Every major group tried to own more of its stack, and spending now outruns cash
Between Oct 9 and Oct 22, 2026, the big tech groups made fewer headline launches than financing and positioning moves. Pick a conclusion to see the evidence behind it.
Every major group tried to own more of its stack, and none left the shared supply of models
Microsoft announced the Maia 300 chip for 2027 and finished pretraining MAI-Base-2. Google opened Gemini 4 Argon to paying users on Oct 13, and Meta shipped Muse Wonder on Oct 12. Yet Copilot still runs OpenAI and Anthropic models, and Google Cloud began hosting OpenAI's GPT-6 models in private preview on Oct 21.
Quality is near parity, so price and distribution decide
On the Artificial Analysis index the top models of four labs sit within six points, 58 to 52. Cost per index task ranges from $0.72 (GPT-6.1 Sol) to $5.98 (Claude Opus 5.5), about 8.3 times, and Argon's introductory API price of $2 / $10 per million tokens doubles on Dec 1.
Agents became the product battleground, and control of agents became the selling point
Microsoft added Guardian Agents on Oct 19, AWS argues for policy enforced at the gateway, and Google's Gemini app now runs tasks across more than 40 connected apps. Adoption claims are still thin: Meta says more than 10 million people set up Muse, while the Wall Street Journal reports nearly 9 million downloads and weak retention.
Spending now outruns cash
Meta's Q3 capex of $34.91B exceeded operating cash flow of $32.79B, leaving free cash flow of negative $2.12B, and a $9.89B legal charge cut operating income 37%. Alphabet's Q2 free cash flow was already negative, and Alibaba has spent RMB 190B of a RMB 380B plan (company guidance, via Zacks). Alphabet, Microsoft and Amazon report on Oct 28-29.
China runs on a separate track of domestic chips and controlled distribution
Huawei powered on its first 1,024-card Atlas 950 SuperPoD on Oct 15. Apple's China launch of Apple Intelligence is reported to run with Alibaba-built guardrails and without Siri AI. IDC puts Nvidia at 34% and Huawei at 28% of China's AI-accelerator market in H1 2026.
Seven groups made seven different bets in the same fortnight
Choose a company to see its dated timeline and the figures that matter, each with its status label.
- Oct 10PublishedNadella blog: model controls must sit outside the model.
- Oct 12AnnouncedMaia 300 for multi-gigawatt scale from 2027; Maia 200 is in production.
- Oct 15Post-training, no public accessMAI-Base-2 pretraining complete (The Elec).
- Oct 19AnnouncedGuardian Agents added to Project Perception and Microsoft 365 E7; per-feature dates not stated.
- Sep 30ContextGemini 4 Argon announced; at first only approved cyber defenders could use it.
- Oct 13AvailableArgon opens to paying users; the Gemini app runs tasks across 40+ apps; Skills begin rolling out.
- Oct 13On holdSkills in Workspace apps "delayed until further notice".
- Oct 21Private previewOpenAI GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna in Model Garden; general availability "in the coming weeks".
- Oct 28UpcomingAlphabet Q3 results.
- Oct 12ShippingMuse Wonder and Wonder Full released; no downloadable weights; safety report still outstanding.
- Oct 20-21ReportedJoel Kaplan plans to leave; C.J. Mahoney takes policy duties; no effective date.
- Oct 21Results and guidanceQ3 revenue $64.60B (+26%); 2026 capex guide $135-145B.
- Jan 19, 2027AnnouncedNon-voting Class C stock takes effect.
- Sep 29Public previewBedrock Managed Agents, powered by OpenAI (just before the window).
- Oct 7Secondary sourceCost allocation by IAM principal now covers the bedrock-mantle endpoint.
- Oct 19PublishedAWS "Trust in Depth": agent limits belong at the gateway, outside the agent's reasoning.
- Oct 21Announced3M and AWS expanded collaboration on optical connectivity for data centers.
- Oct 29UpcomingAmazon Q3 results.
- Oct 13AnnouncedSiri AI in HomePod mini 2, Apple TV 4K and HomeView.
- Oct 16Pre-ordersiPhone Duo; Siri AI in beta, English, with iOS 27.
- Oct 30UpcomingHomePod mini 2 ships; Siri AI "in a future software update".
- Nov 13UpcomingApple TV 4K and HomeView ship with Siri AI at launch.
- DecemberReportedChina launch of Apple Intelligence in iOS 27.3, without Siri AI (Bloomberg).
- Oct 10In effectAlibaba retires 47 legacy API model IDs.
- Oct 14ReportedByteDance Seed creates a model-alignment and preparedness department; 151 apps block Doubao Phone Assistant.
- Oct 15In effectQwen3.8-Flash-Next importable into OCI Generative AI.
- Oct 15ReleasedTencent Hy-Memory-30B-A3B under Apache 2.0 (per ibl.ai).
- Oct 15-16Powered onHuawei Atlas 950 SuperPoD (1,024 cards); a 10,000-card Ascend pool launched with China Telecom.
With near-parity quality, price per task decides the choice
Artificial Analysis (AA) Intelligence Index snapshot of Oct 12, as reported by VentureBeat, and AA cost per index task as reported by The Decoder. Switch the view to see how the order changes.
