AI companies are pressing into older software, scarce infrastructure and customer workflows at the same time. OpenAI reported large gains in legacy programming languages, while Chinese technology groups reportedly placed more than $35 billion in orders for export-compliant Nvidia GPUs. Amazon committed fresh capital to German cloud and AI capacity, Exa upgraded search for agents, and Otter.ai secured new growth financing. Policy pressure also grew around an Anthropic-backed federal AI proposal.
OpenAI targets the software languages modern models often miss
OpenAI released MultiLangBench, an evaluation that asks models to build a SQLite-compatible database engine across 20 programming languages. The company reported that GPT-6 Astra reached an 87.4% average test pass rate, compared with 26.8% for GPT-5.6 Sol. It also said GPT-6.1 Sol reached 97.9% accuracy at high reasoning effort when migrating COBOL programs while preserving their behavior.
The work addresses a practical enterprise gap. COBOL, Fortran and Visual Basic still support banking, weather and industrial systems, but provide less training material than Python or JavaScript. Better results could lower the cost of maintenance and migration, yet benchmark success is not production assurance. Organizations still need domain experts, hidden regression tests and careful checks of business rules before changing software that moves money or controls critical systems.
Chinese technology groups reportedly order $35 billion in Nvidia GPUs
Reuters reported that ByteDance, Alibaba and Tencent ordered more than three million Nvidia RTX PRO 5500 graphics cards with a combined list value above $35 billion. The new cards are designed to comply with US export restrictions and are less capable than Nvidia's most advanced accelerators. The report said deliveries still depend on Washington approving export licenses.
The scale of the reported orders shows how strong Chinese demand remains even for constrained hardware. It also turns licensing decisions into a major variable for Nvidia's revenue and for the pace of domestic AI deployment in China. Orders are not the same as completed sales, and list prices may overstate the final value. Still, the rush suggests companies view access to compliant supply as strategically urgent.
Amazon commits more than 16 billion euros to Germany
Amazon said it plans to invest more than 16 billion euros in Germany during 2027. The program includes 4.8 billion euros for AWS data centers and cloud infrastructure, expansion of the Amazon Leo satellite network, logistics modernization and a new Berlin robotics lab. Amazon said the AWS investment would support demand for cloud and AI services.
The announcement links AI growth to regional power, construction, connectivity and workforce capacity. Germany gains infrastructure and technical jobs, while Amazon deepens its position in Europe's largest economy. The broad headline total includes businesses beyond AI, so the AWS figure is the clearer measure of compute investment. Delivery will also depend on energy availability, permitting and whether customers choose the new capacity.
Anthropic's draft bill faces criticism from safety advocates
Transformer reported that a federal AI discussion draft circulated by Anthropic is drawing criticism for being weaker than the bipartisan FRONTIER Act. The proposal would require third-party catastrophic-risk evaluations but, according to the analysis, would let evaluators self-certify and could limit some state enforcement. Anthropic said the text was a discussion draft and not its final position.
The dispute matters because frontier labs increasingly shape the rules meant to govern them. A national framework could reduce conflicting requirements, but weak evaluator standards or broad preemption could also remove useful safeguards. The draft should be judged against concrete enforcement powers, disclosure duties and independence requirements. Public debate is useful here because informal company proposals can become the starting point for legislation.
Exa upgrades search infrastructure for AI agents
Exa introduced Exa 3.0 and a lower-cost Lite search tier priced at $4 per 1,000 requests. The company says a larger embedding model and indexing improvements raised search quality across public and specialized evaluations. It also redesigned result highlights to return more relevant passages with fewer tokens, a useful change for agents that repeatedly retrieve web evidence.
Search is becoming a core dependency for agents because retrieval quality affects every later step. Lower prices can make frequent browsing practical, while compact evidence can reduce model cost and distraction. Exa's benchmark claims are vendor-reported, however, and should be tested on live, time-sensitive and adversarial queries. Reliability also depends on citation fidelity, freshness and how the system handles inaccessible pages.
Otter.ai secures $320 million without issuing equity
Otter.ai announced $320 million in growth financing from General Catalyst's Customer Value Fund. The structure is non-dilutive capital tied to customer acquisition rather than a conventional equity round. Otter plans to use it to expand sales and marketing for its meeting assistant and conversational knowledge products, while the fund receives returns linked to the resulting customer revenue.
The financing reflects confidence that meeting data can become a broader enterprise knowledge layer. It also places pressure on Otter to convert marketing spend into durable subscriptions. Companies adopting persistent meeting agents will need strong consent, retention and access controls because transcripts can contain sensitive decisions. Growth will depend not only on transcription quality, but on whether summaries and follow-up actions remain accurate enough for routine work.
What matters today
Today's signal is that AI advantage increasingly comes from controlling the systems around models. OpenAI is extending capability into legacy code, Nvidia supply is shaped by export policy, and Amazon is investing in the physical capacity behind cloud demand. Exa is improving how agents find evidence, while Otter is betting on workflow ownership. Anthropic's policy dispute is a reminder that technical influence brings governance responsibility. The winners will need credible evaluation, secure deployment and infrastructure that can survive real operational constraints.