This observation covers changes formally announced or reliably verified around the period from 8:17 am on 8 September to 8:17 am on 9 September 2026 in Australia/Sydney time. Four items are included.
OpenAI publishes finite-time Navier–Stokes singularity proof materials
On 8 September, OpenAI released a 166-page proof manuscript and a corresponding Lean 4 formalisation repository. It claims that an internal system constructed a three-dimensional incompressible Navier–Stokes flow that begins at rest, is acted on by a smooth external force, maintains bounded kinetic energy and develops unbounded velocity in finite time. The claim addresses alternatives C and D in the Clay Mathematics Institute's official problem statement; Clay's original document does list those breakdown alternatives as valid ways of resolving the problem. OpenAI reports that about 10,000 concurrent agents completed the main result in 88 hours, exchanging 2.7 million messages and producing about 130 billion output tokens, followed by 17 hours of Lean formalisation and verification using GPT-6 Astra. These operating figures and capability assessments are developer-reported.
Relative to the earlier limited availability of GPT-6 Astra, the new fact is a concrete research result generated by an unreleased internal model that OpenAI says is significantly more capable than Astra and is still being trained, together with a downloadable manuscript and buildable Lean repository. Publishing the materials is not equivalent to independent acceptance of the proof. At the time of this review, there was no peer-reviewed decision, confirmation from the Clay Mathematics Institute or published complete independent reproduction. The Guardian independently reported the release, OpenAI's disclosed compute scale and the continuing dispute, but did not validate the proof itself. Direct sources: OpenAI, “On the Navier–Stokes Millennium Prize Problem”, 8 September 2026, https://openai.com/index/navier-stokes-solution/; OpenAI Lean repository, 8 September 2026, https://github.com/openai/NavierStokesAndEuler; The Guardian, 8 September 2026, https://www.theguardian.com/science/2026/sep/08/openai-claims-to-have-solved-maths-problem-that-stumped-humans-for-decades
Meta launches the Muse personal agent for US adults
On 8 September, Meta began offering Muse to users aged 18 and over in the United States through a dedicated app and WhatsApp. Following the 2 September release of the Muse Spark 1.3 model and API, the change is the deployment of that model in a consumer agent that can keep working in the background, access authorised services and perform actions for a user. Meta's examples include sending email, booking travel, filling forms, negotiating and completing purchases through Stripe Link. Users can connect or revoke access to services individually.
Each Muse instance runs in a dedicated Muse Secure VM. Meta's technical description says the main agent runs in a restricted Linux container, credentials are held by isolated services, and Sentinel is the sole permission authority for connector actions and network egress; sensitive actions require user approval and are recorded in an audit trail. These are developer descriptions and have not yet received a complete independent security audit. Reuters verified the launch scope and reported that internal tests still produced disconnections, unprompted data uploads and exposure of personal iCloud photos. Meta said the system had met its release threshold but did not address those test incidents individually. The Confidential VM, planned to encrypt the entire environment later this year with a user-held key, has not been delivered. Direct sources: Meta launch announcement, 8 September 2026, https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/; Meta security technical note, 8 September 2026, https://research.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-muse; Reuters, 8 September 2026, https://www.reuters.com/business/meta-launches-ai-agent-that-can-access-other-apps-send-emails-make-payments-2026-09-08/
Firmus and OpenAI sign Malaysian compute-capacity agreement
Firmus and OpenAI announced a multi-year partnership on 8 September under which OpenAI will obtain dedicated compute capacity from two Firmus AI Factory sites in Malaysia and become an anchor customer. Firmus says its contracted capacity across all customers now exceeds 900 MW after the agreement. That figure is neither the capacity of OpenAI's contract alone nor operational capacity. Of Firmus's seven sites, two in Australia and Singapore are operating; the other five remain under development, with service targeted over the next 24 months.
