Raqi-Ai@raqi_ai_uae
Joined October 2025
173 Photos and videos
Nous Research raised $90M at a $1.5B valuation, with Nvidia participating, to take its open-source Hermes agent to enterprises, WSJ reports. GCC: open agents on your own hardware fit data-residency and control needs. wsj.com/pro/venture-capital/… #OpenSource #GCC
1
24
Microsoft and Nvidia unveiled the Surface Laptop Ultra to run AI locally. Memory costs bite: the 128GB DGX Spark now lists at $6,950, and a 64GB model ships Oct 23 from $4,999. GCC: on-prem AI is going mainstream, but budgets are rising. qz.com/microsoft-nvidia-surf… #LocalAI #GCC
64
Meta and Sierra, with Walmart, Stripe, Shopify and others, are building the Personal Agent Protocol, an open standard for AI agents that act for customers. Spec due this month. GCC banks and portals: agent identity is now a core control. cnbc.com/2026/10/06/meta-joi… #AIAgents #GCC
20
Musk says SpaceX's Grok Bot will route each task to the "best back end model," incl. Claude Opus 5.5, Midjourney and Suno. No word on data handling or opt-out. GCC CISOs: ask which models touch your data and if routing can be switched off. runtimewire.com/article/spac… #AIAgents #GCC
1
28
The real agent risk isn't a rogue superintelligence. It's an unapproved payment, a data leak to a public API, or deleted records. Enterprises need auditable, on-prem control planes before agents touch sensitive work. dhia.ai/insights/agent-contr… #AIGovernance #GCC
Made with AI
1
22
GCC/MENA inference teams: MOIRA adapts KV-cache reads per head and layer in vLLM. Reported on H200 at 128K: dense accuracy reading ~30% of pages, 2.2x to 2.5x lower TPOT vs FlashAttention-3. Fewer reads is not less memory. eucax.ai/insights/moira-and-… #LLMInference #KVCache
1
37
TII launched Falcon-Emirati, a 7B model built for Emirati Arabic (84.83% on the Alyah benchmark), plus Falcon-ASR for speech and Falcon-OCR-Arabic for documents. Dialect is where Arabic AI usually breaks. Specialised models are closing the gap. tii.ae/insights/tii-launches…
29
China Telecom AI's Xing4.0-29B-A4B (29B MoE, 4B active) can start on a single consumer GPU. For operators, the bottleneck moves to serving: VRAM fit after quantization, KV cache growth, p95 latency, tenancy and chargeback. clastiq.ai/insights/xing4-se…
25
China Telecom AI's Xing4.0-29B-A4B (29B MoE, 4B active) can start on a single consumer GPU. For operators, the bottleneck moves to serving: VRAM fit after quantization, KV cache growth, p95 latency, tenancy and chargeback. clastiq.ai/insights/xing4-se…
21
GCC/MENA inference teams: TaSQ compresses the KV cache to 1 bit and reports up to 14x larger batches and 1.87x higher peak throughput vs BF16 (SGLang, one RTX 6000 Ada). Bigger batches are not proportional speed, so benchmark. eucax.ai/insights/tasq-and-t… #LLMInference #KVCache
37
GCC/MENA inference teams: faster tokens are not faster answers. SpecScale prunes weak reasoning paths, dedupes compute and defers verification, reporting throughput and latency gains with answer quality preserved. eucax.ai/insights/specscale-… #LLMInference #TestTimeScaling
18
UAE leads 8 markets on AI agents at scale: 15% of UAE CIOs run 500+ (9% global average). But only 5% could contain a problematic agent within 1 to 2 hours (10% worldwide). Deploying is the easy part. Control is the work. The National: thenationalnews.com/future/t…
1
27
Sharon AI's $356m GPU-backed facility is financing, not utilisation. Illustrative 32-GPU month: 69.4% billable vs 56.4% delivered. Cost per billable GPU-hour is $3.075, not the $2.14 installed-hour figure. cognitiongroup.tech/insights…
25
Nscale's $3.36B convertible raise, incl. $1B committed by Nvidia, tests whether AI infrastructure is an investable utility. Binding contracts, power delays and GPU depreciation decide returns. What GCC sovereign buyers should check: raqi.ai/insight/nscales-336b… #SovereignAI #GCC
25
UAE targets 50% of government services and operations on agentic AI within two years. The hard question: who authorised what? Agents need their own identity, scoped permissions and audit trails. Sovereign AI includes control. Khaleej Times: khaleejtimes.com/uae/abu-dha…
2
1
53
From Oct 9, Gemini app users without a plan get only Flash-Lite, per 9to5Google, and AI Plus loses Pro. Access to top models is increasingly tied to paid tiers. Teams planning agents: budget for model tiers and capacity. 9to5google.com/2026/10/03/ge… #Gemini #GCC
31
Bessent told Axios he plans to propose an AI incident notification process with China and thinks Beijing would agree, adding that people at the labs must accept responsibility. Expect incident norms to reach enterprise AI contracts. axios.com/2026/10/03/china-a… #AIPolicy #GCC
20
Trump named DNI Jay Clayton as AI czar to lead a new "Super Intelligence Force" with 120 days to report on AI's risks and opportunities, CNBC reports. Gulf partners: U.S. policy tone shapes export controls and AI deal timing. cnbc.com/2026/10/03/trump-ja… #AIPolicy #GCC
21
DeepSeek released official macOS and Windows desktop apps for DeepSeek Harness, its MIT-licensed open-source agent harness (v0.2 preview). Open runtimes may help keep data in-region, but Chinese-origin tooling needs a security review. marktechpost.com/2026/10/03/… #AIAgents #GCC
1
1
65
David Robinson, who says he led OpenAI's launch safety reports, resigned, writing in The Atlantic that its "culture is broken." OpenAI says it keeps improving safety. CISOs: put vendor safety culture in AI due diligence. techcrunch.com/2026/10/03/op… #AISafety #GCC
26
Load more