer (Nvidia or AMD), per SemiAnalysis / credible reporting
Anthropic announces or closes a new share sale valuing it at $300B or more.85%
still open · check by
last looked
first recorded · before the live record began
as written in the record
Anthropic closes or announces a primary equity round at a valuation ≥$300B (per credible reporting / filings)
SoftBank cuts its ownership stake in Arm below its level when this forecast was recorded.60%
still open · check by
last looked
first recorded · before the live record began
as written in the record
SoftBank reduces its Arm Holdings ownership stake below its current level (per public filings)
ANYbotics ships the ANYmal X, an ATEX explosion-proof, Zone-1-rated four-legged robot.62%
still open · check by
last looked
first recorded · before the live record began
as written in the record
ANYbotics has launched / shipped an ATEX (explosion-proof, Zone-1-rated) quadruped — the ANYmal X — per SemiAnalysis / credible reporting
The US tightens controls on at least one Huawei loophole: HBM in packages, or tool servicing or chemicals to SMIC/CXMT.62%
still open · check by
first recorded · before the live record began
The odds were revised: 60% on → 62% on .
as written in the record
The US (BIS) tightens controls against at least one Huawei-supply loophole flagged here — HBM re-export via HBM-bearing packages/SiP, or WFE tool-servicing / sole-sourced process chemicals to SMIC/CXMT
Anthropic uses AWS Trainium chips in production to train or run a Claude model.97%
still open · check by
last looked
first recorded · before the live record began
The odds were revised: 70% on → 86% on → 90% on → 95% on → 97% on .
as written in the record
Anthropic uses AWS Trainium chips in production to train or serve a Claude model (per SemiAnalysis / credible reporting)
ChatGPT ships in-chat checkout that buys third-party goods or services for users and takes a fee.90%
still open · check by
last looked
first recorded · before the live record began
The odds were revised: 72% on → 90% on .
as written in the record
ChatGPT ships a native agentic-commerce / in-chat checkout capability that completes third-party purchases (goods or services) on a user's behalf, monetized via transaction / affiliate / take-rate fees (per OpenAI announcement or credible reporting)
Nvidia ships a B30A-class Blackwell chip for China that beats the H20 on performance.52%
still open · check by
first recorded · before the live record began
as written in the record
Nvidia ships a Blackwell-based China SKU exceeding the H20 in performance (B30A-class), per SemiAnalysis / credible reporting
Claude Code writes at least 20% of public GitHub commits each day, on SemiAnalysis's count.48%
still open · check by
first recorded · before the live record began
The odds were revised: 35% on → 40% on → 48% on .
as written in the record
Claude Code authors ≥20% of daily public GitHub commits (per SemiAnalysis Tokenomics / GitHub methodology)
CoreWeave or another provider other than Microsoft serves OpenAI's production inference.85%
still open · check by
first recorded · before the live record began
as written in the record
CoreWeave (or another non-Microsoft provider) is serving OpenAI production inference workloads (per SemiAnalysis / credible reporting), breaking Microsoft's historical ~100% share of OpenAI inference
At year-end 2026, Meta's and SpaceXAI's best models each outrank Google's best Gemini on the Artificial Analysis index.55%
still open · check by
last looked
first recorded · live
as written in the record
At 2026 year-end, both Meta Superintelligence Labs' best public model and SpaceXAI's best public model rank above Google's best public Gemini model on the Artificial Analysis Intelligence Index
Neither Anthropic nor OpenAI sees its API revenue pace fall quarter over quarter in the second half of 2026.85%
still open · check by
first recorded · before the live record began
The odds were revised: 80% on → 85% on .
as written in the record
Neither Anthropic nor OpenAI's API/token-revenue run-rate posts a sequential (Q/Q) decline in any quarter of 2H26 — i.e. the enterprise "token-budget wall" does not dent aggregate API growth (per SemiAnalysis Tokenomics / credible reporting)
At year-end 2026, Meta has more total AI compute than OpenAI and more than Anthropic.65%
still open · check by
first recorded · live
as written in the record
Meta's realized total AI compute at year-end 2026 exceeds both OpenAI's and Anthropic's individually (per SemiAnalysis / credible reporting)
Cadence's organic revenue grows faster than Synopsys's, excluding Ansys, in fiscal 2026.72%
still open · check by
first recorded · before the live record began
as written in the record
Cadence's FY2026 organic revenue growth exceeds Synopsys's FY2026 organic (ex-Ansys) revenue growth (per company filings / SemiAnalysis)
DDR5 contract prices end 2026 at least 1.5 times their fourth-quarter 2025 level.75%
still open · check by
first recorded · before the live record began
The odds were revised: 68% on → 75% on .
