Huawei Projects 100,000-Fold AI Token Growth
Huawei Projects 100,000-Fold AI Token Growth
Anonymous
Key Points
1
Epoch AI data cited by HPE’s Vinod Bijlani indicates that performance per dollar of AI-chip spending has improved by about 49% annually since 2023, while the compute needed to reach a given language-model performance level historically fell by roughly half every eight months in 2024.
2
Huawei’s 2026 report marks a change from its earlier editions: rather than focusing mainly on trend forecasts, it sets out ten propositions intended to define concrete industrial pathways toward an intelligent world.
3
Huawei identifies security and privacy protection as two of the ten technological areas needed to support autonomous agents, reflecting concerns about agents’ ability to control equipment, access resources and act without direct human intervention.
4
Huawei is already applying agent-based technology in cybersecurity: the company launched an autonomous security-operations platform in April that uses monitoring and response agents to detect and neutralize cyberthreats.
5
Cisco’s observability expansion is designed to support sovereign-AI deployments by comparing the costs of running models locally with those of using cloud infrastructure; the integration is also intended to require low-effort connections to third-party systems that consume tokens.
Huawei projects that global AI token consumption could explode 100,000-fold by 2035, driven by autonomous agents handling more than 90% of all traffic The Next Web. The company's new Intelligent World 2035 report warns that roughly 900 billion AI agents operating simultaneously will demand massive computing power, better security systems, and new ways to control how agents access resources and make decisions without direct human oversight The Next Web.
Meanwhile, enterprises are scrambling to measure AI spending more precisely. Cisco has added token tracking to its Splunk observability platform, letting companies monitor costs and productivity across tools like Claude, Codex and Cursor Reuters. Industry experts say token pricing alone isn't enough — they're also measuring "dollars per floating-point operation" to understand how efficiently computing power translates into actual AI work Reuters.
Agents Will Generate Almost All AI Traffic by 2035
Huawei's latest report marks a shift in strategy. Instead of just predicting trends, the company now outlines ten specific technological pathways needed to build an intelligent world The Next Web. Autonomous agents rank among the biggest drivers: they're projected to create over 90% of all token usage, replacing human-written requests entirely The Next Web.
This explosion raises urgent questions about control and safety. Huawei lists security and privacy protection as two of the ten critical technology areas required to support agent-based systems The Next Web. The company is already testing agent technology in its own cybersecurity tools — it launched an autonomous security-operations platform in April that uses monitoring and response agents to detect and stop cyberattacks without waiting for human approval The Next Web.
AI Chips Keep Getting Cheaper and Faster
The economics of AI are shifting rapidly. According to Epoch AI data cited by HPE's Vinod Bijlani, the performance-per-dollar of AI chips has improved about 49% every year since 2023 Reuters. At the same time, the amount of computing power needed to train a language model to a given performance level fell roughly in half every eight months throughout 2024 Reuters.
These efficiency gains are reshaping how companies plan AI budgets. The price per million tokens remains useful for estimating consumption, but experts argue it's incomplete without measuring dollars per floating-point operation — a metric that shows how well companies convert money and energy into actual computation Reuters.
Cisco Helps Companies Track AI Spending Across Platforms
Cisco's expansion of its Splunk observability platform gives enterprises visibility into token consumption for the first time. The platform tracks usage, spending, productivity and adoption rates for multiple AI tools and providers, while forecasting end-of-month and end-of-year bills Reuters.
The tool also compares model quality against cost and ties token spending directly back to physical infrastructure expenses like GPUs, memory and networking Reuters. For companies deploying sovereign AI — running models locally instead of in the cloud — this tracking is critical. The platform can help teams decide whether it's cheaper to process tokens on their own servers or pay cloud providers to do the work Reuters.
Sentiment
Mostly NegativeNEGATIVENEUTRALPOSITIVE
Publishers
15
Articles
13
Reach
28
A
AnonymousAuthor
