Imboni yezobuhlakani bokwenziwa isingene esigabeni esisha sentuthuko. Ukuze zibhekane nokuncintisana kwamamodeli esizukulwane esisha ase-China, afana ne-Kimi K3 ye-Moonshot AI ne-V4 Flash ye-DeepSeek, izinkampani ezinkulu ze-AI emhlabeni wonke zehlisa kakhulu amanani emikhiqizo yazo, ngesikhathi zithuthukisa namandla amamodeli ezingeni lokuqala.
I-OpenAI ibe ngeyokugcina ukusebenzisa leli qhinga. Le nkampani yehlise intengo yemodeli yayo eyisisekelo esezingeni eliphambili, i-ChatGPT 5.6 Luna, ngo-80% (kubalwa nge-Token eyisigidi), ngesikhathi yehlisa intengo yemodeli esezingeni eliphakathi, i-5.6 Terra, ngo-20%. Esikhathini esingaphezudlwana kwesonto eledlule, i-Google nayo yethule amamodeli e-Gemini 3.6 Flash ne-3.5 Flash-Lite anamanani aphansi. I-Anthropic yona ayizange yehlise intengo ngokuqondile, kodwa yashintsha imodeli yayo eshibhe kakhulu, i-Opus 4.8, yafaka i-Claude 5.0 enamandla kakhulu, yagcina intengo efanayo.
Ukwanda kokuncintisana emhlabeni wonke kwenza amandla e-AI ashibhe, kodwa lokhu kungase kwehlise amamajini enzuzo alezi zinkampani ezinkulu. Ezinyangeni ezedlule, izinkampani eziningi ezinkulu ze-AI sezimemezele ukunciphisa izindleko nokukhawulela ububanzi bokusetshenziswa kobuchwepheshe be-AI, kuhlanganise ne-xAI ka-Elon Musk, eyake yatshala kakhulu ekuthuthukiseni i-AI. Nakuba abasebenzi abaningi beqala ukusebenzisa i-AI, imibiko yamanje ikhombisa ukuthi ukwanda kwangempela kokukhiqiza akukafiki ezingeni ebelilindelwe.
Umehluko phakathi kwezindlela zokuthuthukisa i-AI e-China nase-US, ngandlela thile, ubonisa amandla ezimboni la mazwe abe nawo isikhathi eside.
Izinkampani zase-US zivame ukusebenzisa izimali ezinkulu kakhulu ekuthuthukiseni ubuchwepheshe obusezingeni eliphambili, kanti onjiniyela base-China basebenzisa amandla okukhiqiza nawobunjiniyela ukuze bakhe amamodeli e-AI ashibhe, amancane ngosayizi, kodwa acishe angasali emuva emandleni asezingeni eliphezulu.
I-DeepSeek yadala ukushaqeka okukhulu kubathuthukisi be-AI baseNtshonalanga ngo-2025, kanti i-Kimi K3 yaphinda yadala umthelela ofanayo ngo-2026. Uma kubhekwa ngayinye, lezi zigameko zodwa sezanele ukwenza izinkampani ezifana ne-OpenAI, Google ne-Anthropic zizizwe zingaphansi kwengcindezi.
Kodwa ezinyangeni ezedlule, izinkampani eziningi ezisebenzisa i-AI kakhulu bezikhala ngokukhuphuka ngokushesha kwezindleko zama-Token. Ngakho lapho kuvela amamodeli amasha asebenza cishe ngendlela efanayo kodwa eshibhe kakhulu, imboni yonke yanyakaziswa kakhulu.
Manje izinkampani ezinkulu ze-AI ngeke zisakwazi ukuncintisana ngokumane zandise inani lamapharamitha amamodeli noma ubukhulu bedatha yokuqeqesha. Kufanele ziqale ukuncintisana ngamanani ngempela, futhi i-OpenAI isivele yehlise kakhulu amanani amanye amamodeli ayo.
Okuphawulekayo ukuthi i-OpenAI ayizange yehlise intengo yemodeli yayo enamandla kakhulu, eyifulegi. Eqinisweni, intengo yenguqulo yayo esheshayo inyukile. Kodwa imodeli yayo esezingeni eliphambili manje isinentengo ephansi kunazo zonke emlandweni, kanti ijubane lokuthuthuka kwamandla e-AI kanye nokwehla kwamanani liyamangaza.