as written in the record
DDR5 contract DRAM pricing exits 2026 (4Q26) at least 1.5× its 4Q25 level (per SemiAnalysis Memory Model / TrendForce contract-price data)
China publishes no comprehensive AI legislation draft by 2027-03-31 containing any of these five frontier controls: a compute/capability threshold, training-run registration, pre-release safety cases, mandatory incident reporting, or regulator shutdown authority80%
still open · check by
first recorded · live
as written in the record
China publishes no comprehensive AI legislation draft by 2027-03-31 containing any of these five frontier controls: a compute/capability threshold, training-run registration, pre-release safety cases, mandatory incident reporting, or regulator shutdown authority
Doosan Enerbility delivers at least one H-class gas turbine to an xAI Colossus site.68%
still open · check by
first recorded · before the live record began
as written in the record
Doosan Enerbility has shipped/delivered at least one H-class gas turbine to xAI for a Colossus site (per SemiAnalysis / credible reporting)
Nvidia ships the Groq-IP LPX (LP30) rack, built for the decode side of inference.70%
still open · check by
first recorded · before the live record began
as written in the record
Nvidia has shipped / made generally available the Groq-IP LPX (LP30) decode-specialized inference rack, per SemiAnalysis / credible reporting
At year-end 2026, US datacenters with their own on-site power run 2.5–3.5GW of IT capacity, on SemiAnalysis's model.70%
still open · check by
first recorded · live
as written in the record
US operational datacenter IT capacity powered behind the meter is 2.5-3.5GW at 2026 year-end, per SemiAnalysis Energy / Datacenter Model
Nvidia's fiscal 2027 filings show at least $75B of commitments backstopping cloud GPU capacity.15%
still open · check by
last looked
first recorded · live
The odds were revised: 55% on → 15% on .
as written in the record
Nvidia discloses cloud-service-agreement / minimum-revenue-guarantee commitments tied to GPU-capacity backstops of at least $75B in its fiscal-year 2027 filings
Meta signs or launches a deal to sell outside customers on-demand compute or hosted Claude or OpenAI access.80%
still open · check by
first recorded · before the live record began
The odds were revised: 75% on → 80% on .
as written in the record
Meta enters the merchant compute / token-as-a-service market — signs or launches a deal to sell external customers large-scale on-demand compute OR hosted private frontier-lab (Anthropic Claude / OpenAI) model instances or token access — per SemiAnalysis / credible reporting
China builds under about 1M domestic AI chips with HBM in 2026.62%
still open · check by
first recorded · before the live record began
as written in the record
China's total 2026 output of HBM-equipped domestic AI accelerators (Huawei Ascend + peers) stays below ~1M units, per SemiAnalysis
Broadcom's Scale-Up Ethernet ships in mass-production AI accelerator racks.58%
still open · check by
first recorded · before the live record began
as written in the record
Broadcom's Scale-Up Ethernet (SUE) is shipping in high-volume production AI accelerator racks (per SemiAnalysis / credible reporting)
Unitree goes public, with shares trading on the STAR Market or a similar exchange.75%
still open · check by
last looked
first recorded · before the live record began
as written in the record
Unitree completes an IPO with shares beginning to trade on a public exchange (Shanghai Stock Exchange STAR Market or equivalent), per credible reporting / filings
Google puts TPUs into production with a major customer beyond Anthropic: Meta, xAI, SSI or OpenAI.75%
still open · check by
first recorded · before the live record began
as written in the record
Google has deployed or sold TPUs (via GCP rental or direct sale) to at least one major external AI customer beyond Anthropic — Meta, xAI, SSI, or OpenAI — in production, per SemiAnalysis / credible reporting
Solaris Energy Infrastructure runs at least 1.1GW of gas turbines for xAI across both Colossus sites.55%
still open · check by
first recorded · before the live record began
as written in the record
Solaris Energy Infrastructure (SEI) has ≥1.1GW of gas turbines operating for xAI (Colossus 1+2 combined), per SemiAnalysis / SEI reporting
Nvidia ships the Rubin CPX accelerator, built for the prefill side of inference.60%
still open · check by
last looked
first recorded · before the live record began
The odds were revised: 78% on → 60% on .