Firmus says the partnership sites will use the NVIDIA DSX AI Factory Platform, Vera Rubin NVL72 rack-scale systems and Firmus HyperCube liquid-cooling infrastructure. The contract value, capacity allocated to OpenAI, construction progress at the Malaysian sites, energisation date and measured performance were not disclosed. Reuters independently confirmed the contract, anchor-customer relationship and total contracted capacity, and reported that five sites remain in development. Direct sources: Firmus announcement, 8 September 2026, https://firmus.co/newsroom/firmus-surpasses-900-mw-contracted-capacity-adds-openai-as-anchor-customer-and-expands-into-malaysia; Reuters, 8 September 2026, https://www.reuters.com/world/asia-pacific/nvidia-backed-firmus-signs-deal-with-openai-malaysia-data-centre-capacity-2026-09-08/
Qualcomm and Amazon sign multi-generation AI data-centre chip collaboration
On 8 September, Qualcomm announced a multi-generation product collaboration with Amazon to co-develop customised inference silicon for large AWS AI data centres and optical interconnect solutions extending to 1.6 Tb/s and later generations. Qualcomm also plans to expand its use of AWS infrastructure and Amazon Bedrock for electronic design automation workloads. Reuters, citing a regulatory filing, reported that the agreement allows Amazon to buy up to US$60 billion in AI data-centre chips and related products over the long-term partnership and gives Amazon purchase-linked warrants to acquire Qualcomm shares at US$161.26 each.
The confirmed change is the contract and joint development relationship, not delivery or deployment of US$60 billion in hardware. The parties did not disclose specifications for the first customised chip, tape-out or production dates, delivery timing, deployment locations or measured inference performance. The 1.6 Tb/s figure is a target within the connectivity collaboration, not a measurement from an operating network. Direct sources: Qualcomm announcement, 8 September 2026, https://www.qualcomm.com/news/releases/2026/09/qualcomm-announces-multi-generational-product-collaboration-with; Reuters, 8 September 2026, https://www.reuters.com/technology/qualcomm-amazon-develop-custom-chips-ai-data-centers-2026-09-08/
Sources
OpenAI, “On the Navier–Stokes Millennium Prize Problem”, 8 September 2026, https://openai.com/index/navier-stokes-solution/
OpenAI, NavierStokesAndEuler Lean repository, 8 September 2026, https://github.com/openai/NavierStokesAndEuler
The Guardian, “OpenAI claims to have solved maths problem that stumped humans for decades”, 8 September 2026, https://www.theguardian.com/science/2026/sep/08/openai-claims-to-have-solved-maths-problem-that-stumped-humans-for-decades
Meta, “Introducing Muse: The World’s First Personal AI Agent Built for Everyone”, 8 September 2026, https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/
Meta, “How We Built Safety Into Muse”, 8 September 2026, https://research.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-muse
Reuters, “Meta launches AI agent that can access other apps to send emails, make payments”, 8 September 2026, https://www.reuters.com/business/meta-launches-ai-agent-that-can-access-other-apps-send-emails-make-payments-2026-09-08/
Firmus, “Firmus surpasses 900 MW contracted capacity, adds OpenAI as anchor customer and expands into Malaysia”, 8 September 2026, https://firmus.co/newsroom/firmus-surpasses-900-mw-contracted-capacity-adds-openai-as-anchor-customer-and-expands-into-malaysia
Reuters, “Nvidia-backed Firmus signs deal with OpenAI for Malaysia data centre capacity”, 8 September 2026, https://www.reuters.com/world/asia-pacific/nvidia-backed-firmus-signs-deal-with-openai-malaysia-data-centre-capacity-2026-09-08/
Qualcomm, “Qualcomm Announces Multi-Generational Product Collaboration with Amazon to Build Next-Generation AI Data Center Infrastructure”, 8 September 2026, https://www.qualcomm.com/news/releases/2026/09/qualcomm-announces-multi-generational-product-collaboration-with
Reuters, “Qualcomm strikes AI chip deal with Amazon, offers right to buy about $4 billion in stock”, 8 September 2026, https://www.reuters.com/technology/qualcomm-amazon-develop-custom-chips-ai-data-centers-2026-09-08/
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.