NgoMashi kulo nyaka, i-OpenAI yakhipha i-ChatGPT 5.4, imodeli enamandla kakhulu kuma-agent. Ngaleso sikhathi, intengo yayingu-2.5 USD nge-Token yokufaka eyisigidi (Qaphela: cishe 16.9 yuan ngesilinganiso samanje sokushintshisana) kanye no-15 USD nge-Token yokukhipha eyisigidi (cishe 101.5 yuan ngesilinganiso samanje sokushintshisana).
Manje, intengo ye-GPT 5.6 Luna yehle yaba ngu-0.20 USD nge-Token yokufaka eyisigidi (cishe 1.4 yuan ngesilinganiso samanje sokushintshisana) kanye no-1.20 USD nge-Token yokukhipha eyisigidi (cishe 8.1 yuan ngesilinganiso samanje sokushintshisana).
Lokhu kusho ukuthi imodeli esezingeni eliphambili idlule isikhathi esingaphansi kwezinyanga ezi-4 kuphela kusukela yethulwa kuya ekwehleni okukhulu kwentengo.
Ngemuva kwalokhu kwehliswa kwentengo, i-Luna manje isingene ebangeni lokuncintisana ne-DeepSeek V4. Intengo yenguqulo yobungcweti ye-DeepSeek V4 ingu-0.435 USD nge-Token yokufaka eyisigidi (cishe 2.9 yuan ngesilinganiso samanje sokushintshisana) kanye no-0.87 USD nge-Token yokukhipha eyisigidi (cishe 5.9 yuan ngesilinganiso samanje sokushintshisana).
I-GPT 5.6 Terra iyimodeli enamandla kakhulu, kodwa ngemuva kokwehliswa ngo-20%, intengo yayo nayo yehle yaba ngu-2 USD nge-Token yokufaka eyisigidi (cishe 13.5 yuan ngesilinganiso samanje sokushintshisana) kanye no-12 USD nge-Token yokukhipha eyisigidi (cishe 81.2 yuan ngesilinganiso samanje sokushintshisana), ngaphansi kwe-Kimi K3 edonsa ukunakwa okukhulu (3 USD nge-Token yokufaka eyisigidi (cishe 20.3 yuan ngesilinganiso samanje sokushintshisana), kanye no-15 USD nge-Token yokukhipha eyisigidi (cishe 101.5 yuan ngesilinganiso samanje sokushintshisana)).
Ngesikhathi esifanayo, i-GPT 5.6 Sol isagcina intengo ka-5 USD nge-Token yokufaka eyisigidi (cishe 33.8 yuan ngesilinganiso samanje sokushintshisana) kanye no-30 USD nge-Token yokukhipha eyisigidi (cishe 203 yuan ngesilinganiso samanje sokushintshisana). I-OpenAI iphinde yanyusa intengo yemodeli yayo esezingeni eliphezulu: imodi ye-Fast ye-5.6 Sol ibiza u-10 USD nge-Token yokufaka eyisigidi (cishe 67.7 yuan ngesilinganiso samanje sokushintshisana) kanye no-60 USD nge-Token yokukhipha eyisigidi (cishe 406 yuan ngesilinganiso samanje sokushintshisana). Inikeza izinga elifanayo lobuhlakani ngokubambezeleka okuphansi, futhi incintisana ngokuqondile namamodeli aphambili asezingeni eliphezulu afana ne-Claude Fable 5 ne-Mythos 5.
Kodwa umbuzo uwukuthi: Ingabe la masu okwehlisa amanani angayilethela ngempela inzuzo i-OpenAI?
Ekuqaleni kwalo nyaka, ngemuva kokuba i-OpenAI imemezele ukuthi empeleni iyalishiya icebo lokuba nezikhungo zedatha zokuqala eziphethwe yiyo, izinkontileka zokuqasha ezisayinwe nabahlinzeki abasha bamafu, ama-“Neoclouds”, zaba yinto ebaluleke kakhulu.
Isibonelo, isivumelwano sokuthenga izinsiza zekhompyutha esingafinyelela ku-300 billion USD (cishe 2.03 trillion yuan ngesilinganiso samanje sokushintshisana) esenziwe ne-Oracle sesibe yisisekelo esibalulekile sokuqhubeka nokusebenza kwezinsizakalo ze-OpenAI.
Nokho, uma ngaphambilini abantu bangaphandle bebebuza ukuthi i-OpenAI izozithwala kanjani izindleko ezinjalo, manje lokho kungabaza sekucace nakakhulu.