as written in the record
Nvidia has shipped / made generally available the Rubin CPX prefill-specialized accelerator (in a VR NVL144 CPX or standalone VR CPX rack), per SemiAnalysis / credible reporting
AMD ships an MI450X-class rack that links at least 64 GPUs as one machine.80%
still open · check by
last looked
first recorded · before the live record began
The odds were revised: 60% on → 72% on → 80% on .
as written in the record
AMD has launched / shipped a rack-scale AI system (MI450X-class, CDNA-Next) with a single scale-up domain of ≥64 GPUs
Cerebras CS-4 hits at least 3,500 output tokens a second per user on GPT-5.6 Sol in a public test.65%
still open · check by
last looked
first recorded · live
as written in the record
Cerebras CS-4 demonstrates ≥3,500 output tokens/sec/user on GPT-5.6 Sol in a public Cerebras, SemiAnalysis, or InferenceX benchmark
NRG announces or signs an ERCOT gas deal of at least 1GW for 10+ years with a hyperscaler or investment-grade buyer.48%
still open · check by
last looked
first recorded · before the live record began
as written in the record
NRG (NRG) announces or signs a large-load BYOG / co-location gas supply deal in ERCOT — a block ≥1GW on a long-dated (10yr+) contract with a hyperscaler / investment-grade counterparty — per SemiAnalysis / credible reporting
CXMT goes public on Shanghai's STAR Market, with its shares trading.90%
still open · check by
last looked
first recorded · before the live record began
as written in the record
CXMT (Changxin Memory) completes its STAR Market IPO — shares begin trading on the Shanghai Stock Exchange (per SemiAnalysis / credible reporting / filings)
TeraWulf's Abernathy datacenter does not deliver its full promised capacity within 2026.65%
still open · check by
first recorded · before the live record began
as written in the record
TeraWulf's Abernathy (Fluidstack) datacenter does NOT deliver its full guided capacity within 2026 (slips into 2027+), per SemiAnalysis / credible reporting
A hyperscaler or Nvidia runs an 800V DC sidecar power rack in a live datacenter.62%
still open · check by
last looked
first recorded · before the live record began
as written in the record
At least one hyperscaler or Nvidia has deployed an 800VDC-distribution HVDC power rack (disaggregated sidecar) in a production datacenter (per SemiAnalysis / credible reporting)
Chinese firms account for at least 50% of industrial robots installed in China.80%
still open · check by
last looked
first recorded · before the live record began
as written in the record
Chinese domestic firms hold ≥50% share of China's industrial-robot installations (per IFR/SemiAnalysis)
Intel raises at least $10B more by selling new shares after June 11, 2026.55%
still open · check by
last looked
first recorded · before the live record began
as written in the record
Intel raises ≥$10B of additional primary equity capital after 2026-06-11 (a marketed follow-on / ATM program / equity-linked convertible, beyond the government + SoftBank + Nvidia + Altera round already closed), per Intel filings / SemiAnalysis / credible reporting
Nvidia launches a Rubin Ultra Kyber rack linking 144 GPU packages, 576 dies, as one NVLink domain.62%
still open · check by
last looked
first recorded · before the live record began
as written in the record
Nvidia has launched a Rubin Ultra–generation product built on the Kyber rack architecture (NVL576-class: 144 GPU packages / 576 compute dies in a single NVLink domain)
SemiAnalysis reports an open model matching or beating its era's first closed model within 90 days of its release.60%
still open · check by
first recorded · live
as written in the record
SemiAnalysis reports an open-source model matching or exceeding the first closed model of its post-agentic era on its era-specific curated composite within 90 days of that closed model's public release
SemiAnalysis's Dark Output Monitor logs a firm winning a court fight over AI work, or an insurer covering it.55%
still open · check by
first recorded · before the live record began
as written in the record