Price card and access, per 1M tokens (input / output)
| Model | List price | Route | Status (as of Oct 23) |
|---|---|---|---|
| Claude Opus 5.5 | $4 / $20 | Claude API, Bedrock, Vertex, Foundry | GA released Sept 22 |
| Claude Sonnet 5.5 | $2 / $10 | Claude API, clouds | GA released Sept 27 or 28 (sources differ) |
| GPT-6.1 Sol | $2 / $10 | OpenAI API, ChatGPT Work, Codex | GA late September (Sept 22 vs Sept 29 in sources) |
| GPT-6 Astra | $10 / $50 | OpenAI API, ChatGPT paid plans | GA Sept 3 |
| Gemini 4 Argon | $2 / $10, then $4 / $20 | Gemini API (paid tier), Gemini app | GA Oct 13; intro price in effect to Nov 30 |
| Muse Wonder | $4 / $20 | Meta Model API, OpenRouter, Muse Code | Shipping since Oct 12 (Meta); one tracker labels the API "preview" |
| Muse Wonder Full | No token price | Muse Code Power $50 or Maximum $200 a month | Subscription only |
How far to trust the charts
In BenchLM's October Arena text snapshot, Argon leads at 1532 (preliminary, 4,904 votes), ahead of Opus 5.5 at 1516 and Fable 5.1 at 1512. On Arena WebDev, Opus 5.5 scores 1813 against Argon's 1677.8.
Argon was measured at high effort, the others at max. RecoupRev reports Opus 5.5 at about 54 on high effort against 58 at max, so a one-point gap between models says little.
Meta's model card (vendor-reported) puts Wonder Full at 57.5 on SWE-Marathon against Opus's 46.3, yet at 57.1 it trails Opus (59.6) and Astra (59.1) on Terminal-Bench 4.0-AA. Use such charts for direction only.
Own stack is a hedge, not an exit, because the independent labs still supply the shelf
Select a company to see what it builds and what it partners or hosts, or switch to the agent-platform comparison.
In-house MAI models; MAI-Base-2 is a pretrained base with no product yet.
Foundry hosts Claude Opus 5.5 and GPT-6 Astra; OpenAI must now be switched on as a subprocessor in Copilot, the first time and not by default.
Copilot Studio, Agent 365 registry and Entra Agent ID. Agents inside the Microsoft graph, "limited to Microsoft surfaces".
Gemini Enterprise Agent Platform (Vertex AI plus Agentspace), 200+ models including Claude; Gemini app Tasks mode with 40+ apps (Oct 13, Pro and Ultra).
Bedrock AgentCore; Web Search tool GA Aug 21 in us-east-1; works with OpenAI models.
Muse Code (plans $5 to $200); Wonder via API, Wonder Full on Power and Maximum plans; Muse app soon, subscribers first.
Capex now outruns the cash that is meant to fund it
Only Meta reported Q3 inside the window. Alphabet and Microsoft report on Oct 28 and Amazon on Oct 29, after the close. Definitions differ (Meta and Microsoft include finance leases; Amazon's is cash capex), so do not rank on headlines alone.
Commitments are announced faster than capacity is delivered, and China runs a separate chip track
Most links behind the capex are announced or reported, not delivered. Switch the view for the compute contracts or the China accelerator market.
Status labels separate what is live from what is only announced
Every dated change in the window and the key dates ahead, with the state on Oct 23, 2026. Filter by company or status.
| Date | Company | Event | State on Oct 23 |
|---|
Choose by where your data lives, keep models swappable, and price every agent per task
The companies compared side by side, then the implication for your profile.
Strategy side by side
| Company | Stance | Chips | Flagship model | 2026 capex stance |
|---|---|---|---|---|
| Microsoft | Builds its own models, still sells OpenAI and Anthropic inside Copilot | Maia 300 announced for 2027; Maia 200 in production | MAI-Base-2 (pretrained, no public access) | About $175B calendar 2026 (reported) |
| Alphabet | Builds Gemini and also hosts OpenAI's models | Rents third-party GPU capacity (reported) | Gemini 4 Argon, GA Oct 13 | $195-205B (Jul 22 guidance) |
| Meta | Builds a proprietary, tool-locked top model | No chip news in the window | Muse Wonder Full / Wonder | $135-145B (Oct 21 guidance) |
| Amazon | Hosts every lab's models on Bedrock | No own-chip news in the window | No own-model score in sources | About $220B cash capex (Jul 30) |
| Apple | Own models for Siri AI; China features built with Alibaba (reported) | On-device and Private Cloud Compute | Siri AI (beta, English) | No figure in sources |
| Alibaba | Builds Qwen, partners with Apple | Zhenwu V900 (mass production Q1 2027) | Qwen3.8 Max | RMB 380B three-year plan |
| Huawei | Builds the domestic chip stack | Atlas 950 SuperPoD live; Ascend 960DT Q1 2027 | None | Not applicable |
What it means for you
Agent platforms are sold as extensions of each suite. Microsoft, AWS and Google all stress security and policy outside the model.
Pick the platform where your data already lives, and make agent permissions and logging a contract term, not a feature.
Guardian Agents (announced Oct 19); AWS gateway policy (Oct 19); Gemini Enterprise Agent Platform with 200+ models.
Top models sit within six index points; Argon's API price doubles on Dec 1; Wonder Full is locked to Muse Code.
Keep the model layer swappable, test on your own tasks, and budget per task, not per list price.
Index 58 to 52; cost per task $0.72 to $5.98; Argon $2 / $10 in effect to Nov 30, then $4 / $20.
Meta's free cash flow turned negative; Alphabet's was negative in Q2; Alibaba's free cash outflow was 2.4 times last year's.
The question is no longer demand but funding: watch debt, equity issues and the 2027 capex guides. Oct 28-29 is the first real test.
Meta -$2.12B (Q3); Alphabet -$5.9B (Q2); Alibaba RMB 44.67B outflow (June quarter, via Zacks).
MAI-Base-2 has no product yet, while OpenAI and Anthropic models are what Copilot customers use today. The clouds are the shelf.
Index span 58 to 52, but cost per task runs about 8.3 times from $0.72 to $5.98.
Meta's $32.79B of operating cash flow did not cover $34.91B of capex. China runs a separate track: Huawei 28% vs Nvidia 34%.