I-OpenAI manje iqhubeka nokulahlekelwa ebhizinisini lokubhalisa, futhi ekuqaleni kwalo nyaka yehluleka ukufinyelela emgomweni obalulekile wemali engenayo. Ngo-2025, nakuba imali engenayo yayiqhubeka nokukhula, le nkampani yalahlekelwa amashumi ezigidigidi zama-dollar.
I-OpenAI ithembise ukutshala cishe u-600 billion USD (cishe 4.06 trillion yuan ngesilinganiso samanje sokushintshisana) ezinsizeni zekhompyutha ngaphambi kuka-2030.
Ngisho noma imali engenayo iqhubeka nokukhula, ingase ingakwazi neze ukumboza izindleko ezinkulu kangaka. Manje njengoba imodeli ethandwa kakhulu neshibhe kakhulu iye yehliswa futhi intengo, indawo yenzuzo yenkampani ingancipha kakhulu noma inyamalale ngokuphelele.
Lokhu kungase kube esinye sezizathu ezenza kuxoxwe ngokuthi i-NVIDIA ingase inikeze i-OpenAI utshalomali olungu-250 billion USD (cishe 1.69 trillion yuan ngesilinganiso samanje sokushintshisana).
Yebo, i-OpenAI akuyona yodwa inkampani ebhekene nale nkinga. Ngonyaka odlule, ukutshalwa kwezimali kwe-Google kwingqalasizinda ye-AI kwaba cishe izikhathi ezi-9 zemali engenayo yebhizinisi layo lamafu. I-Anthropic yona isanda kukwazi ukuthola inzuzo emalini engenayo ebalwa ngonyaka, ikakhulukazi ngenxa yokubambisana okulinganiselwe ne-xAI kokuqasha isikhungo sedatha i-Colossus ngentengo ephansi.
Izindleko zokusebenzisa nokwakha i-AI azikaze zehle ngokuzumayo, kodwa izinkampani ziyaqhubeka nokwehlisa amanani, zinikeze amamodeli ashesha futhi anamandla kakhulu ngezindleko eziphansi.
Uma ubheka nje, lezi zinombolo zibonakala zingenangqondo.
Imboni ye-AI ivame ukusebenzisa “i-Jevons Paradox” ukuchaza ukuthi kungani ukwamukelwa kwe-AI kushesha.
Esikhathini uJevons ayephila kuso, injini yamalahle esebenza kahle kakhulu ayizange inciphise ukusetshenziswa kwamalahle. Kunalokho, ngenxa yokwanda kobubanzi bokusetshenziswa, isidingo samalahle sisonke sanda.
Namuhla, izinkampani ze-AI zikholelwa ukuthi njengoba izindleko zokusebenzisa i-AI zehla, abantu bazothola izindawo eziningi zokuyisebenzisa, ngaleyo ndlela kwande kakhulu ukusetshenziswa kwe-AI kukonke.
Lokhu kungase kube yisona sizathu esibangela isu lamanje lokwehlisa amanani ama-Token.
Uma ama-Token eshibhe ngokwanele, abasebenzisi bazokwandisa kakhulu ukusetshenziswa kwabo, bese kuthi ukukhula kwesikali kulethe imali engenayo ephezulu. Ngaphezu kwalokho, ngemuva kokuphela kwalo nyaka, njengoba ama-chip amasha asheshisa i-AI esetshenziswa kakhulu ezikhungweni zedatha, amandla okucubungula ama-Token ngeyunithi ngayinye yamandla angase anyuke izikhathi ezi-10, okwenza amabhizinisi anemajini enzuzo aphansi akwazi ukwenza inzuzo ngesikali.
Ngaphezu kwalokho, inkundla ye-Vera Rubin yakusasa ye-NVIDIA nayo kufanele ibhekwe. I-NVIDIA ithi le nkundla izophinde inyuse ukusebenza kahle kokucubungula ama-Token izikhathi ezi-10.
Ngokombono, izinzuzo zezinkundla ze-Blackwell ne-Vera Rubin zingenza amaseva okucabanga e-AI abe imboni enamandla amakhulu okwenza inzuzo.
Noma kunjalo, kunzima ukucabanga ukuthi izinkampani ezinkulu ze-AI zingakwazi ukumboza izindleko ezinkulu ezitshaliwe manje ngokuthembela kuphela emalini engenayo yebhizinisi le-AI, ikakhulukazi njengoba ukuncintisana kuqhubeka nokushuba futhi amanani ama-Token eqhubeka nokwehla.