SemiAnalysis's Dark Output Monitor records at least one Tier-5 (a firm successfully defending AI-produced work in a court fight) or Tier-6 (an insurer underwriting AI-work risk) displacement activity, per SemiAnalysis
Apple ships a product with camera image sensors made by Samsung Foundry.60%
still open · check by
first recorded · before the live record began
as written in the record
Apple is shipping a product with CMOS image sensors fabricated by Samsung Foundry (breaking Sony's iPhone-CIS exclusivity), per SemiAnalysis / credible reporting
Single-phase direct-to-chip liquid cooling, not immersion, leads new AI racks above 50kW.85%
still open · check by
first recorded · before the live record began
as written in the record
Single-phase direct-to-chip liquid cooling (DLC), not immersion, is the dominant cooling method for newly-deployed high-density (>50kW/rack) AI accelerator racks
Microsoft's Maia still ships far fewer AI chips than Google TPU, AWS Trainium and Meta MTIA.82%
still open · check by
first recorded · before the live record began
as written in the record
Microsoft remains last among the four in-house-ASIC hyperscalers — its Maia line still ships materially fewer AI-accelerator units than Google TPU, AWS Trainium, AND Meta MTIA (per SemiAnalysis Accelerator Model)
A shipping HBM4 accelerator uses a custom logic-process HBM base die that adds function beyond a standard one.70%
still open · check by
first recorded · before the live record began
The odds were revised: 62% on → 70% on .
as written in the record
A shipping HBM4-generation AI accelerator (e.g. Nvidia Rubin or AMD MI400) uses a custom, logic-process HBM base die adding function beyond a standard off-the-shelf base die (memory-controller offload, off-package LPDDR / 2nd-rank memory, or beachfront I/O), per SemiAnalysis / credible reporting
AI chips take at least 80% of TSMC's N3-family wafer output in 2027, on SemiAnalysis's model.65%
still open · check by
first recorded · before the live record began
The odds were revised: 60% on → 65% on .
as written in the record
AI (accelerator + host CPU + networking) accounts for ≥80% of TSMC N3-family wafer output in 2027, per SemiAnalysis Foundry / Accelerator Model
xAI's Memphis water-recycling plant is running and supplying Colossus's cooling water.68%
still open · check by
last looked
first recorded · before the live record began
as written in the record
xAI's Colossus (Memphis) water-recycling plant (recycling T.E. Maxson municipal wastewater) is operational and supplying the site's cooling water in production, per SemiAnalysis / credible reporting
Anthropic posts at least $1B of GAAP operating profit in the third quarter of 2026.65%
still open · check by
last looked
first recorded · live
as written in the record
Anthropic reports or is credibly reported to have generated ≥$1B GAAP EBIT in 3Q26
SemiAnalysis's modeled Vera Rubin rack cost tops $42 per watt, over 10% above its $38.1-per-watt debut.40%
still open · check by
last looked
first recorded · before the live record began
as written in the record
SemiAnalysis's modeled Vera Rubin rack capex-per-watt exceeds ~$42/W (>10% above the ~$38.1/W VR NVL72 debut level), signaling Nvidia lifting system pricing toward value rather than holding the roughly-flat GB300→VR capex/watt
Qualcomm ships its SD2-generation datacenter CPU for revenue.60%
still open · check by
first recorded · before the live record began
as written in the record
Qualcomm (QCOM) ships its SD2-generation datacenter CPU for revenue (per SemiAnalysis / credible reporting)
ERCOT puts its Performance Credit Mechanism capacity-market reform into effect.55%
still open · check by
last looked
first recorded · before the live record began
as written in the record
ERCOT implements its Performance Credit Mechanism (PCM) capacity-market reform (per PUCT/ERCOT or SemiAnalysis)
One frontier training run reaches 1e27 FLOP, about 100 times GPT-4, on Epoch AI's estimates.57%
still open · check by
first recorded · before the live record began
The odds were revised: 55% on → 57% on .
as written in the record
At least one frontier AI training run reaches ≥1e27 FLOP (≈ +2 OOMs over GPT-4's ~1e25) per Epoch AI's training-compute estimates — the physical-compute leg of the ~5-OOM-by-2027 "counting the OOMs" forecast
Nvidia's Rubin Ultra Kyber rack still takes pluggable optics or cables for scale-out, not on-board optics only.70%
still open · check by
first recorded · before the live record began
as written in the record
Nvidia's production Rubin Ultra Kyber rack supports pluggable optical transceivers / AECs for scale-out (retains an OSFP-cage option, not a Mid-Board-Optical-Module-only design), per SemiAnalysis / credible reporting
OpenAI's Jalapeño serves production inference on hardware past the engineering-sample stage.78%
still open · check by
last looked
first recorded · live
as written in the record
OpenAI Jalapeño is serving production inference tokens on non-engineering-sample hardware, per OpenAI, SemiAnalysis, or credible reporting
Arm ships its own full Phoenix datacenter CPU, with Meta as a customer.70%
still open · check by
first recorded · before the live record began
as written in the record
Arm Holdings (ARM) has shipped / made generally available its own full datacenter CPU (Phoenix) with Meta as a customer (per SemiAnalysis / credible reporting)
Anthropic runs a Claude model in production on Cerebras WSE hardware.55%
still open · check by
last looked
first recorded · before the live record began
as written in the record
Anthropic is serving a Claude model in production on Cerebras WSE hardware (via AWS Bedrock prefill/decode disaggregation or directly), per SemiAnalysis / credible reporting
Oracle announces or breaks ground on a datacenter it owns rather than leases.62%
still open · check by
last looked
first recorded · before the live record began
as written in the record
Oracle (ORCL) has announced or broken ground on a self-built (owned, not leased) datacenter (per SemiAnalysis or credible reporting)
No flagship phone chip from Apple, Qualcomm or MediaTek ships on a TSMC A16-class backside-power node.80%
still open · check by
last looked
first recorded · before the live record began
as written in the record
No flagship smartphone SoC (Apple / Qualcomm / MediaTek) has shipped on a TSMC backside-power node (A16-class)
DDR and LPDDR stay at least 90% of CXMT's revenue through 2026, with HBM minimal.80%
still open · check by
last looked
first recorded · before the live record began
as written in the record
CXMT's revenue stays overwhelmingly commodity DRAM — DDR + LPDDR ≥90% of revenue, HBM a minimal contributor — through 2026 (per SemiAnalysis Memory Model / CXMT filings)
Meta's Hyperion campus in Louisiana has at least 1.5GW of critical IT power running.70%
still open · check by
first recorded · before the live record began
The odds were revised: 55% on → 70% on .
as written in the record
Meta's Hyperion (Louisiana) campus has ≥1.5GW of critical IT power in operation / phase-1 (per SemiAnalysis Datacenter Industry Model or credible reporting)
A DRAM maker reaches mass production on the 1d node, the step after 1c.72%
still open · check by
first recorded · before the live record began
as written in the record
At least one DRAM maker has 1d-node (post-1c) DRAM in high-volume production (per SemiAnalysis / credible reporting)
AWS announces or ships a Trainium4 that uses Nvidia's NVLink Fusion to link chips.45%
still open · check by
first recorded · before the live record began
as written in the record
AWS has announced or shipped a Trainium4 accelerator with a scale-up track using Nvidia's NVLink (NVLink Fusion) interconnect, per SemiAnalysis / credible reporting
Micron supplies under 5% of Nvidia's Rubin R200 HBM in Rubin's first 12 months of shipments.62%
still open · check by
last looked
first recorded · before the live record began
The odds were revised: 55% on → 62% on .
as written in the record
Micron holds <5% share of Nvidia R200 (Rubin) HBM supply in Rubin's first 12 months of shipment (SK Hynix + Samsung split ~all of it), per SemiAnalysis
China's output of competitive datacenter AI chips stays far below the West's ~5M units a year.90%
still open · check by
first recorded · before the live record began
The odds were revised: 85% on → 88% on → 85% on → 90% on .
as written in the record
China's domestic production of competitive datacenter AI accelerators remains far below the ~5M units/yr scale of Western-ecosystem demand
Oracle's Project Jupiter campus in New Mexico reaches neither first power nor first customer delivery before 2028.45%
still open · check by
last looked
first recorded · before the live record began
The odds were revised: 80% on → 45% on .
as written in the record
Oracle's STACK Infrastructure New Mexico Stargate campus (Project Jupiter, Doña Ana County) has NOT reached first power / commercial customer delivery before 2028 — i.e. its 10-June-2026 guide of 1H2027 delivery slips — per SemiAnalysis / credible reporting
Google Cloud's 2027 revenue is at least 2.4 times its 2026 revenue.55%
still open · check by
last looked
first recorded · live
as written in the record
Alphabet reports Google Cloud full-year 2027 revenue at least 2.4× full-year 2026 revenue (≥140% YoY growth)
The US adds at least 38GW of datacenter IT capacity in calendar 2027, on SemiAnalysis's model.60%
still open · check by
first recorded · live
as written in the record
US datacenter IT capacity additions in calendar 2027 reach at least 38GW, per SemiAnalysis Datacenter Industry Model
Anthropic and OpenAI take more than 27% of new AI compute added in 2027, on SemiAnalysis's model.70%
still open · check by
first recorded · live
as written in the record
Anthropic + OpenAI account for >27% of net-new AI-compute GW in calendar 2027 in SemiAnalysis's Tokenomics/Datacenter model, including indirect hyperscaler capacity
SK Hynix's average 2027 HBM4 price to Nvidia runs at least 10% below Samsung's, on SemiAnalysis's numbers.70%
still open · check by
first recorded · live
as written in the record
SemiAnalysis's average 2027 HBM4 price to Nvidia from SK Hynix is at least 10% below its average price to Nvidia from Samsung
SpaceX and xAI's ground datacenters reach at least 10GW of compute at year-end 2027.55%
still open · check by
first recorded · live
as written in the record
SpaceX/xAI's terrestrial datacenter compute capacity reaches at least 10GW at 2027 year-end, per SemiAnalysis / credible reporting
Anthropic and OpenAI together hold at least 56.3% of all AI-lab compute at the end of 2027, on SemiAnalysis's model.50%
still open · check by
first recorded · live
as written in the record
Anthropic + OpenAI account for ≥56.3% of all lab compute at 2027 year-end in SemiAnalysis's Datacenter Industry Model
Microsoft contracts for at least 3GW of SpaceX/xAI compute by the end of 2027.40%
still open · check by
first recorded · live
as written in the record
Microsoft has contracted for at least 3GW of SpaceX/xAI compute capacity by 2027 year-end, per company disclosure / SemiAnalysis / credible reporting
Boom Supersonic's Superpower turbine enters commercial production or first powers a datacenter.45%
still open · check by
first recorded · before the live record began
as written in the record
Boom Supersonic's Superpower aeroderivative gas turbine has entered commercial production or delivered its first datacenter power (per SemiAnalysis / credible reporting)
SpaceX's annual revenue run-rate reaches at least $300B at the end of 2027.30%
still open · check by
first recorded · live
as written in the record
SpaceX's annualized revenue run-rate reaches at least $300B at 2027 year-end, per company disclosure / SemiAnalysis / credible reporting
GE Vernova, Siemens Energy and Mitsubishi take fewer new gas-turbine orders in 2027 than in 2026.55%
still open · check by
first recorded · before the live record began
as written in the record
GE Vernova + Siemens Energy + Mitsubishi combined new gas-turbine order intake for calendar-2027 comes in below their calendar-2026 level (per SemiAnalysis Energy Model / company disclosure), confirming 2026 as the order peak
Google's TPU v8 line includes a MediaTek co-designed model, Zebrafish, in production.62%
still open · check by
last looked
first recorded · before the live record began
as written in the record
Google's TPU v8 generation includes a MediaTek-co-designed SKU (Zebrafish / TPU 8X) in production, per SemiAnalysis / credible reporting
Anthropic ends 2027 at an annual revenue run-rate of at least $300B.45%
still open · check by
first recorded · live
as written in the record
Anthropic's ending ARR for 2027 reaches ≥$300B (per SemiAnalysis Tokenomics / Anthropic filing / credible reporting)
Intel's 14A node uses high-NA EUV on at least one layer in volume or risk production.70%
still open · check by
first recorded · before the live record began
as written in the record
Intel's 14A node uses high-NA EUV (ASML EXE-class) for at least one layer in volume/risk production (per Intel or SemiAnalysis disclosure)
AWS announces or ships a Trainium4 with co-packaged optics or an optical scale-up link.55%
still open · check by
first recorded · before the live record began
as written in the record
AWS Trainium4 ships or is announced with a co-packaged-optics / optical scale-up interconnect (Celestial AI "Photonic Fabric" or equivalent), per SemiAnalysis / credible reporting
No fab uses nanoimprint lithography in mass production for a critical layer of advanced logic or DRAM.90%
still open · check by
first recorded · before the live record began
as written in the record
No fab has used nanoimprint lithography (NIL) in high-volume manufacturing for a critical layer of an advanced logic or DRAM chip (per SemiAnalysis / credible reporting)
Apple qualifies or ships a production chip on an Intel Foundry node: 18A, 18A-P or 14A.35%
still open · check by
first recorded · before the live record began
as written in the record
Apple has qualified or shipped a production chip (any product) on an Intel Foundry node (18A / 18A-P / 14A), per SemiAnalysis / credible reporting
AMD ships an MI500-generation rack linking at least 128 GPU packages as one machine.45%
still open · check by
first recorded · before the live record began
as written in the record
AMD has launched / shipped an MI500-generation rack-scale AI system with a single scale-up domain of ≥128 GPU packages (UAL128/UAL256-class)
AMD or a hyperscaler announces or ships a prefill inference chip with GDDR-class memory, not HBM.55%
still open · check by
last looked
first recorded · before the live record began
as written in the record
At least one non-Nvidia AI-accelerator program (AMD or a hyperscaler ASIC) has announced or shipped a prefill-specialized inference accelerator that omits HBM in favor of GDDR-class memory, per SemiAnalysis / credible reporting
No NAND maker mass-produces 3D NAND that stores five bits per cell.85%
still open · check by
first recorded · before the live record began
as written in the record
No NAND maker has 5-bits-per-cell (penta-level-cell) 3D NAND in high-volume production, per SemiAnalysis / credible reporting
Meta's Rivos-derived Phoebe accelerator finishes its chip design.45%
still open · check by
first recorded · live
as written in the record
Meta's Rivos-derived Phoebe accelerator has taped out (per SemiAnalysis / credible reporting)
A shipping Huawei Kirin or Ascend chip splits one active logic block across bonded 3D tiers.50%
still open · check by
first recorded · before the live record began
as written in the record
A shipping Huawei/HiSilicon product (Kirin or Ascend) uses "LogicFolding" — 3D-stacked ACTIVE logic splitting one logic block across face-to-face hybrid-bonded tiers (beyond cache/IO stacking) — per SemiAnalysis / credible reporting
Nvidia's Rubin Ultra Kyber rack still uses copper, not co-packaged optics, for NVLink scale-up.80%
still open · check by
first recorded · before the live record began
as written in the record
Nvidia's Rubin Ultra (Kyber / NVL576-class) rack still uses COPPER — not co-packaged optics — for its NVLink scale-up interconnect, per SemiAnalysis / credible reporting
Musk's Terafab has not reached 100,000 wafer starts a month.75%
still open · check by
first recorded · before the live record began
as written in the record
Musk's Terafab has NOT reached 100k wafer-starts-per-month of production output (its stated 2028 mass-production entry ramp), per SemiAnalysis / credible reporting
TSMC's Arizona Fab 21 holds under 20% of its advanced-node, N5-and-below wafer capacity.90%
still open · check by
first recorded · before the live record began
The odds were revised: 88% on → 90% on .
as written in the record
TSMC Arizona (Fab 21) accounts for less than 20% of TSMC's total advanced-node (EUV-using / N5-and-below) wafer capacity, per SemiAnalysis / credible reporting
A production AI accelerator ships with 4-Hi HBM4 or HBM4E stacks.68%
still open · check by
first recorded · live
as written in the record
At least one production AI accelerator ships with 4-Hi HBM4 or HBM4E stacks, per SemiAnalysis / credible reporting
Google's TPU v9 uses Intel's EMIB-T advanced packaging.55%
still open · check by
first recorded · before the live record began
as written in the record
Google's TPU v9 uses Intel EMIB-T advanced packaging (per SemiAnalysis / credible reporting)
Modular, prefabricated builds are at least 30% of live datacenter capacity at end-2028, on SemiAnalysis's model.65%
still open · check by
first recorded · live
as written in the record
Modular/prefabricated datacenter construction accounts for ≥30% of total live datacenter capacity at 2028 year-end, per the SemiAnalysis Industrials Model
On-site generation covers more than 50% of new US datacenter capacity added in 2028, on SemiAnalysis's model.80%
still open · check by
first recorded · before the live record began
The odds were revised: 55% on → 60% on → 68% on → 78% on → 80% on .
as written in the record
Behind-the-meter / onsite generation accounts for >50% of new US datacenter capacity (MW) additions in 2028, per SemiAnalysis Datacenter Industry Model
By mid-2029 Microsoft, Google, Meta, Amazon and Oracle double revenue and cut operating costs by $80B or more.15%
still open · check by
first recorded · live
as written in the record
In the latest annual filings available by 2029-07-31, aggregate revenue for MSFT, GOOGL, META, AMZN, and ORCL is ≥2× the aggregate in their latest annual filings available on 2026-07-31, while aggregate reported operating expenses are ≥$80B lower on the same basis
A production AI accelerator rack ships an optical scale-up link built to the OCI-MSA spec.40%
still open · check by
last looked
first recorded · before the live record began
as written in the record
At least one production AI accelerator rack ships an optical (DWDM or CPO) scale-up interconnect built to the OCI-MSA optical-scale-up specification, per SemiAnalysis / credible reporting
China's big cloud companies have at least 3.92GW of leased datacenter capacity outside China at the end of 2029, on SemiAnalysis's model.70%
still open · check by
first recorded · live
as written in the record
Chinese hyperscalers have at least 3.92GW of overseas leased datacenter capacity at 2029 year-end, per SemiAnalysis Datacenter Industry Model
SemiAnalysis reports Anthropic as the biggest TPU user in 2029, ahead of Google DeepMind.70%
still open · check by
last looked
first recorded · live
as written in the record
SemiAnalysis reports Anthropic as the largest TPU user in calendar 2029, ahead of Google DeepMind
South Korea runs at least 8.4GW across its three datacenter megaprojects due by end-2029.60%
still open · check by
first recorded · live
as written in the record
South Korea has at least 8.4GW of operational datacenter capacity across the three megaprojects targeted for completion by year-end 2029, per SemiAnalysis / credible reporting
No foundry mass-produces transistors with 2D-material (TMD) channels.94%
still open · check by
first recorded · before the live record began
The odds were revised: 92% on → 93% on → 94% on .
as written in the record
No foundry has 2D-material (TMD) channel transistors in high-volume manufacturing
The Middle East reaches at least 6GW of operating datacenter capacity, on SemiAnalysis's model.72%
still open · check by
first recorded · before the live record began
as written in the record
Middle East operational datacenter capacity reaches ≥6GW (per SemiAnalysis Datacenter Industry Model)
No AI datacenter of 1MW or more is operating in Earth orbit.80%
still open · check by
first recorded · before the live record began
as written in the record
No scaled AI datacenter (≥1 MW of critical IT compute) is operating in Earth orbit, per SemiAnalysis / credible reporting
No fab uses X-ray lithography in mass production for a critical layer of an advanced logic chip.85%
still open · check by
first recorded · before the live record began
as written in the record
No fab (Substrate or otherwise) has used X-ray lithography (XRL) in high-volume manufacturing for a critical layer of an advanced logic chip (per SemiAnalysis / credible reporting)
TSMC, Samsung or Intel replaces copper with ruthenium in its lowest metal layers on a volume node.50%
still open · check by
first recorded · before the live record began
as written in the record
At least one leading-edge foundry (TSMC / Samsung / Intel) has introduced ruthenium (Ru) as the main conductor for its lowest metal layers (M0/M1-class), replacing copper, in a high-volume logic node, per SemiAnalysis / credible reporting
Alibaba, Tencent, Huawei, ByteDance and Baidu together have at least 20.58GW of datacenter capacity at the end of 2030, on SemiAnalysis's China model.70%
still open · check by
first recorded · live
as written in the record
Alibaba, Tencent, Huawei, ByteDance, and Baidu have combined datacenter capacity of at least 20.58GW at 2030 year-end, per SemiAnalysis China Datacenter Model
No foundry mass-produces complementary-FET (CFET) logic transistors.85%
still open · check by
first recorded · before the live record began
as written in the record
No foundry has complementary-FET (CFET) logic transistors in high-volume manufacturing
Checked against SemiAnalysis's reporting, company filings, or credible press.
Nothing here is a result yet.